Compare commits

..

27 Commits

Author SHA1 Message Date
Mina Choi 5504f79a9d refactor(report): build_overrides + patch_report 통합 / list wholesale merge
- _build_overrides 가 result 받아 deep_merge 까지 처리, _patch_report 제거
- _deep_merge: list by-index → wholesale 치환 (EN 슬롯 누락/라벨 섞임 차단)
- build_facebook_audit: template-copy 대신 LLM logo/logo_description 만 두 페이지에 공통 적용
- _page_patch: language/label 명시 박음 (KR/EN 교차 오염 방지)
- FacebookPage/InstagramAccount/YouTubeAudit: 불필요한 Optional 제거, has_whatsapp/top_content_type 만 Optional 유지
- build_instagram_audit/build_facebook_audit: dict 반환 (overrides[k] = patch 단순 박기)
2026-06-02 17:04:33 +09:00
Mina Choi 9a9ce1319f fix(branding): logo URL 컬럼 일관성 + 잘못된 로고 묘사 회피
- 채널 collectors (instagram/facebook/youtube/tiktok) 가 profileImage 를 raw_info.logo_url 컬럼에도 저장
- collect_brand_basics 가 공식 로고 URL 을 branding row 가 아니라 mainpage row 의 logo_url 컬럼에 저장
- select_branding_logo_url 가 mainpage row 의 logo_url 조회하도록 SQL 수정
- select_run_raw_data 가 logo_url 컬럼도 반환 (_logo_url 합성키) → branding._describe_channel_logos 가 컬럼에서 통일된 이름으로 읽음
- _describe_logo candidates 에서 firecrawl ogImage 제거 (이벤트 배너 잘못 잡히던 케이스)
- extra_channels (tiktok/kakaotalk/naver_cafe) language='KR' 박음

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-02 13:12:58 +09:00
Mina Choi af61713697 refactor(branding): collect/report 단계 분리 + Vision logo hex 추가
- integrations/color_extractor → integrations/site_fetcher (HTTP) + services/brand_parser (파싱) 분리
- integrations/vision → integrations/llm/gemini_vision 이동
- services/collect_extras → services/collect.collect_brand_basics (collect) + services/branding (report) 분리
- Vision prompt 에 logo_colors_hex 5개 강제 + 길이 fallback (4·6개 들어와도 5개로 정규화)
- branding 단계: HTML parser canonical logo URL 을 Vision 에 1순위 전달
  → firecrawl 가 잘못된 이미지 (마케팅 배너 등) 를 logo 로 잡는 케이스 회피
- select_run 에서 큰 JSON 컬럼 (report_data/plan_data) 빼서 meta only
  → generate_plan 만 select_run_report_data 별도 조회. 4군데 호출자는 가벼워짐

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-02 11:08:38 +09:00
Mina Choi b844951ad8 refactor(branding): logo URL 을 raw_info.logo_url 컬럼으로 분리
- collect_brand_assets: Vision 결과의 logo_images 를 JSON 에서 제거하고
  진짜 로고(logo/og 매칭) 인 경우만 raw_info.logo_url 컬럼에 저장.
  favicon-only 매칭은 컬럼 저장 X (옛 logic 동일).
- analysis._build_overrides: select_branding_logo_url 로 컬럼 읽어
  ClinicSnapshot.logo_images 를 horizontal=logo_url 로 재구성.
- branding raw_data 가 "사실 데이터(URL/hex)" vs "Vision 분석 텍스트(묘사)"
  섞이던 문제 일부 해소 — URL 은 컬럼, 텍스트만 JSON 에 잔존.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-01 17:01:19 +09:00
Mina Choi 009d95377a Merge 'b6a0134 db 스키마 변경' on top of db-migration
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-01 16:57:12 +09:00
Mina Choi c23e620fb4 Merge branch 'db-migration': remote_source + raw_info 통합 스키마
- common/db.py 단일 파일 → common/db/ 패키지로 분리 (hospital/source/run/market/file_data)
- 모든 채널 데이터를 raw_info 단일 테이블로 통일 (hospital_baseinfo.raw_data / 채널별 *_data 테이블 제거)
- 부가 채널(tiktok/instagram_en/facebook_en/kakaotalk/naver_cafe)도 remote_source+raw_info 로 일원화
  - EN 채널은 같은 source_type + language='EN' 으로 구분, select_run_raw_data 가 합성키로 반환
- SourceType.BRANDING 추가 — brand_assets/channel_logos 결과를 하나의 raw_info entry 에 머지
- collect.collect_all: main wave gather → branding 2단계 순차 실행
- mock_urls 매칭 + _with_scheme 보정 유지

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-01 16:57:08 +09:00
jaehwang b6a0134ba7 db 스키마 변경 2026-06-01 16:51:30 +09:00
Mina Choi 86af23b56d feat(kpi): 규모별 성장률 공식으로 KPI dashboard 코드 산출
Perplexity Sonar가 KPI target schema 필드를 구조적으로 못 채우는 한계 검증됨 (프롬프트 강화·sonar-pro·sonar-reasoning-pro·hint 주입 다 실패). mockup 7개(irum/grand/o2o/ts/banobagi/wonjin/viewclinic) 역분석으로 추출한 채널 규모별 성장률 공식을 코드에서 결정적으로 산출 → 100% 재현성 확보.

- kpi_dashboard.py(신규): _target_multiplier 4단계 + _blog_frequency cadence + 강남언니 리뷰 보수적 multiplier
- 8 metric 산출: YouTube 구독자 / Instagram KR·EN 팔로워 / Facebook KR·EN 팔로워 / TikTok 팔로워 / Naver Cafe 회원 / 네이버 블로그 포스팅 빈도 / 강남언니 리뷰
- analysis.py: _build_overrides에서 build_kpi_dashboard 호출, _patch_report에서 LLM 출력 무시하고 코드값 강제
- common/utils.parse_ts: facebook_audit._parse_ts 옮겨 공용화 (FB·블로그 RSS 둘 다 사용). ISO 8601 / epoch / RFC 2822(네이버 RSS) 통합 처리
- report_prompt: kpi_dashboard는 코드 강제 치환 안내 + overall_score는 channel_scores 평균으로 0/null 금지 가드 추가

mockup viewclinic YT 구독자 104K→115K→200K 정확 일치 검증. 라이프사이클 4단계로 같은 raw_data 입력 시 매번 동일 output 보장.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-01 15:45:06 +09:00
jaehwang 3b4c154fb2 db migration done 2026-06-01 15:31:33 +09:00
Mina Choi e5a9036e47 fix(report+analysis): Instagram/Facebook Optional 완화 + viewclinic mock 제거 + brand_assets 강제주입
- schemas/report.py: InstagramAccount/InstagramAudit/FacebookPage/FacebookAudit 필드 Optional 완화
  (LLM이 page 1·2개 모두 language/label/logo/has_whatsapp 등 빼먹는 케이스 차단)
- analysis.py: viewclinic mock 분기(_is_mock, _load_mock_report, _load_mock_plan) 제거 — raw_data 충분
- analysis.py: _build_clinic_snapshot에 brandAssets.logo_images/brand_colors 강제 주입
  (LLM 프롬프트 가드 무시하고 null 두는 케이스 차단)
- analysis.py: facebook_audit.pages 머지 방식 변경 — LLM 첫 페이지 템플릿 복제 후 코드 patch로 인덱스별 덮어쓰기
  (EN(index 1) 드랍 + label/logo 누락 검증 실패 동시 회피)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-01 08:50:35 +09:00
Mina Choi 5dbc7d7ffe fix(report): ClinicSnapshot/YouTubeAudit/Instagram*/Facebook* Optional 완화
required로 두면 LLM 응답이나 수집 데이터 누락 시 pydantic ValidationError로
리포트 endpoint 전체가 500으로 죽음. 실제 테스트(청담오라클)에서 LLM이
weekly_view_growth, established 등 10개 필드를 null 반환하는 케이스 확인.

- ClinicSnapshot/YouTubeAudit: schemas + models 양쪽 모두 Optional (LLM 입력 검증
  + FastAPI 응답 검증 둘 다 통과 필요)
- InstagramAccount/InstagramAudit/FacebookPage/FacebookAudit: models만 (인스타·페북 빈
  계정/페이지 케이스 대응)
- list[T] 필드는 기본값 [] 부여

트레이드오프: 스키마 레벨 데이터 완결성 보장 약화. 운영하며 자주 비는 필드
패턴 보고 collection 단계 보강 필요.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 16:42:04 +09:00
Mina Choi 71b605eaa6 Merge branch 'wip/channel-brand-2026-05-29'
채널 확장 + 브랜드 자산 파이프라인을 main에 통합.

신규/주요 변경:
- 5채널 외 부가 수집 (틱톡/IG·FB 영문/네이버 카페/카카오톡) — collect_extras.py
- 브랜드 자산: 홈페이지 로고 URL + CSS 색상 추출 (color_extractor.py) + Gemini Vision 로고 묘사 (vision.py)
- 채널 로고 비교: 공식 로고와 각 채널 프로필 이미지 일치 여부 평가
- 인스타/페북 audit 빌더 분리 (instagram_audit.py, facebook_audit.py)
- mock_urls.py: 78개 병원 영문 채널 51건 + 필드 캐노니컬 순서 정규화
- ReportInput/PlanInput 신규 채널 필드 추가
- ChannelBrandingRule literal "missing" → "N/A"

teammate eed5772와의 conflict 해결:
- ClinicSnapshot/YouTubeAudit: teammate가 신뢰 못하는 필드 제거 (established/years_in_business/price_range/media_appearances/medical_tourism/nearest_station/subscriber_rank)
- services/analysis.py: teammate의 _build_clinic_snapshot/_build_youtube_audit/duration helpers + 우리의 _naver_blog_summary 둘 다 보존
- imports: youtube_diagnosis_prompt + build_instagram_accounts/build_facebook_pages 모두 채택

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 16:40:21 +09:00
jaehwang c9c5ee9177 Merge branch 'main' into db-migration 2026-05-29 16:31:47 +09:00
Mina Choi aff2b2720d WIP: channel-brand merge + Optional 모델 완화 + collect_extras rename + mock_urls 영문 채널 51건
머지 본체:
- 5채널 외 부가 수집(틱톡/IG·FB EN/네이버 카페/카카오톡)
- 브랜드 자산/채널 로고 Vision 분석
- ReportInput/PlanInput에 신규 채널 필드 추가
- ChannelBrandingRule literal "missing" → "N/A"

후속 로컬 작업 (분리 커밋 예정):
- fix(report): ClinicSnapshot/YouTubeAudit/Instagram*/Facebook* required→Optional (LLM null 응답 대응)
- refactor: enrichment.py → collect_extras.py (네이밍 명확화)
- data(mock_urls): 38개 병원 영문 채널 51건 추가 + 78개 필드 캐노니컬 순서 정규화

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 16:22:17 +09:00
jaehwang eed57729d9 clinic_overview , youtube analysis 정리 2026-05-29 16:19:06 +09:00
Mina Choi 56fa2c6238 chore: schema/model 잔여 sync (이전 커밋에 빠진 스키마 필드)
- ReportInput / Channels: kakao_talk, naver_cafe 필드 (이전 카카오/카페 채널 커밋 092bfe7 에서 누락)
- PlanInput: naver_blog 필드 (이번 네이버 블로그 채널 커밋 9da285e 에서 누락)
- ChannelBrandingRule literal: "missing" → "N/A" 통일 (이전 missing→N/A 커밋 5f1eee8 에서 누락)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:46:15 +09:00
Mina Choi 4bc7c9652c data(mock_urls): 카카오톡·네이버 카페 URL 일괄 추가 + 필드 정렬
78개 클리닉에 kakaoTalk / naverCafe 필드 추가, 검색 agent 가 일괄 조회한 결과
적용:
- kakaoTalk: 68개 (한국 클리닉 87% 가 카카오톡 채널 운영 — pf.kakao.com/_X 형태)
- naverCafe: 3개 (의료 클리닉 공식 카페 운영은 드물어 적음)

URL 형식 정규화: https://, www. 접두사 제거하고 호스트부터 시작.

확실하지 않은 케이스는 agent 가 의도적으로 빈값으로 둠 (개인 카톡 친구 추가
링크나 오픈채팅, 동명 다른 병원 카페 같이 false positive 위험 있는 케이스).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:45:36 +09:00
Mina Choi bed5f0c274 chore: TIKTOK_ACTOR 상수 + 수집기 옵저버빌리티 정리
apify.py: 라이브 actor id 들을 모두 모듈 상단 상수로 통일 (TIKTOK_ACTOR 추가).
fetch_tiktok_profile 이 raw 문자열 'clockworks~tiktok-scraper' 쓰던 것 정리.
이제 IG_PROFILE / IG_HIGHLIGHTS / FB_PAGES / FB_POSTS / TIKTOK 5개 상수.

수집기 옵저버빌리티 정리:
- collect.py: 채널별 done 로그에 붙이던 _summarize (followers/posts 등 데이터
  shape inspection) 제거 — production 로그가 아니라 진단용에 가까워 test_raw.py
  의 summarize() 로 대신 충분.
- enrichment.py / pipeline.py / collect.py: 저레벨 수집기의 timing instrumentation
  은 정리. orchestrator 레벨(pipeline 의 stage_times, analysis/market 의 LLM
  호출 timing)은 유지.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:45:23 +09:00
Mina Choi fa32109658 fix(color_extractor): CSS .logo 패턴 우선순위 + lang/flag noise 필터 강화
문제: JK 성형외과 (jkplastic.com) 처럼 <h1 class="logo"><a>JK PLASTIC</a></h1>
형태로 logo 텍스트만 있고 진짜 이미지는 외부 CSS의 .logo { background-image: url(...) }
로 들어가는 사이트에서, generic <header> 첫 img 패턴이 한국어 깃발(lang-kor.png)을
먼저 잡아 잘못된 로고가 박혔음.

수정:
- find_logo_url_in_html 흐름 재정렬:
  1) class/id/alt/src 명시 + 부모 class="logo" + 중첩 img (specific)
  2) **외부 CSS 의 .logo background-image** ← generic 보다 앞으로 (class-based 라
     더 specific)
  3) <header>/<nav> 첫 img (가장 generic, 잘못 잡힐 위험)
- noise 필터 강화: lang-kor / lang-eng / flag / country / icon- / btn- / arrow /
  prev / next / search 같이 logo 아닌 게 명백한 src 는 모든 단계에서 skip

검증: JK 는 lang-kor.png → logo-color.png 로 정확히 잡힘.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:45:08 +09:00
Mina Choi dca0c78860 fix(url): _with_scheme 강화 — www 자동 보강 + 중첩 https:// 정리 + API 입력 적용
문제 1: gangnamunni.com 의 SSL 인증서가 www.gangnamunni.com 에만 유효 →
  사용자가 'gangnamunni.com/hospitals/189' 같이 줬을 때 클릭 시 브라우저 SSL warning.
문제 2: LLM 출력에 'https://www.facebook.com/https://facebook.com/X' 같이 중첩된
  URL이 가끔 박힘.

수정 (_with_scheme):
- 중첩된 'http(s)://' 발견 시 마지막 URL 만 잘라 사용
- _WWW_REQUIRED 도메인 (gangnamunni / facebook / instagram) 은 bare 도메인이면
  www. 자동 보강

api/analysis.py: main 채널(instagram/facebook/naver_blog/youtube/gangnam_unni)
URL 도 _with_scheme 적용해서 DB에 정규화된 형태로 저장. 이전엔 extra channels
(tiktok/EN/카카오톡/카페) 에만 적용돼있어서 강남언니 같은 main 채널이 빠져있었음.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:44:53 +09:00
Mina Choi db42805fdb fix(report): LLM 환각 잠금 — channel mapping 보호 + URL prefix + registry_data
brand_inconsistencies 데이터 보호:
- 채널-묘사 mapping 을 LLM이 swap·재해석해서 Brand Consistency Map 이 어긋났던
  문제 (VIEW 한국페북에 영문 인스타 묘사가 박힌다든가) 해결.
- channel_logos.channel_logos[] 의 channel / logo_description / is_official 을
  **그대로 박을 것** 명시. 절대 swap·변형 금지.

URL 환각 잠금:
- LLM이 'https://www.facebook.com/' 같은 prefix를 raw URL 앞에 붙여서
  'https://www.facebook.com/https://facebook.com/THEPS16445998' 같이 깨지던 문제 차단.
- "URL prefix 절대 직접 만들지 마세요. 받은 URL = 출력 URL" 강제.

registry_data 환각 잠금:
- registry_data.website_en 같은 자유 필드를 LLM이 그럴듯하게 ('thepsclinic.com'
  같이) 지어내던 문제. "데이터에 없으면 반드시 null" 강제.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:44:38 +09:00
Mina Choi 9da285e905 feat(plan): 네이버 블로그 채널 + brand_guide profile_photo 시스템 박기
네이버 블로그 채널 추가:
- naver.fetch_blog_total_count: RSS에 totalCount 없으면 blog.naver.com 의 PostList
  페이지 HTML에서 '(\d+)개의 글' 패턴으로 진짜 전체 글 수 추출
  (RSS는 최근 50개만 줘서 그동안 totalResults=50 으로 잘못 박혔음 — 뷰성형외과 실제 554개)
- analysis._naver_blog_summary 다이어트: totalPosts + latestPostDate 만 LLM에 보냄
  (posts 본문/링크/제목 빼서 토큰 절약 + LLM의 무관 정보 hallucinate 방지)
- plan_prompt: channelStrategies 리스트에 네이버 블로그 명시 포함

brand_guide.channel_branding.profile_photo 코드 박기:
- 기존: LLM이 "공식 로고로 통일 (가이드 미보유)" 같은 fallback 문구 hallucinate
- 수정: analysis._patch_plan 이 모든 채널의 profile_photo 를 brand_assets.logo_description
  으로 일괄 박음 (채널 통일 전략이라 모두 동일 값)
- plan_prompt: "profilePhoto 는 빈 문자열로 두세요 — 시스템이 채웁니다" 명시

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:44:18 +09:00
Mina Choi 8c1e513dc0 fix(vision): channel logo describe — 3채널씩 청크 호출로 매칭 정확도 향상
기존: 공식 로고 + 모든 채널 프로필 이미지를 한 번에 묶어 Gemini에 보냄 →
LLM이 채널-이미지 매칭을 헷갈려 같은 묘사를 여러 채널에 복사하는 문제.
VIEW 케이스에서 한국 페북·영문 인스타가 둘 다 "보라/노란 V자형 공식 로고" 묘사로
잘못 박혔음 (실제로는 흰배경 V자 심볼 vs 금색 VIEW로 완전히 다름).

수정: describe_channel_logos를 3채널씩 청크로 분리 + 명시적 이미지 번호 매핑:
- "이미지 1 = 공식 로고, 이미지 2 = Instagram 채널, 이미지 3 = Facebook..." 식
- "공식 로고 묘사를 절대 복사하지 마세요" 강한 지시
- 청크별 병렬 호출 (asyncio.gather)
- inconsistency_summary / recommendation 은 LLM 한 번 더 안 부르고 결정적 산출

비용: 호출 1회 → 청크 수 만큼 (보통 2회), 페니 수준 증가
시간: 병렬이라 거의 동일
정확도: 사용자가 본 실제 묘사와 일치하게 됨 (개별 호출 테스트로 검증)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-29 10:44:00 +09:00
Mina Choi 652265cd19 페북 수집·지표·저장 파이프라인 정리
수집:
- pages + posts 두 actor 병렬 호출 (facebook-pages-scraper, facebook-posts-scraper)
- 저장 필드 슬림화: 페이지 메타에서 likes/rating/email/phone/address 제거
  (followers/reviews와 중복이거나 클리닉 raw_data에 이미 있음)
- 게시물 저장은 캡션 160자 + likes/reactions/shares/views/isVideo/timestamp만

지표 계산 위치 이동: 리포트 시점 → 수집 시점:
- recent_post_age / post_frequency / engagement 를 transform_for_storage에서
  결정적으로 산출해 DB에 박음 (재계산 불필요)
- 저장된 게시물은 LLM용 캡션·타입 2필드만 — 추가 슬림 단계 제거

리팩토링:
- services/facebook_audit.py 신설 (instagram_audit 패턴) — _build_overrides의
  인라인 클로저(_fb_page_patch)와 analysis.py의 _fb_post_metrics 분리
- collect.py / enrichment.py 가 transform_for_storage를 호출하도록

엔게이지먼트 표기:
- 범위(min~max)로 표시, 전부 0인 지표는 제외
- 댓글은 actor 미제공이라 "댓글 거의 없음" 고정 부가

콘텐츠 유형:
- top_content_type 은 캡션 본문 주제 추론이 필요해 LLM에 위임
- report_prompt.txt 에 facebook_audit.pages[].top_content_type 작성 지침 추가

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-28 13:49:22 +09:00
jaehwang ab215395c6 db ready 2026-05-28 13:13:30 +09:00
jaehwang d1293f9188 뷰성형외과 전용 mock데이터 외삽 알고리즘 추가 2026-05-28 09:31:41 +09:00
jaehwang 0e68cbe71b 뷰성형외과 전용 mock데이터 외삽 알고리즘 추가 2026-05-21 15:41:43 +09:00
47 changed files with 5664 additions and 1495 deletions

View File

@ -1,158 +1,88 @@
-- 테이블 순서는 관계를 고려하여 한 번에 실행해도 에러가 발생하지 않게 정렬되었습니다.
-- instagram_data Table Create SQL
-- 테이블 생성 SQL - instagram_data
CREATE TABLE instagram_data
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`url` VARCHAR(500) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
);
-- Index 설정 SQL - instagram_data(hospital_id)
CREATE INDEX IX_instagram_data_1
ON instagram_data(hospital_id);
-- facebook_data Table Create SQL
-- 테이블 생성 SQL - facebook_data
CREATE TABLE facebook_data
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`url` VARCHAR(500) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
);
-- Index 설정 SQL - facebook_data(hospital_id)
CREATE INDEX IX_facebook_data_1
ON facebook_data(hospital_id);
-- naver_blog_data Table Create SQL
-- 테이블 생성 SQL - naver_blog_data
CREATE TABLE naver_blog_data
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`url` VARCHAR(500) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
);
-- Index 설정 SQL - naver_blog_data(hospital_id)
CREATE INDEX IX_naver_blog_data_1
ON naver_blog_data(hospital_id);
-- hospital_baseinfo Table Create SQL
-- 테이블 생성 SQL - hospital_baseinfo
CREATE TABLE hospital_baseinfo
(
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL,
`hospital_name` VARCHAR(50) NOT NULL,
`hospital_name_en` VARCHAR(50) NULL,
`brn` VARCHAR(50) NOT NULL,
`road_address` VARCHAR(100) NULL,
`site_address` VARCHAR(100) NULL,
`url` VARCHAR(500) NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (hospital_id)
);
-- Index 설정 SQL - hospital_baseinfo(owner_user_id)
CREATE INDEX IX_hospital_baseinfo_1
ON hospital_baseinfo(owner_user_id);
-- user_info Table Create SQL
-- 테이블 생성 SQL - user_info
-- user_info
CREATE TABLE user_info
(
`user_id` INT NOT NULL AUTO_INCREMENT,
`username` VARCHAR(50) NOT NULL,
`password` VARCHAR(50) NOT NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`user_id` INT NOT NULL AUTO_INCREMENT,
`username` VARCHAR(50) NOT NULL,
`password` VARCHAR(50) NOT NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (user_id)
);
-- youtube_data Table Create SQL
CREATE TABLE youtube_data
-- hospital_baseinfo
CREATE TABLE hospital_baseinfo
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`url` VARCHAR(500) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL,
`hospital_name` VARCHAR(50) NOT NULL,
`hospital_name_en` VARCHAR(50) NULL,
`brn` VARCHAR(50) NOT NULL,
`road_address` VARCHAR(100) NULL,
`site_address` VARCHAR(100) NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (hospital_id)
);
-- Index 설정 SQL - youtube_data(hospital_id)
CREATE INDEX IX_youtube_data_1
ON youtube_data(hospital_id);
CREATE INDEX IX_hospital_baseinfo_1 ON hospital_baseinfo (owner_user_id);
-- gangnam_unni_data Table Create SQL
CREATE TABLE gangnam_unni_data
-- remote_source: 병원별 채널 소스 정보 (instagram/facebook/naver_blog/youtube/gangnam_unni 등)
CREATE TABLE remote_source
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`url` VARCHAR(500) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
`source_id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`source_type` VARCHAR(50) NOT NULL,
`language` CHAR(2) NULL,
`url` VARCHAR(500) NOT NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (source_id)
);
-- Index 설정 SQL - gangnam_unni_data(hospital_id)
CREATE INDEX IX_gangnam_unni_data_1
ON gangnam_unni_data(hospital_id);
CREATE INDEX IX_remote_source_1 ON remote_source (hospital_id);
CREATE INDEX IX_remote_source_2 ON remote_source (hospital_id, source_type);
-- analysis_runs Table Create SQL
-- analysis_runs
CREATE TABLE analysis_runs
(
`analysis_run_id` CHAR(36) NOT NULL,
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL DEFAULT 0,
`status` VARCHAR(50) NOT NULL DEFAULT 'discovering',
`instagram_data_id` INT NULL,
`facebook_data_id` INT NULL,
`naver_blog_data_id` INT NULL,
`youtube_data_id` INT NULL,
`gangnam_unni_data_id` INT NULL,
`report_data` JSON NULL,
`plan_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`analysis_run_id` CHAR(36) NOT NULL,
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL DEFAULT 0,
`status` VARCHAR(50) NOT NULL DEFAULT 'discovering',
`report_data` JSON NULL,
`plan_data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (analysis_run_id)
);
-- Index 설정 SQL - analysis_runs(hospital_id)
CREATE INDEX IX_analysis_runs_1
ON analysis_runs(hospital_id);
-- Index 설정 SQL - analysis_runs(owner_user_id)
CREATE INDEX IX_analysis_runs_2
ON analysis_runs(owner_user_id);
CREATE INDEX IX_analysis_runs_1 ON analysis_runs (hospital_id);
CREATE INDEX IX_analysis_runs_2 ON analysis_runs (owner_user_id);
-- file_data Table Create SQL
-- raw_info: 분석 실행별 수집 원시 데이터
CREATE TABLE raw_info
(
`info_id` INT NOT NULL AUTO_INCREMENT,
`source_id` INT NOT NULL,
`analysis_run_id` CHAR(36) NOT NULL,
`data_tag` VARCHAR(50) NOT NULL DEFAULT 'default',
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`raw_data` JSON NULL,
`logo_url` VARCHAR(500) NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`updated_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (info_id)
);
CREATE INDEX IX_raw_info_1 ON raw_info (analysis_run_id);
CREATE INDEX IX_raw_info_2 ON raw_info (source_id);
-- file_data
CREATE TABLE file_data
(
`id` INT NOT NULL AUTO_INCREMENT,
@ -169,48 +99,38 @@ CREATE TABLE file_data
);
-- hospital_history Table Create SQL
-- hospital_history
CREATE TABLE hospital_history
(
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL,
`hospital_name` VARCHAR(50) NOT NULL,
`hospital_name_en` VARCHAR(50) NULL,
`brn` VARCHAR(50) NOT NULL,
`road_address` VARCHAR(100) NULL,
`site_address` VARCHAR(100) NULL,
`url` VARCHAR(500) NULL,
`status` VARCHAR(20) NOT NULL,
`raw_data` JSON NULL,
`analysis_run_id` CHAR(36) NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`id` INT NOT NULL AUTO_INCREMENT,
`hospital_id` CHAR(36) NOT NULL,
`owner_user_id` INT NOT NULL,
`hospital_name` VARCHAR(50) NOT NULL,
`hospital_name_en` VARCHAR(50) NULL,
`brn` VARCHAR(50) NOT NULL,
`road_address` VARCHAR(100) NULL,
`site_address` VARCHAR(100) NULL,
`status` VARCHAR(20) NOT NULL,
`analysis_run_id` CHAR(36) NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id)
);
-- Index 설정 SQL - hospital_history(hospital_id)
CREATE INDEX IX_hospital_history_1
ON hospital_history(hospital_id);
-- Index 설정 SQL - hospital_history(analysis_run_id)
CREATE INDEX IX_hospital_history_2
ON hospital_history(analysis_run_id);
CREATE INDEX IX_hospital_history_1 ON hospital_history (hospital_id);
CREATE INDEX IX_hospital_history_2 ON hospital_history (analysis_run_id);
-- market_analysis Table Create SQL
-- market_analysis
CREATE TABLE market_analysis
(
`id` INT NOT NULL AUTO_INCREMENT,
`analysis_run_id` CHAR(36) NOT NULL,
`analysis_type` VARCHAR(50) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
`id` INT NOT NULL AUTO_INCREMENT,
`analysis_run_id` CHAR(36) NOT NULL,
`analysis_type` VARCHAR(50) NOT NULL,
`status` VARCHAR(20) NOT NULL DEFAULT 'start',
`data` JSON NULL,
`created_at` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (id),
UNIQUE KEY UQ_market_analysis (analysis_run_id, analysis_type)
);
-- Index 설정 SQL - market_analysis(analysis_run_id)
CREATE INDEX IX_market_analysis_1
ON market_analysis(analysis_run_id);
CREATE INDEX IX_market_analysis_1 ON market_analysis (analysis_run_id);

View File

@ -2,21 +2,23 @@ import logging
import uuid6
from fastapi import APIRouter, BackgroundTasks, Depends, File, Form, HTTPException, UploadFile, status
from common.deps import verify_api_key
from common.db import fetchone, insert_instagram_row, insert_facebook_row, insert_naver_blog_row, insert_youtube_row, insert_gangnam_unni_row, insert_analysis_run
from common.db.hospital import select_hospital
from common.db.source import select_source_mainpage, insert_source, insert_raw_info
from common.db.run import insert_run, select_run_status
from common.utils import _normalize_homepage, _with_scheme
from models.analysis import AnalysisCreate, AnalysisStartResponse, AnalysisStatusResponse
from models.file import FileListItem, FileType, FileUploadResponse
from models.status import AnalysisStatus
from models.status import AnalysisStatus, SourceType
from services.pipeline import run_pipeline
from services.file import get_analysis_files_response, handle_analysis_file_upload, soft_delete_analysis_file
from services.file_data import get_analysis_files_response, handle_analysis_file_upload, soft_delete_analysis_file
from mock_urls import MOCK_CLINICS
from common.utils import _normalize_homepage, _with_scheme
router = APIRouter(prefix="/api/analysis", tags=["analysis"], dependencies=[Depends(verify_api_key)])
logger = logging.getLogger(__name__)
# 추후 DB에 클리닉별로 매핑할 채널(틱톡/영문 인스타·페북). 지금은 mock_urls에서 homepage 매칭으로 보충.
def _extra_channels_from_mockurls(homepage_url: str) -> dict:
"""homepage로 mock_urls에서 클리닉을 찾아 틱톡/영문 인스타·페북 URL 반환 (없으면 {})."""
# 클라가 일부만 보내거나 빈 값이면 mock_urls 의 동일 homepage 매칭으로 채워줌 (메인 + 부가 채널 동일 규칙).
def _channels_from_mockurls(homepage_url: str) -> dict:
target = _normalize_homepage(homepage_url)
if not target:
return {}
@ -24,9 +26,18 @@ def _extra_channels_from_mockurls(homepage_url: str) -> dict:
urls = c["urls"]
if _normalize_homepage(urls.get("homepage", "")) == target:
return {
"tiktok": _with_scheme(urls.get("tiktok")),
# main
"instagram": _with_scheme(urls.get("instagram")),
"facebook": _with_scheme(urls.get("facebook")),
"naver_blog": _with_scheme(urls.get("naverBlog")),
"youtube": _with_scheme(urls.get("youtube")),
"gangnam_unni": _with_scheme(urls.get("gangnamUnni")),
# extra
"tiktok": _with_scheme(urls.get("tiktok")),
"instagram_en": _with_scheme(urls.get("instagramEn")),
"facebook_en": _with_scheme(urls.get("facebookEn")),
"facebook_en": _with_scheme(urls.get("facebookEn")),
"kakao_talk": _with_scheme(urls.get("kakaoTalk")),
"naver_cafe": _with_scheme(urls.get("naverCafe")),
}
return {}
@ -37,34 +48,51 @@ async def start_analysis(body: AnalysisCreate, background_tasks: BackgroundTasks
analysis_run_id = str(uuid6.uuid7())
hospital_id = body.clinic_id
# 사실 hospital과 owner_user_id 비교 후 검증이 필요한 거지만 일단 PoC 니까. 나중에 바꿉니다.
hospital = await fetchone(
"SELECT owner_user_id, url FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
# 사실 hospital 과 owner_user_id 비교 후 검증이 필요한 거지만 일단 PoC 니까. 나중에 바꿉니다.
hospital = await select_hospital(hospital_id)
if not hospital:
raise HTTPException(status_code=409, detail="Clinic not found")
ig_id = await insert_instagram_row(hospital_id, body.channels.instagram) if body.channels.instagram else None
fb_id = await insert_facebook_row(hospital_id, body.channels.facebook) if body.channels.facebook else None
nb_id = await insert_naver_blog_row(hospital_id, body.channels.naver_blog) if body.channels.naver_blog else None
yt_id = await insert_youtube_row(hospital_id, body.channels.youtube) if body.channels.youtube else None
gu_id = await insert_gangnam_unni_row(hospital_id, body.channels.gangnam_unni) if body.channels.gangnam_unni else None
analysis_run_id = await insert_run(analysis_run_id, hospital_id, hospital["owner_user_id"])
analysis_run_id = await insert_analysis_run(
analysis_run_id, hospital_id, hospital["owner_user_id"],
ig_id, fb_id, nb_id, yt_id, gu_id,
)
mainpage = await select_source_mainpage(hospital_id)
if mainpage:
await insert_raw_info(mainpage["source_id"], analysis_run_id, data_tag=SourceType.MAINPAGE)
# branding (HTML/CSS + Vision 로고 매칭) — mainpage 와 같은 homepage URL 을 source 로 사용.
branding_id = await insert_source(hospital_id, SourceType.BRANDING, mainpage["url"], language="KR")
await insert_raw_info(branding_id, analysis_run_id, data_tag=SourceType.BRANDING)
# 클라 값 우선, 없으면 보충 (추후 DB에서 클리닉별로 가져올 값)
mock_extra = _extra_channels_from_mockurls(hospital["url"])
extra_channels = {
"tiktok": body.channels.tiktok or mock_extra.get("tiktok"),
"instagram_en": body.channels.instagram_en or mock_extra.get("instagram_en"),
"facebook_en": body.channels.facebook_en or mock_extra.get("facebook_en"),
}
logger.info("[analysis] extra_channels=%s (mock_matched=%s)", extra_channels, bool(mock_extra))
background_tasks.add_task(run_pipeline, analysis_run_id, extra_channels)
# 클라가 안 보낸 채널은 mock_urls 에서 homepage 매칭으로 보충 (main + extra 동일 규칙).
mock = _channels_from_mockurls((mainpage or {}).get("url") or "")
# 메인 5채널 (KR). _with_scheme 으로 'gangnamunni.com/...' 같이 scheme/www 없이 와도 보강.
main_channels = [
(SourceType.INSTAGRAM, _with_scheme(body.channels.instagram) or mock.get("instagram")),
(SourceType.FACEBOOK, _with_scheme(body.channels.facebook) or mock.get("facebook")),
(SourceType.NAVER_BLOG, _with_scheme(body.channels.naver_blog) or mock.get("naver_blog")),
(SourceType.YOUTUBE, _with_scheme(body.channels.youtube) or mock.get("youtube")),
(SourceType.GANGNAM_UNNI, _with_scheme(body.channels.gangnam_unni) or mock.get("gangnam_unni")),
]
for source_type, url in main_channels:
if url:
source_id = await insert_source(hospital_id, source_type, url, language="KR")
await insert_raw_info(source_id, analysis_run_id, data_tag=source_type)
# 부가 채널 — instagram_en/facebook_en 은 동일 source_type 에 language='EN' 으로 구분, 나머지는 자체 source_type.
extra_channels = [
(SourceType.INSTAGRAM, "EN", _with_scheme(body.channels.instagram_en) or mock.get("instagram_en")),
(SourceType.FACEBOOK, "EN", _with_scheme(body.channels.facebook_en) or mock.get("facebook_en")),
(SourceType.TIKTOK, "KR", _with_scheme(body.channels.tiktok) or mock.get("tiktok")),
(SourceType.KAKAOTALK, "KR", _with_scheme(body.channels.kakao_talk) or mock.get("kakao_talk")),
(SourceType.NAVER_CAFE, "KR", _with_scheme(body.channels.naver_cafe) or mock.get("naver_cafe")),
]
for source_type, language, url in extra_channels:
if url:
source_id = await insert_source(hospital_id, source_type, url, language=language)
await insert_raw_info(source_id, analysis_run_id, data_tag=source_type)
logger.info("[analysis] main+extra channels resolved (mock_matched=%s)", bool(mock))
background_tasks.add_task(run_pipeline, analysis_run_id)
return AnalysisStartResponse(
analysis_run_id=analysis_run_id,
@ -101,12 +129,12 @@ async def delete_analysis_run_file(run_id: str, file_id: int) -> None:
@router.get("/{run_id}/status", response_model=AnalysisStatusResponse)
async def get_analysis_status(run_id: str):
logger.info("GET /api/analysis/%s/status", run_id)
row = await fetchone("SELECT status FROM analysis_runs WHERE analysis_run_id = %s", (run_id,))
if not row:
run_status = await select_run_status(run_id)
if run_status is None:
raise HTTPException(status_code=404, detail="Run not found")
return AnalysisStatusResponse(
analysis_run_id=run_id,
status=AnalysisStatus(row["status"]),
status=AnalysisStatus(run_status),
progress=50.0,
current_step="",
channel_errors={},

View File

@ -2,7 +2,8 @@ import logging
import uuid6
from fastapi import APIRouter, Depends, HTTPException, status
from common.deps import verify_api_key
from common.db import insert_hospital, fetchone
from common.db.hospital import select_hospital, insert_hospital
from common.db.source import insert_source
from common.utils import get_env
from integrations.firecrawl import FirecrawlClient
from models.clinic import ClinicCreate, ClinicCreateResponse, ClinicResponse, ClinicHistoryResponse, RunSummary
@ -30,9 +31,8 @@ async def create_clinic(body: ClinicCreate):
name=info["clinicName"],
name_en=info.get("clinicNameEn"),
road_address=info.get("address"),
url=body.url,
raw_data=info,
)
await insert_source(hospital_id, "mainpage", body.url)
return ClinicCreateResponse(
id=hospital_id,
url=body.url,
@ -44,11 +44,7 @@ async def create_clinic(body: ClinicCreate):
@router.get("/{hospital_id}", response_model=ClinicResponse)
async def get_clinic(hospital_id: str):
logger.info("GET /api/clinics/%s", hospital_id)
row = await fetchone(
"SELECT hospital_id, hospital_name, hospital_name_en, road_address, url, status, raw_data, created_at, updated_at"
" FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
row = await select_hospital(hospital_id)
if not row:
raise HTTPException(status_code=404, detail="Clinic not found")
return ClinicResponse(**{**row, "created_at": str(row["created_at"]), "updated_at": str(row["updated_at"])})

View File

@ -1,10 +1,13 @@
import json
import logging
from fastapi import APIRouter, Depends, HTTPException, Response
from common.db import fetchone
from common.db.run import select_run_with_clinic
from common.db.source import select_run_source_raw
from common.deps import verify_api_key
from common.utils import _with_scheme
from integrations.llm.schemas.plan import PlanOutput
from models.plan import PlanApiResponse
from models.status import SourceType
router = APIRouter(prefix="/api/plan", tags=["plan"], dependencies=[Depends(verify_api_key)])
logger = logging.getLogger(__name__)
@ -13,24 +16,21 @@ logger = logging.getLogger(__name__)
@router.get("/{run_id}", response_model=PlanApiResponse, response_model_by_alias=True)
async def get_plan(run_id: str):
logger.info("GET /api/plan/%s", run_id)
row = await fetchone(
"SELECT ar.plan_data, ar.created_at, h.hospital_name, h.hospital_name_en, h.url"
" FROM analysis_runs ar"
" JOIN hospital_baseinfo h ON ar.hospital_id = h.hospital_id"
" WHERE ar.analysis_run_id = %s",
(run_id,),
)
row = await select_run_with_clinic(run_id)
if row is None:
raise HTTPException(status_code=404, detail="Run not found")
if row["plan_data"] is None:
return Response(status_code=204)
data = json.loads(row["plan_data"]) if isinstance(row["plan_data"], str) else row["plan_data"]
plan = PlanOutput(**data)
# 강남언니에서 긁어온 이름이 있으면 우선 (hospital_baseinfo 의 정식 이름보다 강남언니가 더 광고용 표기).
gu = await select_run_source_raw(run_id, SourceType.GANGNAM_UNNI) or {}
clinic_name = gu.get("name") or row["hospital_name"]
return PlanApiResponse(
id=run_id,
clinic_name=row["hospital_name"],
clinic_name=clinic_name,
clinic_name_en=row["hospital_name_en"],
created_at=str(row["created_at"]),
target_url=row["url"],
target_url=_with_scheme(row["target_url"]),
**plan.model_dump(),
)

View File

@ -1,8 +1,9 @@
import json
import logging
from fastapi import APIRouter, Depends, HTTPException, Response
from common.db import fetchone
from common.db.run import select_run_with_clinic
from common.deps import verify_api_key
from common.utils import _with_scheme
from integrations.llm.schemas.report import ReportOutput
from models.report import MarketingReportResponse
@ -13,13 +14,7 @@ logger = logging.getLogger(__name__)
@router.get("/{run_id}", response_model=MarketingReportResponse, response_model_by_alias=True)
async def get_report(run_id: str):
logger.info("GET /api/report/%s", run_id)
row = await fetchone(
"SELECT ar.report_data, ar.created_at, h.hospital_name, h.hospital_name_en, h.url"
" FROM analysis_runs ar"
" JOIN hospital_baseinfo h ON ar.hospital_id = h.hospital_id"
" WHERE ar.analysis_run_id = %s",
(run_id,),
)
row = await select_run_with_clinic(run_id)
if row is None:
raise HTTPException(status_code=404, detail="Run not found")
if row["report_data"] is None:
@ -31,6 +26,6 @@ async def get_report(run_id: str):
clinic_name=row["hospital_name"],
clinic_name_en=row["hospital_name_en"],
created_at=str(row["created_at"]),
target_url=row["url"],
target_url=_with_scheme(row["target_url"]),
**llm_output.model_dump(exclude={"id", "created_at", "target_url"}),
)

View File

@ -1,287 +0,0 @@
import json
import os
import aiomysql
from common.utils import get_env
_pool: aiomysql.Pool | None = None
async def get_pool() -> aiomysql.Pool:
global _pool
if _pool is None:
_pool = await aiomysql.create_pool(
host=get_env("MYSQL_HOST"),
port=int(os.getenv("MYSQL_PORT", "3306")),
user=get_env("MYSQL_USER"),
password=get_env("MYSQL_PASSWORD"),
db=get_env("MYSQL_DB"),
charset="utf8mb4",
minsize=0,
maxsize=30,
connect_timeout=10,
)
return _pool
# 쓰기 (INSERT/UPDATE/DELETE)
async def execute(sql: str, args: tuple = ()) -> int:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor() as cur:
await cur.execute(sql, args)
await conn.commit()
return cur.lastrowid
finally:
conn.close()
# 읽기 (SELECT)
async def fetchone(sql: str, args: tuple = ()) -> dict | None:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor(aiomysql.DictCursor) as cur:
await cur.execute(sql, args)
return await cur.fetchone()
finally:
conn.close()
async def fetchall(sql: str, args: tuple = ()) -> list[dict]:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor(aiomysql.DictCursor) as cur:
await cur.execute(sql, args)
return await cur.fetchall()
finally:
conn.close()
async def insert_instagram_row(hospital_id: str, url: str) -> int:
return await execute("INSERT INTO instagram_data (hospital_id, url) VALUES (%s, %s)", (hospital_id, url))
async def insert_facebook_row(hospital_id: str, url: str) -> int:
return await execute("INSERT INTO facebook_data (hospital_id, url) VALUES (%s, %s)", (hospital_id, url))
async def insert_naver_blog_row(hospital_id: str, url: str) -> int:
return await execute("INSERT INTO naver_blog_data (hospital_id, url) VALUES (%s, %s)", (hospital_id, url))
async def insert_youtube_row(hospital_id: str, url: str) -> int:
return await execute("INSERT INTO youtube_data (hospital_id, url) VALUES (%s, %s)", (hospital_id, url))
async def insert_gangnam_unni_row(hospital_id: str, url: str) -> int:
return await execute("INSERT INTO gangnam_unni_data (hospital_id, url) VALUES (%s, %s)", (hospital_id, url))
async def insert_file_row(
analysis_run_id: str,
file_type: str,
file_name: str,
file_url: str,
size_bytes: int | None = None,
hospital_id: str | None = None,
) -> int:
return await execute(
"INSERT INTO file_data (analysis_run_id, hospital_id, file_type, file_name, file_url, size_bytes)"
" VALUES (%s, %s, %s, %s, %s, %s)",
(analysis_run_id, hospital_id, file_type, file_name, file_url, size_bytes),
)
async def insert_analysis_run(
analysis_run_id: str,
hospital_id: str,
owner_user_id: int,
instagram_data_id: int | None,
facebook_data_id: int | None,
naver_blog_data_id: int | None,
youtube_data_id: int | None,
gangnam_unni_data_id: int | None,
) -> str:
await execute(
"INSERT INTO analysis_runs"
" (analysis_run_id, hospital_id, owner_user_id, instagram_data_id, facebook_data_id, naver_blog_data_id, youtube_data_id, gangnam_unni_data_id)"
" VALUES (%s, %s, %s, %s, %s, %s, %s, %s)",
(analysis_run_id, hospital_id, owner_user_id, instagram_data_id, facebook_data_id, naver_blog_data_id, youtube_data_id, gangnam_unni_data_id),
)
return analysis_run_id
async def save_analysis_report(analysis_run_id: str, data: dict) -> None:
await execute(
"UPDATE analysis_runs SET report_data = %s WHERE analysis_run_id = %s",
(json.dumps(data, ensure_ascii=False), analysis_run_id),
)
async def is_done(table: str, row_id: int | None) -> bool:
if row_id is None:
return True
r = await fetchone(f"SELECT status FROM {table} WHERE id = %s", (row_id,))
return r["status"] == "done"
async def fetch_raw(table: str, row_id: int | None) -> dict | None:
if row_id is None:
return None
row = await fetchone(f"SELECT raw_data FROM {table} WHERE id = %s", (row_id,))
if not row or not row["raw_data"]:
return None
return json.loads(row["raw_data"]) if isinstance(row["raw_data"], str) else row["raw_data"]
async def get_analysis_raw_data(analysis_run_id: str) -> dict:
run = await fetchone(
"SELECT instagram_data_id, facebook_data_id, naver_blog_data_id, youtube_data_id, gangnam_unni_data_id"
" FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
return {
"instagram": await fetch_raw("instagram_data", run["instagram_data_id"]),
"facebook": await fetch_raw("facebook_data", run["facebook_data_id"]),
"naver_blog": await fetch_raw("naver_blog_data", run["naver_blog_data_id"]),
"youtube": await fetch_raw("youtube_data", run["youtube_data_id"]),
"gangnam_unni": await fetch_raw("gangnam_unni_data", run["gangnam_unni_data_id"]),
}
async def set_instagram_status(row_id: int, status: str) -> None:
await execute("UPDATE instagram_data SET status = %s WHERE id = %s", (status, row_id))
async def set_facebook_status(row_id: int, status: str) -> None:
await execute("UPDATE facebook_data SET status = %s WHERE id = %s", (status, row_id))
async def set_naver_blog_status(row_id: int, status: str) -> None:
await execute("UPDATE naver_blog_data SET status = %s WHERE id = %s", (status, row_id))
async def set_youtube_status(row_id: int, status: str) -> None:
await execute("UPDATE youtube_data SET status = %s WHERE id = %s", (status, row_id))
async def set_gangnam_unni_status(row_id: int, status: str) -> None:
await execute("UPDATE gangnam_unni_data SET status = %s WHERE id = %s", (status, row_id))
async def save_instagram_raw_data(row_id: int, data: dict) -> None:
await execute("UPDATE instagram_data SET raw_data = %s, status = 'done' WHERE id = %s", (json.dumps(data, ensure_ascii=False), row_id))
async def save_facebook_raw_data(row_id: int, data: dict) -> None:
await execute("UPDATE facebook_data SET raw_data = %s, status = 'done' WHERE id = %s", (json.dumps(data, ensure_ascii=False), row_id))
async def save_naver_blog_raw_data(row_id: int, data: dict) -> None:
await execute("UPDATE naver_blog_data SET raw_data = %s, status = 'done' WHERE id = %s", (json.dumps(data, ensure_ascii=False), row_id))
async def save_youtube_raw_data(row_id: int, data: dict) -> None:
await execute("UPDATE youtube_data SET raw_data = %s, status = 'done' WHERE id = %s", (json.dumps(data, ensure_ascii=False), row_id))
async def save_gangnam_unni_raw_data(row_id: int, data: dict) -> None:
await execute("UPDATE gangnam_unni_data SET raw_data = %s, status = 'done' WHERE id = %s", (json.dumps(data, ensure_ascii=False), row_id))
async def _insert_hospital_history(hospital_id: str, analysis_run_id: str | None) -> None:
row = await fetchone(
"SELECT owner_user_id, hospital_name, hospital_name_en, brn, road_address, site_address, url, status, raw_data"
" FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
if not row:
return
await execute(
"INSERT INTO hospital_history"
" (hospital_id, owner_user_id, hospital_name, hospital_name_en, brn, road_address, site_address, url, status, raw_data, analysis_run_id)"
" VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)",
(
hospital_id,
row["owner_user_id"],
row["hospital_name"],
row["hospital_name_en"],
row["brn"],
row["road_address"],
row["site_address"],
row["url"],
row["status"],
row["raw_data"] if isinstance(row["raw_data"], str) else json.dumps(row["raw_data"], ensure_ascii=False) if row["raw_data"] else None,
analysis_run_id,
),
)
async def insert_hospital(
hospital_id: str,
name: str,
name_en: str | None = None,
road_address: str | None = None,
site_address: str | None = None,
url: str | None = None,
raw_data: dict | None = None,
owner_user_id: int = 0,
brn: str = "",
) -> dict:
await execute(
"INSERT INTO hospital_baseinfo (hospital_id, hospital_name, hospital_name_en, road_address, site_address, url, raw_data, status, owner_user_id, brn)"
" VALUES (%s, %s, %s, %s, %s, %s, %s, 'done', %s, %s)",
(hospital_id, name, name_en, road_address, site_address, url,
json.dumps(raw_data, ensure_ascii=False) if raw_data else None,
owner_user_id, brn),
)
await _insert_hospital_history(hospital_id, analysis_run_id=None)
return await fetchone(
"SELECT created_at FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
async def save_hospital_raw_data(hospital_id: str, data: dict, analysis_run_id: str | None = None) -> None:
await execute(
"UPDATE hospital_baseinfo"
" SET raw_data = %s, status = 'done',"
" hospital_name = COALESCE(%s, hospital_name),"
" hospital_name_en = COALESCE(%s, hospital_name_en),"
" road_address = COALESCE(%s, road_address)"
" WHERE hospital_id = %s",
(
json.dumps(data, ensure_ascii=False),
data.get("clinicName"),
data.get("clinicNameEn"),
data.get("address"),
hospital_id,
),
)
await _insert_hospital_history(hospital_id, analysis_run_id)
async def merge_hospital_raw_data(hospital_id: str, patch: dict) -> None:
"""hospital_baseinfo.raw_data를 읽어 patch를 top-level 병합 후 저장 (read-modify-write).
부가 수집 단계들이 순차로 raw_data에 키를 덧붙일 사용."""
row = await fetchone("SELECT raw_data FROM hospital_baseinfo WHERE hospital_id = %s", (hospital_id,))
raw = row["raw_data"] if row else None
raw_data = json.loads(raw) if isinstance(raw, str) else (raw or {})
raw_data.update(patch)
await execute(
"UPDATE hospital_baseinfo SET raw_data = %s WHERE hospital_id = %s",
(json.dumps(raw_data, ensure_ascii=False), hospital_id),
)
async def get_market_analysis(analysis_run_id: str) -> dict:
rows = await fetchall(
"SELECT analysis_type, data FROM market_analysis WHERE analysis_run_id = %s AND status = 'done'",
(analysis_run_id,),
)
return {
row["analysis_type"]: json.loads(row["data"]) if isinstance(row["data"], str) else row["data"]
for row in rows
}

16
app/common/db/__init__.py Normal file
View File

@ -0,0 +1,16 @@
from common.db.base import execute, fetchone, fetchall
from common.db.hospital import select_hospital, update_hospital_status, insert_hospital, update_hospital
from common.db.source import (
insert_source, select_source_mainpage, select_source_by_type,
insert_raw_info, update_raw_info_status, update_raw_info, update_raw_info_merge,
update_raw_info_logo_url, select_mainpage_logo_url, select_branding_info_id,
select_raw_info_data,
select_run_sources, select_run_raw_data, select_run_source_raw,
select_run_mainpage_url,
)
from common.db.run import (
insert_run, select_run, select_run_status, update_run_status,
update_run_report, update_run_plan, select_run_with_clinic, select_run_report_data,
)
from common.db.market import upsert_market_status, upsert_market_result, select_market
from common.db.file_data import insert_file, select_run_files, select_file, delete_file

56
app/common/db/base.py Normal file
View File

@ -0,0 +1,56 @@
import os
import aiomysql
from common.utils import get_env
_pool: aiomysql.Pool | None = None
async def get_pool() -> aiomysql.Pool:
global _pool
if _pool is None:
_pool = await aiomysql.create_pool(
host=get_env("MYSQL_HOST"),
port=int(os.getenv("MYSQL_PORT", "3306")),
user=get_env("MYSQL_USER"),
password=get_env("MYSQL_PASSWORD"),
db=get_env("MYSQL_DB"),
charset="utf8mb4",
minsize=0,
maxsize=30,
connect_timeout=10,
)
return _pool
async def execute(sql: str, args: tuple = ()) -> int:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor() as cur:
await cur.execute(sql, args)
await conn.commit()
return cur.lastrowid
finally:
conn.close()
async def fetchone(sql: str, args: tuple = ()) -> dict | None:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor(aiomysql.DictCursor) as cur:
await cur.execute(sql, args)
return await cur.fetchone()
finally:
conn.close()
async def fetchall(sql: str, args: tuple = ()) -> list[dict]:
pool = await get_pool()
async with pool.acquire() as conn:
try:
async with conn.cursor(aiomysql.DictCursor) as cur:
await cur.execute(sql, args)
return await cur.fetchall()
finally:
conn.close()

View File

@ -0,0 +1,39 @@
from common.db.base import execute, fetchone, fetchall
async def insert_file(
analysis_run_id: str,
file_type: str,
file_name: str,
file_url: str,
size_bytes: int | None = None,
hospital_id: str | None = None,
) -> int:
return await execute(
"INSERT INTO file_data (analysis_run_id, hospital_id, file_type, file_name, file_url, size_bytes)"
" VALUES (%s, %s, %s, %s, %s, %s)",
(analysis_run_id, hospital_id, file_type, file_name, file_url, size_bytes),
)
async def select_run_files(analysis_run_id: str) -> list[dict]:
return await fetchall(
"SELECT id, file_type, file_name, file_url, size_bytes, created_at"
" FROM file_data WHERE analysis_run_id = %s AND is_deleted = FALSE"
" ORDER BY created_at DESC",
(analysis_run_id,),
)
async def select_file(file_id: int, analysis_run_id: str) -> dict | None:
return await fetchone(
"SELECT id FROM file_data WHERE id = %s AND analysis_run_id = %s",
(file_id, analysis_run_id),
)
async def delete_file(file_id: int) -> None:
await execute(
"UPDATE file_data SET is_deleted = TRUE WHERE id = %s AND is_deleted = FALSE",
(file_id,),
)

78
app/common/db/hospital.py Normal file
View File

@ -0,0 +1,78 @@
from common.db.base import execute, fetchone
async def select_hospital(hospital_id: str) -> dict | None:
return await fetchone(
"SELECT hospital_id, owner_user_id, hospital_name, hospital_name_en,"
" brn, road_address, site_address, status, created_at, updated_at"
" FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
async def update_hospital_status(hospital_id: str, status: str) -> None:
await execute(
"UPDATE hospital_baseinfo SET status = %s WHERE hospital_id = %s",
(status, hospital_id),
)
async def _insert_hospital_history(hospital_id: str, analysis_run_id: str | None) -> None:
row = await fetchone(
"SELECT owner_user_id, hospital_name, hospital_name_en, brn, road_address, site_address, status"
" FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
if not row:
return
await execute(
"INSERT INTO hospital_history"
" (hospital_id, owner_user_id, hospital_name, hospital_name_en, brn, road_address, site_address, status, analysis_run_id)"
" VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s)",
(
hospital_id,
row["owner_user_id"],
row["hospital_name"],
row["hospital_name_en"],
row["brn"],
row["road_address"],
row["site_address"],
row["status"],
analysis_run_id,
),
)
async def insert_hospital(
hospital_id: str,
name: str,
name_en: str | None = None,
road_address: str | None = None,
site_address: str | None = None,
owner_user_id: int = 0,
brn: str = "",
) -> dict:
await execute(
"INSERT INTO hospital_baseinfo"
" (hospital_id, hospital_name, hospital_name_en, road_address, site_address, status, owner_user_id, brn)"
" VALUES (%s, %s, %s, %s, %s, 'done', %s, %s)",
(hospital_id, name, name_en, road_address, site_address, owner_user_id, brn),
)
await _insert_hospital_history(hospital_id, analysis_run_id=None)
return await fetchone(
"SELECT created_at FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
async def update_hospital(hospital_id: str, data: dict, analysis_run_id: str | None = None) -> None:
await execute(
"UPDATE hospital_baseinfo"
" SET status = 'done',"
" hospital_name = COALESCE(%s, hospital_name),"
" hospital_name_en = COALESCE(%s, hospital_name_en),"
" road_address = COALESCE(%s, road_address)"
" WHERE hospital_id = %s",
(data.get("clinicName"), data.get("clinicNameEn"), data.get("address"), hospital_id),
)
await _insert_hospital_history(hospital_id, analysis_run_id)

31
app/common/db/market.py Normal file
View File

@ -0,0 +1,31 @@
import json
from common.db.base import execute, fetchall
async def upsert_market_status(analysis_run_id: str, analysis_type: str, status: str) -> None:
await execute(
"INSERT INTO market_analysis (analysis_run_id, analysis_type, status)"
" VALUES (%s, %s, %s)"
" ON DUPLICATE KEY UPDATE status = VALUES(status)",
(analysis_run_id, analysis_type, status),
)
async def upsert_market_result(analysis_run_id: str, analysis_type: str, data: dict) -> None:
await execute(
"INSERT INTO market_analysis (analysis_run_id, analysis_type, status, data)"
" VALUES (%s, %s, 'done', %s)"
" ON DUPLICATE KEY UPDATE status = 'done', data = VALUES(data)",
(analysis_run_id, analysis_type, json.dumps(data, ensure_ascii=False)),
)
async def select_market(analysis_run_id: str) -> dict:
rows = await fetchall(
"SELECT analysis_type, data FROM market_analysis WHERE analysis_run_id = %s AND status = 'done'",
(analysis_run_id,),
)
return {
row["analysis_type"]: json.loads(row["data"]) if isinstance(row["data"], str) else row["data"]
for row in rows
}

76
app/common/db/run.py Normal file
View File

@ -0,0 +1,76 @@
import json
from common.db.base import execute, fetchone
async def insert_run(
analysis_run_id: str,
hospital_id: str,
owner_user_id: int,
) -> str:
await execute(
"INSERT INTO analysis_runs (analysis_run_id, hospital_id, owner_user_id) VALUES (%s, %s, %s)",
(analysis_run_id, hospital_id, owner_user_id),
)
return analysis_run_id
async def select_run(analysis_run_id: str) -> dict | None:
return await fetchone(
"SELECT analysis_run_id, hospital_id, owner_user_id, status, created_at, updated_at"
" FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
async def select_run_report_data(analysis_run_id: str) -> dict | None:
"""report 결과가 필요할 때만 호출. raw JSON 파싱해서 dict 반환."""
import json
row = await fetchone(
"SELECT report_data FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
if not row or not row["report_data"]:
return None
return json.loads(row["report_data"]) if isinstance(row["report_data"], str) else row["report_data"]
async def select_run_status(analysis_run_id: str) -> str | None:
row = await fetchone(
"SELECT status FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
return row["status"] if row else None
async def update_run_status(analysis_run_id: str, status: str) -> None:
await execute(
"UPDATE analysis_runs SET status = %s WHERE analysis_run_id = %s",
(status, analysis_run_id),
)
async def update_run_report(analysis_run_id: str, data: dict) -> None:
await execute(
"UPDATE analysis_runs SET report_data = %s WHERE analysis_run_id = %s",
(json.dumps(data, ensure_ascii=False), analysis_run_id),
)
async def update_run_plan(analysis_run_id: str, data: dict) -> None:
await execute(
"UPDATE analysis_runs SET plan_data = %s WHERE analysis_run_id = %s",
(json.dumps(data, ensure_ascii=False), analysis_run_id),
)
async def select_run_with_clinic(analysis_run_id: str) -> dict | None:
return await fetchone(
"SELECT ar.report_data, ar.plan_data, ar.created_at,"
" h.hospital_name, h.hospital_name_en,"
" rs.url AS target_url"
" FROM analysis_runs ar"
" JOIN hospital_baseinfo h ON ar.hospital_id = h.hospital_id"
" LEFT JOIN remote_source rs ON rs.hospital_id = h.hospital_id AND rs.source_type = 'mainpage'"
" WHERE ar.analysis_run_id = %s",
(analysis_run_id,),
)

162
app/common/db/source.py Normal file
View File

@ -0,0 +1,162 @@
import json
from common.db.base import execute, fetchone, fetchall
from models.status import SourceType
async def insert_source(
hospital_id: str,
source_type: SourceType,
url: str,
language: str | None = None,
) -> int:
return await execute(
"INSERT INTO remote_source (hospital_id, source_type, language, url) VALUES (%s, %s, %s, %s)",
(hospital_id, source_type, language, url),
)
async def select_source_mainpage(hospital_id: str) -> dict | None:
return await fetchone(
"SELECT source_id, url FROM remote_source WHERE hospital_id = %s AND source_type = 'mainpage'",
(hospital_id,),
)
async def insert_raw_info(
source_id: int,
analysis_run_id: str,
data_tag: SourceType,
) -> int:
return await execute(
"INSERT INTO raw_info (source_id, analysis_run_id, data_tag) VALUES (%s, %s, %s)",
(source_id, analysis_run_id, data_tag),
)
async def update_raw_info_status(info_id: int, status: str) -> None:
await execute("UPDATE raw_info SET status = %s WHERE info_id = %s", (status, info_id))
async def update_raw_info(info_id: int, data: dict) -> None:
await execute(
"UPDATE raw_info SET raw_data = %s, status = 'done' WHERE info_id = %s",
(json.dumps(data, ensure_ascii=False), info_id),
)
async def select_raw_info_data(info_id: int | None) -> dict | None:
if info_id is None:
return None
row = await fetchone("SELECT raw_data FROM raw_info WHERE info_id = %s", (info_id,))
if not row or not row["raw_data"]:
return None
return json.loads(row["raw_data"]) if isinstance(row["raw_data"], str) else row["raw_data"]
async def select_run_sources(analysis_run_id: str) -> list[dict]:
return await fetchall(
"SELECT ri.info_id, rs.source_type, rs.url"
" FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s",
(analysis_run_id,),
)
async def select_run_raw_data(analysis_run_id: str) -> dict:
rows = await fetchall(
"SELECT rs.source_type, rs.language, ri.raw_data, ri.logo_url"
" FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s",
(analysis_run_id,),
)
result: dict = {}
for row in rows:
raw = row["raw_data"]
key = row["source_type"]
if (row.get("language") or "").upper() == "EN":
key = f"{key}_en"
data = json.loads(raw) if isinstance(raw, str) else (raw or {})
if isinstance(data, dict) and row.get("logo_url"):
data["_logo_url"] = row["logo_url"]
result[key] = data
return result
async def select_run_source_raw(
analysis_run_id: str, source_type: str, language: str | None = None,
) -> dict | None:
sql = (
"SELECT ri.raw_data FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s AND rs.source_type = %s"
)
args: tuple = (analysis_run_id, source_type)
if language:
sql += " AND rs.language = %s"
args = (*args, language)
sql += " LIMIT 1"
row = await fetchone(sql, args)
if not row or not row["raw_data"]:
return None
return json.loads(row["raw_data"]) if isinstance(row["raw_data"], str) else row["raw_data"]
async def update_raw_info_logo_url(info_id: int, logo_url: str) -> None:
"""raw_info.logo_url 컬럼에 로고 URL 저장 (JSON raw_data 와 분리해 컬럼 인덱스/조회 용이)."""
await execute(
"UPDATE raw_info SET logo_url = %s WHERE info_id = %s",
(logo_url, info_id),
)
async def select_branding_info_id(analysis_run_id: str) -> int | None:
row = await fetchone(
"SELECT ri.info_id FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s AND rs.source_type = 'branding' LIMIT 1",
(analysis_run_id,),
)
return (row or {}).get("info_id")
async def select_mainpage_logo_url(analysis_run_id: str) -> str | None:
row = await fetchone(
"SELECT ri.logo_url FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s AND rs.source_type = 'mainpage' LIMIT 1",
(analysis_run_id,),
)
return (row or {}).get("logo_url")
async def update_raw_info_merge(info_id: int, patch: dict) -> None:
"""raw_info.raw_data 를 read-modify-write 로 top-level 머지.
source 단계별로 (: branding brandAssets channelLogos) 키를 덧붙일 사용."""
row = await fetchone("SELECT raw_data FROM raw_info WHERE info_id = %s", (info_id,))
if not row:
return
raw = row["raw_data"]
data = json.loads(raw) if isinstance(raw, str) else (raw or {})
data.update(patch)
await execute(
"UPDATE raw_info SET raw_data = %s, status = 'done' WHERE info_id = %s",
(json.dumps(data, ensure_ascii=False), info_id),
)
async def select_source_by_type(
hospital_id: str, source_type: str, language: str | None = None,
) -> dict | None:
sql = "SELECT source_id, url FROM remote_source WHERE hospital_id = %s AND source_type = %s"
args: tuple = (hospital_id, source_type)
if language:
sql += " AND language = %s"
args = (*args, language)
sql += " LIMIT 1"
return await fetchone(sql, args)
async def select_run_mainpage_url(analysis_run_id: str) -> str:
row = await fetchone(
"SELECT rs.url FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s AND rs.source_type = 'mainpage'",
(analysis_run_id,),
)
return (row or {}).get("url") or ""

View File

@ -1,6 +1,7 @@
import os
import asyncio
import logging
from datetime import datetime, timezone
from http import HTTPMethod
import httpx
@ -9,6 +10,27 @@ logger = logging.getLogger(__name__)
REQUEST_TIMEOUT = 60
def parse_ts(v) -> datetime | None:
"""수집기마다 다른 timestamp 포맷을 통일된 datetime으로 변환.
파싱 실패 None.
"""
# 숫자면 epoch (Unix timestamp) — apify가 가끔 epoch로 줌
if isinstance(v, (int, float)):
return datetime.fromtimestamp(v, tz=timezone.utc)
if isinstance(v, str):
# 1순위: ISO 8601 (대부분 apify/firecrawl 출력)
try:
return datetime.fromisoformat(v.replace("Z", "+00:00"))
except ValueError:
pass
# 2순위: RFC 2822 (네이버 블로그 RSS 등 — 표준 라이브러리 파서로)
try:
from email.utils import parsedate_to_datetime
return parsedate_to_datetime(v)
except (TypeError, ValueError):
return None
return None
def get_env(key: str) -> str:
v = os.environ.get(key, "")
@ -61,6 +83,27 @@ def _normalize_homepage(url: str) -> str:
return u.rstrip("/")
# SSL 인증서가 www.* 에만 유효한 도메인 — bare 도메인이면 사용자 클릭 시 브라우저 SSL warning 뜸.
_WWW_REQUIRED = ("gangnamunni.com", "facebook.com", "instagram.com", "toxnfill.com")
def _with_scheme(u: str | None) -> str | None:
"""scheme 없는 URL에 https:// 보정 (수집기 파싱용). 빈 값은 None."""
return (u if "://" in u else "https://" + u) if u else None
"""scheme 없는 URL에 https:// 보정 (수집기/링크 표시용). 빈 값은 None.
+ 중첩된 https:// 끼어있으면 마지막 URL만 추출 (LLM이 가끔 'https://www.X/https://Y' 같이 만듦).
+ SSL 엄격 도메인(gangnamunni/facebook/instagram) www. 자동 보강."""
if not u:
return None
u = u.strip()
# 'https://www.facebook.com/https://facebook.com/X' 같은 중첩 → 마지막 'http(s)://' 부터 잘라 사용
last = max(u.rfind("https://"), u.rfind("http://"))
if last > 0:
u = u[last:]
if "://" not in u:
u = "https://" + u
# scheme 뒤가 www. 없이 SSL 엄격 도메인이면 www. 추가
for dom in _WWW_REQUIRED:
for scheme in ("https://", "http://"):
if u.startswith(scheme + dom):
u = scheme + "www." + u[len(scheme):]
break
return u

View File

@ -9,6 +9,13 @@ APIFY_BASE = "https://api.apify.com/v2"
IG_PROFILE_ACTOR = "coderx~instagram-profile-scraper-bio-posts"
IG_HIGHLIGHTS_ACTOR = "igview-owner~instagram-highlights-scraper"
# Facebook: pages + posts 두 actor 직접 호출.
FB_PAGES_ACTOR = "apify~facebook-pages-scraper"
FB_POSTS_ACTOR = "apify~facebook-posts-scraper"
# TikTok
TIKTOK_ACTOR = "clockworks~tiktok-scraper"
def _ig_username(url: str) -> str:
return urlparse(url).path.strip("/").split("/")[0] if "://" in url else url.lstrip("@")
@ -19,7 +26,7 @@ class ApifyClient:
self.token = token
self.wait_for_finish = wait_for_finish
async def _run_actor(self, actor_id: str, input_data: dict) -> list[dict]:
async def _run_actor(self, actor_id: str, input_data: dict, limit: int = 20) -> list[dict]:
resp = await http_request(
HTTPMethod.POST,
url=f"{APIFY_BASE}/acts/{actor_id}/runs",
@ -35,7 +42,7 @@ class ApifyClient:
items_resp = await http_request(
HTTPMethod.GET,
url=f"{APIFY_BASE}/datasets/{dataset_id}/items",
params={"token": self.token, "limit": 20},
params={"token": self.token, "limit": limit},
label=f"apify-dataset-{dataset_id}",
)
if not items_resp or not items_resp.is_success:
@ -61,6 +68,13 @@ class ApifyClient:
return None
if isinstance(highlights, Exception):
highlights = []
# 프로필상 하이라이트가 있다고 하면(highlight_reel_count>0) 빈 결과일 때 최대 2회 재시도.
if not highlights and (profile.get("highlight_reel_count", 0) or profile.get("highlightReelCount", 0)) > 0:
for _ in range(2):
retry = await self.fetch_instagram_highlights(username)
if retry:
highlights = retry
break
return {
"username": profile["username"],
"profileImage": profile.get("hdProfilePicUrl") or profile.get("profilePicUrl"),
@ -116,31 +130,52 @@ class ApifyClient:
# }
async def fetch_facebook_page(self, page_url: str) -> dict | None:
items = await self._run_actor("apify~facebook-pages-scraper", {"startUrls": [{"url": page_url}]})
items = await self._run_actor(FB_PAGES_ACTOR, {"startUrls": [{"url": page_url}]})
return items[0] if items else None
async def fetch_facebook_posts(self, page_url: str, limit: int = 20) -> list[dict]:
return await self._run_actor(
FB_POSTS_ACTOR, {"startUrls": [{"url": page_url}], "resultsLimit": limit}, limit=limit,
)
async def get_facebook_page(self, page_url: str) -> dict | None:
page = await self.fetch_facebook_page(page_url)
if not page:
# pages·posts 두 task 병렬 호출 (posts 실패해도 page만 있으면 진행)
page, posts = await asyncio.gather(
self.fetch_facebook_page(page_url),
self.fetch_facebook_posts(page_url),
return_exceptions=True,
)
if isinstance(page, Exception) or not page:
return None
if isinstance(posts, Exception):
posts = []
return {
"pageName": page.get("title") or page.get("name"),
"profileImage": page.get("profilePictureUrl") or page.get("profilePhoto") or page.get("profilePic"),
"pageUrl": page.get("pageUrl", page_url),
"followers": page.get("followers", 0),
"likes": page.get("likes", 0),
"following": page.get("followings", 0),
"reviews": page.get("ratingCount", 0),
"categories": page.get("categories", []),
"email": page.get("email"),
"phone": page.get("phone"),
"website": page.get("website"),
"address": page.get("address"),
"website": page.get("website") or page.get("websites"),
"intro": page.get("intro"),
"rating": page.get("rating"),
"latestPosts": [
{
"text": (p.get("text") or "")[:160],
"likes": p.get("likes", 0),
"reactions": p.get("topReactionsCount", 0),
"shares": p.get("shares", 0),
"views": p.get("viewsCount") or 0,
"isVideo": p.get("isVideo", False),
"timestamp": p.get("time") or p.get("timestamp"),
}
for p in (posts or []) if isinstance(p, dict)
],
}
async def fetch_tiktok_profile(self, url: str) -> list[dict]:
user = urlparse(url).path.strip("/").lstrip("@").split("/")[0] if "://" in url else url.lstrip("@")
return await self._run_actor("clockworks~tiktok-scraper", {
return await self._run_actor(TIKTOK_ACTOR, {
"profiles": [user],
"resultsPerPage": 10,
"profileScrapeSections": ["videos"],

View File

@ -1,250 +0,0 @@
"""홈페이지 HTML/CSS에서 hex 색상 직접 추출 + 빈도 기반 brand palette 산출.
Vision LLM에 의존하지 않고 페이지의 실제 CSS 값을 정규식으로 잡음.
로고만 분석하는 Vision보다 사이트 전체 컬러 시스템 (primary/secondary/background/text) 정확히 추출.
"""
import logging
import re
import ssl
from collections import Counter
from urllib.parse import urljoin, urlparse
import httpx
logger = logging.getLogger(__name__)
def _make_ssl_context() -> ssl.SSLContext:
"""오래된 한국 의료 사이트들이 SSL DH_KEY_TOO_SMALL / cipher 약함 등으로 차단되는 문제 우회.
보안 등급 1 낮춤 + cert 검증 유지."""
ctx = ssl.create_default_context()
try:
ctx.set_ciphers("DEFAULT@SECLEVEL=1")
except ssl.SSLError:
pass
return ctx
async def _fetch_html(url: str, timeout: float = 20.0) -> tuple[int, str]:
"""SSL/검증 단계별 fallback으로 HTML 받기. 그랜드/톡스앤필 같은 oldsite 대응."""
headers = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"}
# 1차: 표준 검증
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers) as c:
r = await c.get(url)
return r.status_code, r.text
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError) as e:
logger.info("[fetch] %s standard SSL failed: %s — fallback to weak cipher", url, e)
# 2차: 약한 cipher 허용
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers, verify=_make_ssl_context()) as c:
r = await c.get(url)
return r.status_code, r.text
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError) as e:
logger.info("[fetch] %s weak cipher failed: %s — fallback to verify=False", url, e)
# 3차: SSL 검증 끔 (host mismatch 등)
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers, verify=False) as c:
r = await c.get(url)
return r.status_code, r.text
except Exception as e:
logger.warning("[fetch] %s all fallbacks failed: %s", url, e)
return 0, ""
LOGO_IMG_PATTERNS = [
# 1) <img class="...logo..." src="...">
re.compile(r'<img[^>]*\bclass=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
# 2) <img src="..." class="...logo...">
re.compile(r'<img[^>]*\bsrc=["\']([^"\']+)["\'][^>]*\bclass=["\'][^"\']*\blogo\b[^"\']*["\']', re.IGNORECASE),
# 3) <img id="...logo..." src="...">
re.compile(r'<img[^>]*\bid=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
# 4) <img alt="...logo..." src="...">
re.compile(r'<img[^>]*\balt=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
# 5) <a/h1 class="logo"><...nested...><img src="...">
re.compile(r'<(?:a|h[1-6]|div|span)[^>]*\b(?:class|id)=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE | re.DOTALL),
# 6) inline background-image: <a/div class="logo" style="background-image: url(...)">
re.compile(r'<(?:a|div|span|h[1-6])[^>]*\b(?:class|id)=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bstyle=["\'][^"\']*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)', re.IGNORECASE),
# 7) inline background-image: <a/div style="background-image: url(...)" class="logo"> (속성 순서 반대)
re.compile(r'<(?:a|div|span|h[1-6])[^>]*\bstyle=["\'][^"\']*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)[^"\']*["\'][^>]*\b(?:class|id)=["\'][^"\']*\blogo\b', re.IGNORECASE),
# 8) src 자체에 "logo" 포함 (header_logo.png, brand-logo.svg 등)
re.compile(r'<img[^>]*\bsrc=["\']([^"\']*\blogo\b[^"\']*\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE),
# 9) <header>...<img src="..."> (헤더 영역 첫 img)
re.compile(r'<header\b[^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE | re.DOTALL),
# 10) <nav>...<img src="..."> (nav 영역 첫 img)
re.compile(r'<nav\b[^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE | re.DOTALL),
# 11) Open Graph image (대표 이미지) - 최후 fallback
re.compile(r'<meta[^>]*\bproperty=["\']og:image["\'][^>]*\bcontent=["\']([^"\']+)["\']', re.IGNORECASE),
re.compile(r'<meta[^>]*\bcontent=["\']([^"\']+)["\'][^>]*\bproperty=["\']og:image["\']', re.IGNORECASE),
]
# CSS 파일에서 .logo { background-image: url(...) } 추출용
LOGO_CSS_PATTERN = re.compile(
r'\.[\w-]*\blogo\b[\w-]*\s*(?:,\s*\.[\w-]+\s*)*\{[^}]*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)',
re.IGNORECASE | re.DOTALL,
)
def find_logo_url_in_html(html: str, base_url: str, css_texts: list[str] | None = None) -> str | None:
"""HTML에서 logo URL 찾기. class/id/alt → 부모 + 중첩 img → background-image → src에 logo → header/nav → og:image 순."""
for pat in LOGO_IMG_PATTERNS:
for m in pat.finditer(html):
src = m.group(1)
if not src or src.startswith("data:"):
continue
if re.search(r"(blank|spacer|pixel|transparent|1x1)\b", src, re.IGNORECASE):
continue
return urljoin(base_url, src)
# 외부 CSS에서 .logo background-image 추출
for css in (css_texts or []):
m = LOGO_CSS_PATTERN.search(css)
if m:
src = m.group(1)
if src and not src.startswith("data:"):
return urljoin(base_url, src)
return None
HEX6 = re.compile(r"#([0-9a-fA-F]{6})\b")
HEX3 = re.compile(r"#([0-9a-fA-F]{3})\b(?![0-9a-fA-F])")
RGB = re.compile(r"rgba?\(\s*(\d{1,3})\s*,\s*(\d{1,3})\s*,\s*(\d{1,3})\s*(?:,\s*[\d.]+\s*)?\)")
CSS_VAR_HEX = re.compile(r"--[\w-]+\s*:\s*(#[0-9a-fA-F]{3,8})", re.IGNORECASE)
CSS_LINK = re.compile(r'<link[^>]+rel=["\']stylesheet["\'][^>]+href=["\']([^"\']+)["\']', re.IGNORECASE)
STYLE_BLOCK = re.compile(r"<style[^>]*>(.*?)</style>", re.IGNORECASE | re.DOTALL)
# 무채색·아주 흔한 노이즈 컬러 (이런 건 brand color로 잡지 않음)
NOISE = {
"#ffffff", "#000000", "#fff", "#000",
"#333", "#222", "#111", "#444", "#555", "#666", "#777", "#888", "#999",
"#aaa", "#bbb", "#ccc", "#ddd", "#eee", "#f0f0f0", "#f5f5f5", "#fafafa",
}
def _normalize(hex_str: str) -> str:
h = hex_str.lstrip("#").lower()
if len(h) == 3:
h = "".join(c * 2 for c in h)
if len(h) == 8:
h = h[:6]
return f"#{h}"
def _rgb_to_hex(r: int, g: int, b: int) -> str:
return f"#{r:02x}{g:02x}{b:02x}"
def _hex_to_rgb(h: str) -> tuple[int, int, int]:
h = h.lstrip("#")
return int(h[0:2], 16), int(h[2:4], 16), int(h[4:6], 16)
def _distance(a: str, b: str) -> float:
ar, ag, ab = _hex_to_rgb(a)
br, bg, bb = _hex_to_rgb(b)
return ((ar - br) ** 2 + (ag - bg) ** 2 + (ab - bb) ** 2) ** 0.5
def _is_grayscale(h: str, tol: int = 12) -> bool:
r, g, b = _hex_to_rgb(h)
return max(r, g, b) - min(r, g, b) < tol
def _extract_hex(text: str) -> list[str]:
"""텍스트에서 모든 hex 색상 추출 (정규화)."""
out: list[str] = []
out.extend(_normalize(m.group(0)) for m in HEX6.finditer(text))
out.extend(_normalize(m.group(0)) for m in HEX3.finditer(text))
for m in RGB.finditer(text):
r, g, b = int(m.group(1)), int(m.group(2)), int(m.group(3))
if 0 <= r <= 255 and 0 <= g <= 255 and 0 <= b <= 255:
out.append(_rgb_to_hex(r, g, b))
return out
def _cluster(colors: Counter, threshold: float = 25.0) -> list[tuple[str, int]]:
"""비슷한 색은 묶음. 가장 빈도 높은 색을 대표로."""
ranked = colors.most_common()
clusters: list[tuple[str, int]] = []
for color, count in ranked:
merged = False
for i, (rep, rep_count) in enumerate(clusters):
if _distance(color, rep) < threshold:
clusters[i] = (rep, rep_count + count)
merged = True
break
if not merged:
clusters.append((color, count))
return clusters
async def _fetch_html_and_css(homepage_url: str, max_css_files: int = 8) -> tuple[str, list[str]]:
"""홈페이지 HTML + 외부 CSS(Top N)를 한 번에 fetch. 로고/색상 추출이 사이트를 중복으로 긁지 않도록 공유.
_fetch_html이 SSL 약함/host mismatch까지 fallback 처리. 실패 ("", [])."""
status, html = await _fetch_html(homepage_url)
if status != 200 or not html:
logger.warning("[color_extractor] homepage fetch failed status=%s url=%s", status, homepage_url)
return "", []
css_texts: list[str] = []
for css_href in CSS_LINK.findall(html)[:max_css_files]:
cstatus, ctext = await _fetch_html(urljoin(homepage_url, css_href), timeout=15.0)
if cstatus == 200 and ctext:
css_texts.append(ctext)
return html, css_texts
def _colors_from_text(html: str, css_texts: list[str], source_url: str = "") -> dict:
"""이미 받아온 HTML + CSS 텍스트에서 hex 빈도 분석 → primary/accent/text + palette. (fetch 없음, 순수 계산)"""
# 1. HTML 내 <style> 블록 + 통째(inline style="color:#...") + 외부 CSS
all_text_chunks: list[str] = list(STYLE_BLOCK.findall(html))
all_text_chunks.append(html)
all_text_chunks.extend(css_texts)
# 2. 모든 hex 추출 (NOISE 제외)
counter: Counter = Counter()
for text in all_text_chunks:
for color in _extract_hex(text):
if color in NOISE:
continue
counter[color] += 1
if not counter:
logger.info("[color_extractor] no colors extracted from %s", source_url)
return {}
# 3. 비슷한 색 클러스터링
clustered = _cluster(counter)
# 4. primary = 빈도 높은 채도 있는 색 / accent = 두번째 채도 있는 색 / text = 빈도 높은 무채색
chromatic = [c for c, _ in clustered if not _is_grayscale(c)]
grayscale = [c for c, _ in clustered if _is_grayscale(c)]
palette_top = clustered[:8]
palette = [{"name": f"색상 {i+1}", "hex": h, "usage": f"빈도 {n}"} for i, (h, n) in enumerate(palette_top)]
return {
"brand_colors": {
"primary": chromatic[0] if chromatic else None,
"accent": chromatic[1] if len(chromatic) > 1 else None,
"text": grayscale[0] if grayscale else None,
},
"color_palette": palette,
"extracted_from": "html+css",
}
async def extract_brand_colors_from_site(homepage_url: str, max_css_files: int = 8) -> dict:
"""홈페이지 HTML + 외부 CSS fetch → hex 색상 빈도 분석 → primary/accent/text + palette 5종."""
html, css_texts = await _fetch_html_and_css(homepage_url, max_css_files)
if not html:
return {}
return _colors_from_text(html, css_texts, homepage_url)
async def extract_brand_assets_from_site(homepage_url: str, max_css_files: int = 8) -> dict:
"""사이트를 한 번만 fetch해서 로고 URL과 brand 색상을 함께 추출.
반환: {"logo_url": str | None, "colors": {brand_colors, color_palette, ...} | {}}"""
html, css_texts = await _fetch_html_and_css(homepage_url, max_css_files)
if not html:
return {"logo_url": None, "colors": {}}
return {
"logo_url": find_logo_url_in_html(html, homepage_url, css_texts=css_texts),
"colors": _colors_from_text(html, css_texts, homepage_url),
}

View File

@ -0,0 +1,334 @@
"""Gemini Vision — 로고/브랜드 비주얼 자동 분석 (OpenAI 호환 모드).
정확한 hex 색상은 color_extractor가 CSS에서 직접 뽑음 (Vision은 근사값밖에 ).
Vision은 사람이 봐야 있는 정성 정보 심볼 형태/워드마크/ 담당.
"""
import asyncio
import base64
import json
import logging
import re
import ssl
import httpx
import resvg_py
from openai import AsyncOpenAI
logger = logging.getLogger(__name__)
DEFAULT_MODEL = "gemini-2.5-flash"
class VisionClient:
"""Gemini Vision을 OpenAI 호환 endpoint로 호출. GEMINI_API_KEY만 필요."""
def __init__(self, api_key: str, model: str = DEFAULT_MODEL, timeout: float = 30.0, max_retries: int = 2):
self.client = AsyncOpenAI(
api_key=api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
timeout=timeout,
max_retries=max_retries,
)
self.model = model
@staticmethod
def _extract_json(text: str) -> dict | None:
if not text:
return None
m = re.search(r"```(?:json)?\s*(\{.*?\})\s*```", text, re.DOTALL)
if m:
try:
return json.loads(m.group(1))
except json.JSONDecodeError:
pass
m = re.search(r"\{.*\}", text, re.DOTALL)
if m:
try:
return json.loads(m.group(0))
except json.JSONDecodeError:
return None
return None
@staticmethod
async def _fetch_as_data_url(url: str) -> str | None:
"""Gemini는 URL 직접 fetch가 막힌 호스트가 많아 base64 인라인으로 변환.
+ 'image does not exist' 같은 placeholder 이미지 거부 (작은 bytes / 잘못된 content-type).
+ 한국 의료 사이트 SSL이 약해서 표준 검증에 실패하는 대응 (3 SSL fallback)."""
headers = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"}
def _weak_ctx() -> ssl.SSLContext:
ctx = ssl.create_default_context()
try:
ctx.set_ciphers("DEFAULT@SECLEVEL=1")
except ssl.SSLError:
pass
return ctx
last_err: Exception | None = None
for verify in (True, _weak_ctx(), False):
try:
async with httpx.AsyncClient(
timeout=15.0, follow_redirects=True, headers=headers, verify=verify,
) as c:
resp = await c.get(url)
if resp.status_code != 200:
logger.warning("[vision] fetch %s status=%s", url, resp.status_code)
return None
mime = resp.headers.get("content-type", "").split(";")[0].strip()
# 실제 이미지가 아니면 거부 (HTML 페이지가 404 대신 200으로 리다이렉트 되는 경우)
if not mime.startswith("image/"):
logger.warning("[vision] %s not an image (content-type=%s)", url, mime)
return None
# SVG는 Gemini가 못 보므로 즉시 PNG로 래스터화 (resvg, in-memory ~1ms)
content = resp.content
if mime == "image/svg+xml" or url.lower().split("?")[0].endswith(".svg"):
try:
content = bytes(resvg_py.svg_to_bytes(svg_string=resp.text))
mime = "image/png"
except Exception as e:
logger.warning("[vision] svg rasterize failed %s: %s", url, e)
return None
size = len(content)
if size < 500:
logger.warning("[vision] %s too small (%d bytes) — likely placeholder", url, size)
return None
b64 = base64.b64encode(content).decode("ascii")
return f"data:{mime};base64,{b64}"
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError) as e:
last_err = e
continue
except Exception as e:
logger.warning("[vision] fetch error %s: %s", url, e)
return None
logger.warning("[vision] fetch %s SSL fallback all failed: %s", url, last_err)
return None
async def _ask(self, image_urls: list[str], prompt: str, max_tokens: int = 4000) -> dict | None:
content: list[dict] = []
for u in image_urls:
if not u:
continue
data_url = await self._fetch_as_data_url(u)
if not data_url:
continue
content.append({"type": "image_url", "image_url": {"url": data_url}})
if not any(c.get("type") == "image_url" for c in content):
logger.warning("[vision] no images could be fetched")
return None
content.append({"type": "text", "text": prompt})
try:
resp = await self.client.chat.completions.create(
model=self.model,
messages=[{"role": "user", "content": content}],
max_tokens=max_tokens,
)
choice = resp.choices[0]
if choice.finish_reason != "stop":
logger.warning("[vision] unexpected finish_reason=%s", choice.finish_reason)
return self._extract_json(choice.message.content or "")
except Exception as e:
logger.warning("[vision] error: %s", e)
return None
async def describe_svg_text(self, svg_url: str) -> dict | None:
"""SVG는 Gemini Vision이 못 보지만 XML 텍스트 자체는 LLM이 읽을 수 있음.
SVG 소스를 받아 그대로 text endpoint에 던지고 ·심볼·텍스트를 추론하게 .
analyze_brand_assets와 동일한 스키마(logo_description/style/has_symbol/...) 반환."""
headers = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"}
def _weak_ctx() -> ssl.SSLContext:
ctx = ssl.create_default_context()
try:
ctx.set_ciphers("DEFAULT@SECLEVEL=1")
except ssl.SSLError:
pass
return ctx
svg_text: str | None = None
for verify in (True, _weak_ctx(), False):
try:
async with httpx.AsyncClient(
timeout=15.0, follow_redirects=True, headers=headers, verify=verify,
) as c:
resp = await c.get(svg_url)
if resp.status_code == 200:
svg_text = resp.text
break
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError):
continue
except Exception as e:
logger.warning("[vision] svg fetch error %s: %s", svg_url, e)
return None
if not svg_text:
logger.warning("[vision] svg fetch failed %s", svg_url)
return None
# 페이로드 폭주 방지 — 평범한 로고 SVG는 수 KB 수준
if len(svg_text) > 60000:
svg_text = svg_text[:60000]
prompt = (
"아래는 병원 로고 SVG 소스 코드입니다. SVG 마크업(path/circle/text/fill/stroke 등)을 "
"읽고 로고의 시각적 특징을 추론해 아래 JSON 스키마로만 응답하세요. 코드펜스 없이 순수 JSON.\n"
"{\n"
' "logo_description": "심볼 형태 + 워드마크 + 톤을 1~2문장 한국어로",\n'
' "logo_style": "minimal | illustrative | typographic | abstract 중 하나",\n'
' "has_symbol": "심볼/아이콘이 있으면 true, 글자만 있으면 false (boolean)",\n'
' "logo_symbol": "심볼 묘사 (예: \'잎사귀\'). 없으면 빈 문자열",\n'
' "logo_text": "워드마크 텍스트 그대로. <text> 태그 내용 우선",\n'
' "logo_colors_desc": "쓰인 색감을 사람이 부르는 이름으로 (예: \'딥네이비 + 골드\'). hex 출력 금지"\n'
"}\n"
"주의: hex 값이나 URL은 출력하지 마세요 (별도 추출 로직 처리). 모든 텍스트는 한국어로.\n\n"
"SVG 소스:\n"
f"{svg_text}"
)
try:
resp = await self.client.chat.completions.create(
model=self.model,
messages=[{"role": "user", "content": prompt}],
max_tokens=8000, # Gemini 2.5는 thinking 토큰을 max_tokens에서 차감하므로 여유 필요
)
choice = resp.choices[0]
if choice.finish_reason != "stop":
logger.warning("[vision] svg describe finish_reason=%s", choice.finish_reason)
result = self._extract_json(choice.message.content or "")
except Exception as e:
logger.warning("[vision] svg describe error: %s", e)
return None
if not result:
return None
result["logo_images"] = {"circle": None, "horizontal": svg_url, "korean": None}
return result
async def analyze_brand_assets(
self,
logo_url: str | None,
homepage_url: str | None,
additional_images: list[str] | None = None,
) -> dict:
"""로고 이미지를 보고 정성 분석. 정확한 hex는 color_extractor가 따로 처리하므로 여기선 안 뽑음."""
urls = [u for u in [logo_url] + list(additional_images or []) if u]
if not urls:
return {}
prompt = (
"당신은 브랜드 로고 시각 분석가입니다. 첨부된 이미지(첫 번째가 병원의 대표 로고)를 보고 "
"아래 JSON 스키마로만 응답하세요. 코드펜스 없이 순수 JSON만 출력.\n"
"{\n"
' "logo_description": "로고를 1~2문장으로 설명 (심볼 형태 + 워드마크 + 전반적 톤). 예: \'둥근 잎사귀를 감싼 추상 심볼에 세리프 한글 워드마크, 차분하고 고급스러운 톤\'",\n'
' "logo_style": "minimal | illustrative | typographic | abstract 중 하나",\n'
' "has_symbol": "심볼/아이콘이 있으면 true, 글자만 있으면 false (boolean)",\n'
' "logo_symbol": "심볼이 묘사하는 대상 (예: \'잎사귀\', \'추상 곡선\'). 없으면 빈 문자열",\n'
' "logo_text": "로고에 보이는 워드마크 텍스트 그대로 (한글/영문). 없으면 빈 문자열",\n'
' "logo_colors_desc": "로고에 쓰인 색감을 사람이 부르는 이름으로 서술 (예: \'딥네이비 + 골드\')",\n'
' "logo_colors_hex": ["로고에서 시각적으로 두드러진 색 정확히 5개의 hex 근사값 배열. 예: [\'#1A2B3C\', \'#D4A017\', \'#FFFFFF\', \'#9E5C2A\', \'#1F1F1F\']. 강한 색이 5개 안 되면 음영/명도 차이로 5개 채울 것. 빈 배열 금지."]\n'
"}\n"
"주의: logo_colors_hex 는 시각 추정이라 정확도 떨어질 수 있음. CSS 추출이 우선이고 이건 fallback/보완 용.\n"
"모든 설명/텍스트 값은 반드시 한국어로 작성하세요 (영어 금지)."
)
result = await self._ask(urls, prompt)
if not result:
return {}
# logo_images는 우리가 직접 채움 (Vision은 묘사만)
result["logo_images"] = {"circle": None, "horizontal": logo_url, "korean": None}
# logo_colors_hex 5개 강제 정규화 — LLM 이 4개나 6개 줄 수도 있어서 길이 fallback.
hex_list = [h for h in (result.get("logo_colors_hex") or []) if isinstance(h, str) and h.startswith("#")]
if hex_list:
while len(hex_list) < 5:
hex_list.append(hex_list[-1]) # 마지막 색 복제로 패딩
result["logo_colors_hex"] = hex_list[:5]
else:
result["logo_colors_hex"] = []
return result
async def describe_channel_logos(
self,
official_logo_url: str | None,
channel_logos: list[dict],
) -> dict | None:
"""채널별 프로필 이미지(로고)를 보고 각각 설명 + 공식 로고와 일치 여부 평가.
channel_logos: [{"channel": "Instagram", "url": "..."}, ...]
반환: {"channel_logos": [{"channel","logo_description","is_official"}], "inconsistency_summary", "recommendation"}
**3채널씩 묶어 병렬 호출** ( 번에 묶으면 LLM이 채널-이미지 매칭 헷갈려 같은 묘사를
여러 채널에 복사하는 문제 VIEW 한국페북·영문인스타가 "공식 로고" 묘사로 잘못
박혔던 케이스 있어서 분리. 1채널씩 N번보다 가성비 좋음)."""
items = [c for c in channel_logos if c.get("url")]
if not items:
return None
CHUNK = 3
async def _chunk(batch: list[dict]) -> list[dict]:
urls = [official_logo_url] + [c["url"] for c in batch] if official_logo_url else [c["url"] for c in batch]
n = len(batch)
# 이미지 번호 ↔ 채널 매핑 명시
if official_logo_url:
mapping = "이미지 1 = 공식 로고\n" + "\n".join(
f"이미지 {i+2} = {c.get('channel','?')} 채널 프로필" for i, c in enumerate(batch)
)
instruction = (
f"{mapping}\n\n"
f"이미지 2~{n+1}(채널 프로필 {n}개)을 각각 **그 이미지에 실제로 보이는 그대로** "
"한국어 1문장으로 묘사하세요 (색·형태·텍스트·배경 그대로).\n"
"❗ 공식 로고(이미지 1) 묘사를 절대 복사하지 마세요. 각 채널 이미지에 보이는 실제 특징만.\n"
"각 채널이 공식 로고와 시각적으로 거의 동일하면 is_official=true, "
"심볼/색/배경/텍스트가 다르거나 모델 사진이면 false.\n"
)
else:
mapping = "\n".join(f"이미지 {i+1} = {c.get('channel','?')} 채널 프로필" for i, c in enumerate(batch))
instruction = (
f"{mapping}\n\n"
f"각 이미지를 보이는 그대로 한국어 1문장으로 묘사 (색·형태·텍스트·배경).\n"
)
schema_lines = ",\n".join(
f' {{"channel": "{c.get("channel","?")}", "logo_description": "...", "is_official": true}}'
for c in batch
)
p = (
instruction
+ "\n아래 JSON으로만 응답 (코드펜스 없이, 순수 JSON):\n{\n"
+ f' "channel_logos": [\n{schema_lines}\n ]\n'
+ "}\n"
+ f"channel 필드는 위 매핑 그대로 ({', '.join(c.get('channel','?') for c in batch)}). "
+ "logo_description은 반드시 한국어 (영어 금지)."
)
r = await self._ask(urls, p)
if not r:
return []
out = []
for c in r.get("channel_logos", []):
out.append({
"channel": c.get("channel", ""),
"logo_description": c.get("logo_description", ""),
"is_official": bool(c.get("is_official", False)) if official_logo_url else None,
})
return out
# 3개씩 청크 → 병렬
chunks = [items[i:i+CHUNK] for i in range(0, len(items), CHUNK)]
results = await asyncio.gather(*[_chunk(b) for b in chunks], return_exceptions=True)
channel_logos_out: list[dict] = []
for r in results:
if isinstance(r, Exception):
logger.warning("[vision] channel_logo chunk error: %s", r)
continue
channel_logos_out.extend(r)
if not channel_logos_out:
return None
# 일관성 요약 + 권고는 결정적 산출 (LLM 한번 더 안 부름)
if official_logo_url:
mismatches = [c["channel"] for c in channel_logos_out if not c.get("is_official")]
if not mismatches:
summary = "모든 채널이 공식 로고를 일관되게 사용하고 있습니다."
rec = "현재 일관성 유지."
else:
summary = f"{len(mismatches)}개 채널({', '.join(mismatches)})이 공식 로고와 다른 이미지를 사용해 브랜드 일관성이 부족합니다."
rec = "비공식 채널 프로필을 공식 로고로 통일 권고."
else:
summary, rec = "", ""
return {
"channel_logos": channel_logos_out,
"inconsistency_summary": summary,
"recommendation": rec,
}

View File

@ -1,7 +1,7 @@
import os
from pydantic import BaseModel
from common.utils import get_env
from integrations.llm.schemas.report import ReportInput, ReportOutput
from integrations.llm.schemas.report import ReportInput, ReportOutput, YouTubeDiagnosisInput, YouTubeDiagnosisOutput
from integrations.llm.schemas.plan import PlanInput, PlanOutput
from integrations.llm.schemas.market import (
MarketCompetitorsInput, MarketCompetitorsOutput,
@ -80,3 +80,10 @@ market_target_audience_prompt = Prompt(
input_class=MarketTargetAudienceInput,
output_class=MarketTargetAudienceOutput,
)
youtube_diagnosis_prompt = Prompt(
file_name="youtube_diagnosis_prompt.txt",
prompt_model="REPORT_MODEL",
input_class=YouTubeDiagnosisInput,
output_class=YouTubeDiagnosisOutput,
)

View File

@ -18,6 +18,9 @@ class PlanInput(BaseModel):
tiktok: str | None = None
instagram_en: str | None = None
facebook_en: str | None = None
naver_blog: str | None = None
naver_cafe: str | None = None
kakao_talk: str | None = None
channel_logos: str | None = None
brand_assets: str | None = None
@ -56,7 +59,7 @@ class ChannelBrandingRule(BaseModel):
profile_photo: str
banner_spec: str
bio_template: str
current_status: Literal["correct", "incorrect", "missing"]
current_status: Literal["correct", "incorrect", "N/A"]
class BrandPlanInconsistencyValue(BaseModel):

View File

@ -68,22 +68,18 @@ class RegistryData(BaseModel):
class ClinicSnapshot(BaseModel):
name: str
name_en: str
established: str
years_in_business: int
staff_count: int
lead_doctor: LeadDoctor
overall_rating: float
total_reviews: int
price_range: PriceRange
certifications: list[str]
media_appearances: list[str]
medical_tourism: list[str]
location: str
nearest_station: str
phone: str
domain: str
# _build_clinic_snapshot은 source 데이터 있을 때만 채움 (`if x:` 가드).
# required면 강남언니/홈페이지 누락 병원에서 ValidationError로 리포트 실패.
name: str | None = None
name_en: str | None = None
staff_count: int | None = None
lead_doctor: LeadDoctor | None = None
overall_rating: float | None = None
total_reviews: int | None = None
certifications: list[str] = []
location: str | None = None
phone: str | None = None
domain: str | None = None
logo_images: LogoImages | None = None
brand_colors: BrandColors | None = None
source: DataSource | None = None
@ -137,7 +133,6 @@ class YouTubeAudit(BaseModel):
avg_video_length: str
upload_frequency: str
channel_created_date: str
subscriber_rank: str
channel_description: str
linked_urls: list[LinkedUrl]
playlists: list[str]
@ -164,8 +159,8 @@ class InstagramAccount(BaseModel):
class InstagramAudit(BaseModel):
accounts: list[InstagramAccount]
diagnosis: list[DiagnosisItem]
accounts: list[InstagramAccount] = []
diagnosis: list[DiagnosisItem] = []
# --- Facebook ---
@ -198,17 +193,17 @@ class FacebookPage(BaseModel):
linked_domain: str
reviews: int
recent_post_age: str
has_whatsapp: bool
post_frequency: str | None = None
has_whatsapp: bool | None = None
post_frequency: str
top_content_type: str | None = None
engagement: str | None = None
engagement: str
class FacebookAudit(BaseModel):
pages: list[FacebookPage]
diagnosis: list[DiagnosisItem]
brand_inconsistencies: list[BrandInconsistency]
consolidation_recommendation: str
pages: list[FacebookPage] = []
diagnosis: list[DiagnosisItem] = []
brand_inconsistencies: list[BrandInconsistency] = []
consolidation_recommendation: str | None = None
# --- 기타 채널 / 웹사이트 ---
@ -326,6 +321,8 @@ class ReportInput(BaseModel):
tiktok: str | None = None
instagram_en: str | None = None
facebook_en: str | None = None
kakao_talk: str | None = None
naver_cafe: str | None = None
channel_logos: str | None = None
@ -351,3 +348,20 @@ class MarketingReport(BaseModel):
ReportOutput = MarketingReport
# --- YouTubeDiagnosis ---
class YouTubeDiagnosisInput(BaseModel):
channel_name: str | None = None
subscribers: int | None = None
total_videos: int | None = None
total_views: int | None = None
avg_video_length: str | None = None
upload_frequency: str | None = None
top_videos: str | None = None
playlists: str | None = None
class YouTubeDiagnosisOutput(BaseModel):
diagnosis: list[DiagnosisItem]

View File

@ -32,8 +32,11 @@
## 분석 리포트
{report}
## 추가 채널 데이터 (틱톡 / 인스타그램 EN / 페이스북 EN)
아래에 데이터가 있는 채널은 channelStrategies와 channelBranding에 **반드시 포함**하세요 (틱톡, 영문 인스타그램, 영문 페이스북). null이면 제외.
## 추가 채널 데이터 (네이버 블로그 / 틱톡 / 인스타그램 EN / 페이스북 EN / 네이버 카페 / 카카오톡)
아래에 데이터가 있는 채널은 channelStrategies에 **반드시 포함**하세요 (네이버 블로그, 틱톡, 영문 인스타그램, 영문 페이스북, 네이버 카페, 카카오톡). channelBranding은 SNS·블로그·카페까지만 포함(카카오톡은 메신저라 제외). null이면 제외.
### 네이버 블로그 (Naver Blog)
{naver_blog}
### 틱톡 (TikTok)
{tiktok}
@ -44,10 +47,25 @@
### 페이스북 (영문 페이지)
{facebook_en}
### 네이버 카페 (공식 카페 운영 신호)
{naver_cafe}
- naver_cafe.cafeName: 카페명, naver_cafe.memberCount: 회원수
- currentStatus는 "회원 N명" 형태로 간단하게. 게시글 수·최근 활동은 수집 불가 (추측 금지).
- targetGoal은 회원 확보 목표 수치 + 운영 권장 (예: "회원 5,000명, 주 1~2회 공지 발행").
### 카카오톡 채널 (URL only — 콘텐츠 수집 X, 존재 여부만)
{kakao_talk}
- channelStrategies 카드 하나로 포함. currentStatus는 "공식 카카오톡 채널 운영 중" 정도, targetGoal은 친구 추가 유도·상담 전환·자동응답 시나리오 구체화 등.
## 채널별 로고 분석 (Gemini Vision) — 채널룰/일관성의 근거
{channel_logos}
- 위 channel_logos[]의 각 항목: channel(채널명), logo_description(프로필이 어떻게 생겼는지), is_official(공식 로고와 일치 여부).
- **channelBranding[]를 이 데이터로 채우세요**: 채널별로 profilePhoto=해당 채널의 logo_description, currentStatus=is_official이 true면 "correct" / false면 "incorrect" (데이터 없는 채널은 "missing"). bannerSpec은 권장 배너 규격(크기/디자인)을 작성.
- **channelBranding[]은 "어떻게 해야 하는지 권장 가이드라인" 섹션입니다.** 채널 통일 전략 기준으로 권장값 박을 것:
- profilePhoto: **빈 문자열 ""로 두세요.** 시스템이 brand_assets.logo_description으로 직접 채우므로 LLM은 만들지 마세요.
- bannerSpec: 권장 배너 규격 (크기·디자인 가이드)
- bioTemplate: 권장 bio 템플릿 (구조·필수 요소·예약 링크 포함 여부)
- currentStatus: is_official=true면 "correct" / false면 "incorrect" (데이터 없는 채널은 "N/A") — 현재 상태 마커는 이 필드 하나로만.
- 현재 채널 프로필 이미지의 실제 묘사(channel_logos.channel_logos[].logo_description)는 brandInconsistencies에서만 사용. channelBranding에서 채널별로 다른 묘사를 박지 마세요.
- **brandInconsistencies[]에 "로고" 항목을 반드시 만드세요**: values[]에 채널마다 channel(채널명) / value(logo_description 그대로) / is_correct(is_official 값) 세 필드를 넣고, impact는 inconsistency_summary, recommendation은 channel_logos.recommendation 기반으로 작성 (공식 로고로 통일 권고 포함).
## 브랜드 자산 (홈페이지 CSS에서 추출 — 결정적 데이터)
@ -69,10 +87,10 @@
- brandInconsistencies: 채널 간 브랜딩 불일치 항목 및 개선 권고
### Section 2: channelStrategies
- 리포트에 데이터가 있는 채널만 포함
- **currentStatus는 현재 채널 상태를 실제 수치로 서술** (예: "14,047 팔로워, Reels 0개", "104K 구독자, 주 2~3회 업로드"). `excellent`/`warning`/`good` 같은 등급·평가어를 절대 쓰지 마세요.
- targetGoal은 구체적 목표 수치로 작성 (예: "50K 팔로워, Reels 주 5개")
- 각 채널의 우선순위(P0/P1/P2), 콘텐츠 유형, 게시 빈도, 포맷 가이드라인 작성
- 메인 SNS 채널(Instagram, Facebook, YouTube, TikTok, 네이버 블로그) + 영문 계정(Instagram EN, Facebook EN) + **네이버 카페 / 카카오톡** (URL 있을 때) 카드를 **모두 포함**. 데이터 없는 채널도 빠뜨리지 말 것.
- **currentStatus**: 데이터 있는 채널은 실제 수치로 서술 (예: "14,047 팔로워, Reels 0개", "104K 구독자, 주 2~3회 업로드"). **데이터 없는 채널은 "계정 없음"** 으로 표시. `excellent`/`warning`/`good` 같은 등급·평가어 금지.
- **targetGoal은 모든 채널에 반드시 채울 것** — 구체적 목표 수치(예: "50K 팔로워, Reels 주 5개"). 데이터 없는 채널도 시작 시 권장 목표를 작성하고 비우지 말 것.
- 각 채널의 우선순위(P0/P1/P2), 콘텐츠 유형, 게시 빈도, 포맷 가이드라인 모두 권장값으로 작성 — 데이터 없어도 시작 권장값으로 채울 것.
- customerJourneyStage는 해당 채널의 주요 기여 단계로 설정
### Section 3: contentStrategy

View File

@ -63,26 +63,43 @@
### 페이스북 (영문 페이지)
{facebook_en}
### 카카오톡 채널 (URL only — 수집 데이터 없음, 존재 여부만 확인)
{kakao_talk}
### 네이버 카페 (공식 카페 운영 신호)
{naver_cafe}
- naver_cafe.cafeName: 카페명
- naver_cafe.memberCount: 회원수
- 게시글 총 수·최근 게시일은 로그인 필요라 수집 불가. 추측 금지. 위 두 값만 사용.
### 채널별 로고 분석 (Gemini Vision)
{channel_logos}
- channel_logos.channel_logos[]에 각 채널의 로고 설명(logo_description)과 공식 로고 일치 여부(is_official)가 있습니다.
- **facebook_audit.pages[].logo** 는 짧은 판정 타이틀로: is_official=true면 `"일치 (공식 로고)"`, false면 `"불일치 (비공식 변형)"`. 그리고 **facebook_audit.pages[].logo_description** 에 해당 채널의 logo_description(설명문)을 넣으세요.
- 위 값들은 channel_logos 데이터 기반으로만 작성하고 추측하지 마세요.
- 채널 간 로고 불일치(is_official=false)는 brand 일관성 진단(problem_diagnosis/weaknesses)에 반영하세요.
- **brand_inconsistencies[]에 "로고" 항목을 반드시 만드세요**: values[]에 channel_logos.channel_logos[] 각 채널마다 다음 3필드를 **그대로** 박을 것 — channel(채널명 그대로), value(해당 채널의 logo_description 문자열 그대로 복붙), is_correct(해당 채널의 is_official 값 그대로). ❗ **채널-묘사 매핑을 절대 swap·재해석·임의 변형 금지**. channel_logos에 적힌 그대로 사용. impact는 channel_logos.inconsistency_summary 사용, recommendation은 channel_logos.recommendation 사용.
## clinic_snapshot / 채널 audit 작성 지침 (수집 데이터 그대로, 추측 금지)
- clinic_snapshot.name 은 {clinic_name} 을 **그대로** 사용 (강남언니 표기명 '-본원' 등으로 바꾸지 말 것).
- clinic_snapshot 의 overall_rating/total_reviews/staff_count/location/certifications/lead_doctor 는 강남언니({gangnam_unni}) 데이터의 값을 그대로 사용.
- **instagram_audit.accounts 는 반드시 빈 배열 []로 두세요.** 계정 정보는 시스템이 수집 데이터로 직접 채우니 LLM은 만들지 말고, instagram_audit.diagnosis(진단)만 작성하세요.
- facebook_audit.pages: KR 페북({facebook})·영문 페북({facebook_en}) 데이터가 있으면 **각각 별도 페이지**로 넣고, url/page_name/followers 등은 그 데이터 그대로. language/label 동일 규칙.
- facebook_audit.pages[].top_content_type 은 해당 페이지 latestPosts의 **캡션·미디어를 읽고** 주로 올리는 콘텐츠를 의미 기반으로 짧게 묘사하세요 (예: "Before/After 사진 + 환자 여정 Reels", "이벤트·프로모션 카드뉴스", "다국어 시술 소개"). 단순 "동영상/이미지 위주"가 아니라 **무슨 주제**인지 쓰세요. (recent_post_age·post_frequency·engagement 수치는 시스템이 덮어쓰니 대략 적어도 됩니다.)
- 위 수치·URL·이름은 제공된 데이터에서 그대로 쓰고 절대 지어내지 마세요.
## 기타 채널 현황 (other_channels) 작성 지침
- other_channels에는 메인 audit(YouTube/Instagram/Facebook/Website)에 **포함되지 않은** 채널만 넣으세요.
- 위 '채널 데이터'에 **실제 수집된 데이터가 있는 채널만** status=active와 실제 url로 일관되게 포함: 네이버 블로그, 강남언니, 틱톡, 영문 인스타그램({instagram_en}), 영문 페이스북({facebook_en}).
- **영문 인스타그램·영문 페이스북은 KR 메인 audit(Instagram/Facebook)과 별개 계정이므로, 데이터가 있으면 반드시 other_channels에 "Instagram EN" / "Facebook EN"으로 각각 포함하세요 (절대 누락 금지).**
- **수집 데이터에 없는 채널(카카오톡/네이버플레이스/네이버카페/Threads 등)은 절대 임의로 만들지 마세요.** 데이터 없으면 그 채널은 생략 (랜덤 생성·추측 금지).
- **카카오톡·네이버 카페**: {kakao_talk} 또는 {naver_cafe}에 url이 있으면 other_channels에 각각 "KakaoTalk" / "Naver Cafe"로 status=active + 해당 url로 포함. 수집된 콘텐츠 데이터는 없으므로 URL 존재 자체가 활성 채널 신호. **둘 다 null/빈 값이면 절대 만들지 마세요.**
- **그 외 데이터 없는 채널(네이버플레이스/Threads 등)은 절대 임의로 만들지 마세요.** 데이터 없으면 그 채널은 생략 (랜덤 생성·추측 금지).
- url은 수집 데이터의 실제 URL만 사용. 없으면 빈 문자열.
- **URL에 'https://www.facebook.com/' 같은 prefix를 절대 직접 만들지 마세요.** 수집 데이터의 URL을 그대로 사용. 이미 'https://...' 가 붙은 URL에 또 prefix 붙이면 'https://www.facebook.com/https://facebook.com/X' 같이 깨집니다. 받은 URL = 출력 URL.
## registry_data 작성 지침 (clinic_snapshot 안)
- **registry_data.website_en / district / branches / brand_group / naver_place_url / gangnam_unni_url / google_maps_url 모두 제공된 데이터에 명시되지 않으면 반드시 null로 두세요.**
- 영문 사이트 URL, 영문명, 지점 정보 같은 거 데이터에 없으면 **절대 추측하거나 그럴듯해 보이는 도메인을 지어내지 마세요** (예: 'thepsclinic.com', '*-eng.com' 같은 거).
## 분석 지침
@ -94,5 +111,5 @@
- 데이터가 null인 계정은 항목을 만들지 마세요. icon은 instagram/facebook/video 등 플랫폼에 맞게 설정.
- strengths와 weaknesses는 각 3개 이상 작성하세요.
- roadmap은 우선순위 순으로 실행 가능한 액션으로 작성하세요.
- kpis는 실제 수집된 수치 기반으로 현실적인 측정 가능 지표로 작성하세요.
- kpi_dashboard는 코드가 결정적으로 산출해 후처리 강제 치환하므로 LLM 출력 무시됩니다. 빈 배열 또는 placeholder로 두세요.
- conversion_strategy의 actions는 구체적인 실행 방안으로 작성하세요.

View File

@ -0,0 +1,24 @@
다음은 성형외과/피부과 유튜브 채널 데이터입니다.
채널명: {channel_name}
구독자 수: {subscribers}
총 영상 수: {total_videos}
총 조회수: {total_views}
평균 영상 길이: {avg_video_length}
업로드 주기: {upload_frequency}
인기 영상 목록: {top_videos}
플레이리스트: {playlists}
위 데이터를 바탕으로 이 채널의 마케팅 문제점과 개선사항을 진단해줘.
각 항목은 category(진단 카테고리), detail(상세 설명), severity(critical/warning/info) 형식의 JSON 배열로 출력해줘.
진단 카테고리들은 다음과 같아. :
구독자 대비 조회수 비율,
최근 롱폼 조회수,
Shorts 조회수,
업로드 빈도,
콘텐츠 톤앤매너,
썸네일 디자인,
최고 성과 Shorts
출처 번호([1], [2] 등)는 굳이 포함하지 마.

View File

@ -1,4 +1,5 @@
import re
import httpx
from http import HTTPMethod
from urllib.parse import urlparse
from common.utils import http_request
@ -64,6 +65,20 @@ class NaverClient:
return None
return resp.text
async def fetch_blog_total_count(self, handle: str) -> int | None:
"""블로그 전체 글 수는 RSS에 없어서 PostList HTML에서 '554개의 글' 패턴 추출.
<h4 class="category_title pcol2">... 554개의 </h4> 구조."""
resp = await http_request(
HTTPMethod.GET,
url=f"https://blog.naver.com/PostList.naver?blogId={handle}&from=postList&directAccess=true",
timeout=15,
label="naver-blog-postlist",
)
if not resp or not resp.is_success:
return None
m = re.search(r"(\d+)개의 글", resp.text)
return int(m.group(1)) if m else None
async def get_blog_rss(self, url: str) -> dict | None:
blog_handle = urlparse(url).path.strip("/").split("/")[0] if "://" in url else url
xml = await self.fetch_blog_rss(blog_handle)
@ -82,10 +97,71 @@ class NaverClient:
"postDate": date.group(1) if date else "",
"description": re.sub(r"<[^>]*>", "", desc.group(1) if desc else "").strip()[:150],
})
# RSS의 totalCount 우선, 없으면 블로그 PostList 페이지에서 "N개의 글" 파싱, 그것도 없으면 RSS 글수
total_match = re.search(r"<totalCount>(\d+)</totalCount>", xml)
if total_match:
total = int(total_match.group(1))
else:
total = await self.fetch_blog_total_count(blog_handle) or len(posts)
return {
"officialBlogUrl": f"https://blog.naver.com/{blog_handle}",
"officialBlogHandle": blog_handle,
"totalResults": int(total_match.group(1)) if total_match else len(posts),
"totalResults": total,
"posts": posts[:10],
}
async def get_cafe_info(self, cafe_url: str, *_args, **_kwargs) -> dict | None:
"""네이버 카페 운영 신호 수집. 2단계 fetch:
1) https://cafe.naver.com/{handle} cafeId 추출
2) ArticleList.nhn?search.clubid={cafeId} memberCount + cafeName 추출
본문/게시글은 로그인 필요라 가져옴. 회원수·카페명만 잡히면 충분.
common.http_request는 redirect 따라가서 카페 페이지에 맞아 httpx 직접 사용."""
handle = urlparse(cafe_url).path.strip("/").split("/")[0] if "://" in cafe_url else cafe_url.split("/")[-1]
if not handle:
return None
async with httpx.AsyncClient(
timeout=10, follow_redirects=True,
headers={"User-Agent": "Mozilla/5.0"},
) as c:
# 1. cafeId 추출
try:
main = await c.get(f"https://cafe.naver.com/{handle}")
except Exception:
return {"url": f"https://cafe.naver.com/{handle}", "cafeHandle": handle, "accessible": False}
if main.status_code != 200:
return {"url": f"https://cafe.naver.com/{handle}", "cafeHandle": handle, "accessible": False}
cid_match = re.search(r'cafeId["\']?\s*[:=]\s*["\']?(\d+)', main.text)
cafe_id = cid_match.group(1) if cid_match else None
result: dict = {
"url": f"https://cafe.naver.com/{handle}",
"cafeHandle": handle,
"cafeId": cafe_id,
"accessible": True,
"cafeName": None,
"memberCount": None,
}
if not cafe_id:
return result
# 2. ArticleList 페이지에서 회원수 + 카페명 추출 (로그인 없이 접근 가능한 유일한 endpoint)
try:
listing = await c.get(
f"https://cafe.naver.com/ArticleList.nhn?search.clubid={cafe_id}&search.menuid=&search.boardtype=L",
headers={"Referer": f"https://cafe.naver.com/{handle}"},
)
except Exception:
return result
if listing.status_code != 200:
return result
mc = re.search(r'memberCount[^0-9]+(\d[\d,]*)', listing.text)
if mc:
result["memberCount"] = int(mc.group(1).replace(",", ""))
tm = re.search(r"<title>(.+?)\s*:\s*네이버 카페</title>", listing.text)
if tm:
name = re.sub(r"&amp;", "&", tm.group(1)).strip()
if "," in name:
name = name.split(",", 1)[1].strip()
result["cafeName"] = name
return result

View File

@ -0,0 +1,66 @@
"""홈페이지 HTML + 외부 CSS 를 가져오는 fetch 전용 모듈.
오래된 한국 의료 사이트들이 SSL DH_KEY_TOO_SMALL / cipher 약함 / host mismatch 등으로
표준 fetch 차단되는 케이스가 많아 단계별 SSL fallback 으로 받는다.
파싱·도메인 로직은 들어가지 않음 순수 HTTP 응답 본문 반환.
"""
import logging
import re
import ssl
from urllib.parse import urljoin
import httpx
logger = logging.getLogger(__name__)
CSS_LINK = re.compile(
r'<link[^>]+rel=["\']stylesheet["\'][^>]+href=["\']([^"\']+)["\']',
re.IGNORECASE,
)
def _make_ssl_context() -> ssl.SSLContext:
"""보안 등급 1로 낮춤 + cert 검증 유지 (옛 한국 의료 사이트 cipher 약함 회피)."""
ctx = ssl.create_default_context()
try:
ctx.set_ciphers("DEFAULT@SECLEVEL=1")
except ssl.SSLError:
pass
return ctx
async def fetch_html(url: str, timeout: float = 20.0) -> tuple[int, str]:
"""SSL 검증 단계별 fallback 으로 HTML 본문 받기. 실패 시 (0, "")."""
headers = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"}
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers) as c:
r = await c.get(url)
return r.status_code, r.text
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError) as e:
logger.info("[fetch] %s standard SSL failed: %s — fallback to weak cipher", url, e)
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers, verify=_make_ssl_context()) as c:
r = await c.get(url)
return r.status_code, r.text
except (httpx.ConnectError, httpx.ReadError, ssl.SSLError) as e:
logger.info("[fetch] %s weak cipher failed: %s — fallback to verify=False", url, e)
try:
async with httpx.AsyncClient(timeout=timeout, follow_redirects=True, headers=headers, verify=False) as c:
r = await c.get(url)
return r.status_code, r.text
except Exception as e:
logger.warning("[fetch] %s all fallbacks failed: %s", url, e)
return 0, ""
async def fetch_html_and_css(homepage_url: str, max_css_files: int = 8) -> tuple[str, list[str]]:
"""홈페이지 HTML + 외부 CSS(Top N) 한 번에 fetch. 실패 시 ("", [])."""
status, html = await fetch_html(homepage_url)
if status != 200 or not html:
logger.warning("[fetch] homepage fetch failed status=%s url=%s", status, homepage_url)
return "", []
css_texts: list[str] = []
for css_href in CSS_LINK.findall(html)[:max_css_files]:
cstatus, ctext = await fetch_html(urljoin(homepage_url, css_href), timeout=15.0)
if cstatus == 200 and ctext:
css_texts.append(ctext)
return html, css_texts

View File

@ -1,173 +0,0 @@
"""Gemini Vision — 로고/브랜드 비주얼 자동 분석 (OpenAI 호환 모드).
정확한 hex 색상은 color_extractor가 CSS에서 직접 뽑음 (Vision은 근사값밖에 ).
Vision은 사람이 봐야 있는 정성 정보 심볼 형태/워드마크/ 담당.
"""
import base64
import json
import logging
import re
import httpx
from openai import AsyncOpenAI
logger = logging.getLogger(__name__)
DEFAULT_MODEL = "gemini-2.5-flash"
class VisionClient:
"""Gemini Vision을 OpenAI 호환 endpoint로 호출. GEMINI_API_KEY만 필요."""
def __init__(self, api_key: str, model: str = DEFAULT_MODEL, timeout: float = 30.0, max_retries: int = 2):
self.client = AsyncOpenAI(
api_key=api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
timeout=timeout,
max_retries=max_retries,
)
self.model = model
@staticmethod
def _extract_json(text: str) -> dict | None:
if not text:
return None
m = re.search(r"```(?:json)?\s*(\{.*?\})\s*```", text, re.DOTALL)
if m:
try:
return json.loads(m.group(1))
except json.JSONDecodeError:
pass
m = re.search(r"\{.*\}", text, re.DOTALL)
if m:
try:
return json.loads(m.group(0))
except json.JSONDecodeError:
return None
return None
@staticmethod
async def _fetch_as_data_url(url: str) -> str | None:
"""Gemini는 URL 직접 fetch가 막힌 호스트가 많아 base64 인라인으로 변환.
+ 'image does not exist' 같은 placeholder 이미지 거부 (작은 bytes / 잘못된 content-type)."""
try:
async with httpx.AsyncClient(timeout=15.0, follow_redirects=True) as c:
resp = await c.get(url)
if resp.status_code != 200:
logger.warning("[vision] fetch %s status=%s", url, resp.status_code)
return None
mime = resp.headers.get("content-type", "").split(";")[0].strip()
# 실제 이미지가 아니면 거부 (HTML 페이지가 404 대신 200으로 리다이렉트 되는 경우)
if not mime.startswith("image/"):
logger.warning("[vision] %s not an image (content-type=%s)", url, mime)
return None
size = len(resp.content)
if size < 500:
logger.warning("[vision] %s too small (%d bytes) — likely placeholder", url, size)
return None
b64 = base64.b64encode(resp.content).decode("ascii")
return f"data:{mime};base64,{b64}"
except Exception as e:
logger.warning("[vision] fetch error %s: %s", url, e)
return None
async def _ask(self, image_urls: list[str], prompt: str, max_tokens: int = 4000) -> dict | None:
content: list[dict] = []
for u in image_urls:
if not u:
continue
data_url = await self._fetch_as_data_url(u)
if not data_url:
continue
content.append({"type": "image_url", "image_url": {"url": data_url}})
if not any(c.get("type") == "image_url" for c in content):
logger.warning("[vision] no images could be fetched")
return None
content.append({"type": "text", "text": prompt})
try:
resp = await self.client.chat.completions.create(
model=self.model,
messages=[{"role": "user", "content": content}],
max_tokens=max_tokens,
)
choice = resp.choices[0]
if choice.finish_reason != "stop":
logger.warning("[vision] unexpected finish_reason=%s", choice.finish_reason)
return self._extract_json(choice.message.content or "")
except Exception as e:
logger.warning("[vision] error: %s", e)
return None
async def analyze_brand_assets(
self,
logo_url: str | None,
homepage_url: str | None,
additional_images: list[str] | None = None,
) -> dict:
"""로고 이미지를 보고 정성 분석. 정확한 hex는 color_extractor가 따로 처리하므로 여기선 안 뽑음."""
urls = [u for u in [logo_url] + list(additional_images or []) if u]
if not urls:
return {}
prompt = (
"당신은 브랜드 로고 시각 분석가입니다. 첨부된 이미지(첫 번째가 병원의 대표 로고)를 보고 "
"아래 JSON 스키마로만 응답하세요. 코드펜스 없이 순수 JSON만 출력.\n"
"{\n"
' "logo_description": "로고를 1~2문장으로 설명 (심볼 형태 + 워드마크 + 전반적 톤). 예: \'둥근 잎사귀를 감싼 추상 심볼에 세리프 한글 워드마크, 차분하고 고급스러운 톤\'",\n'
' "logo_style": "minimal | illustrative | typographic | abstract 중 하나",\n'
' "has_symbol": "심볼/아이콘이 있으면 true, 글자만 있으면 false (boolean)",\n'
' "logo_symbol": "심볼이 묘사하는 대상 (예: \'잎사귀\', \'추상 곡선\'). 없으면 빈 문자열",\n'
' "logo_text": "로고에 보이는 워드마크 텍스트 그대로 (한글/영문). 없으면 빈 문자열",\n'
' "logo_colors_desc": "로고에 쓰인 색감을 사람이 부르는 이름으로 서술 (예: \'딥네이비 + 골드\'). 정확한 hex는 출력하지 말 것"\n'
"}\n"
"주의: 색상 hex 값이나 logo URL 같은 필드는 출력하지 마세요 (별도 추출 로직이 처리).\n"
"모든 설명/텍스트 값은 반드시 한국어로 작성하세요 (영어 금지)."
)
result = await self._ask(urls, prompt)
if not result:
return {}
# logo_images는 우리가 직접 채움 (Vision은 묘사만)
result["logo_images"] = {"circle": None, "horizontal": logo_url, "korean": None}
return result
async def describe_channel_logos(
self,
official_logo_url: str | None,
channel_logos: list[dict],
) -> dict | None:
"""채널별 프로필 이미지(로고)를 보고 각각 설명 + 공식 로고와 일치 여부 평가.
channel_logos: [{"channel": "Instagram", "url": "..."}, ...]
반환: {"channel_logos": [{"channel","logo_description","is_official"}], "inconsistency_summary", "recommendation"}"""
items = [c for c in channel_logos if c.get("url")]
if not items:
return None
# 공식 로고가 있으면 맨 앞에 두고 기준으로 삼음
urls: list[str] = []
if official_logo_url:
urls.append(official_logo_url)
urls.extend(c["url"] for c in items)
channel_order = ", ".join(c.get("channel", "?") for c in items)
if official_logo_url:
header = (
"첨부 이미지 중 **첫 번째가 이 병원의 공식 로고**입니다. "
f"이어지는 이미지들은 채널별 프로필 이미지이며 순서는: {channel_order}.\n"
"각 채널 로고를 1문장으로 설명하고, 공식 로고(첫 번째)와 일치하면 is_official=true, "
"비공식 변형/모델사진/다른 이미지면 false로 평가하세요.\n"
)
else:
header = (
f"첨부 이미지는 한 병원의 채널별 프로필 이미지입니다. 순서: {channel_order}.\n"
"각 채널 로고를 1문장으로 설명하세요 (공식 로고 기준이 없으므로 is_official은 판단 가능하면만).\n"
)
prompt = (
header
+ "아래 JSON으로만 응답 (코드펜스 없이 순수 JSON):\n"
"{\n"
' "channel_logos": [{"channel": "...", "logo_description": "...", "is_official": true}],\n'
' "inconsistency_summary": "채널 간 로고 일관성 1~2문장 요약",\n'
' "recommendation": "통합 권고 1문장"\n'
"}\n"
"모든 logo_description·inconsistency_summary·recommendation은 반드시 한국어로 작성하세요 (영어 금지)."
)
return await self._ask(urls, prompt)

View File

@ -79,7 +79,17 @@ class YouTubeClient:
if resp and resp.is_success:
videos = resp.json().get("items", [])[:10]
return {"channelId": channel_id, "channel": channel, "videos": videos}
playlists: list[dict] = []
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/playlists",
params={"part": "snippet", "channelId": channel_id, "maxResults": 50, "key": self.api_key},
label="yt-playlists",
)
if resp and resp.is_success:
playlists = resp.json().get("items", [])
return {"channelId": channel_id, "channel": channel, "videos": videos, "playlists": playlists}
async def get_channel(self, url: str) -> dict | None:
raw = await self.fetch_channel(url)
@ -111,6 +121,11 @@ class YouTubeClient:
}
for v in raw["videos"]
],
"playlists": [
p.get("snippet", {}).get("title")
for p in raw["playlists"]
if p.get("snippet", {}).get("title")
],
}
async def search_channels(self, query: str, max_results: int = 3) -> list[str]:

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -11,6 +11,8 @@ class Channels(BaseModel):
tiktok: str | None = None
instagram_en: str | None = None
facebook_en: str | None = None
kakao_talk: str | None = None
naver_cafe: str | None = None
class AnalysisOptions(BaseModel):

View File

@ -10,9 +10,7 @@ class ClinicResponse(BaseModel):
hospital_name: str
hospital_name_en: str | None
road_address: str | None
url: str | None
status: str
raw_data: dict | None
created_at: str
updated_at: str

View File

@ -49,7 +49,7 @@ class ChannelBrandingRule(CamelModel):
profile_photo: str
banner_spec: str
bio_template: str
current_status: Literal["correct", "incorrect", "missing"]
current_status: Literal["correct", "incorrect", "N/A"]
class BrandGuide(CamelModel):

View File

@ -66,22 +66,18 @@ class RegistryData(CamelModel):
class ClinicSnapshot(CamelModel):
name: str
name_en: str
established: str
years_in_business: int
staff_count: int
lead_doctor: LeadDoctor
overall_rating: float
total_reviews: int
price_range: PriceRange
certifications: list[str]
media_appearances: list[str]
medical_tourism: list[str]
location: str
nearest_station: str
phone: str
domain: str
# _build_clinic_snapshot은 source 데이터 있을 때만 필드 추가 (`if x:` 가드).
# 강남언니/홈페이지 수집 누락된 병원에서 required면 ValidationError로 리포트 전체 실패.
name: str | None = None
name_en: str | None = None
staff_count: int | None = None
lead_doctor: LeadDoctor | None = None
overall_rating: float | None = None
total_reviews: int | None = None
certifications: list[str] = []
location: str | None = None
phone: str | None = None
domain: str | None = None
logo_images: LogoImages | None = None
brand_colors: BrandColors | None = None
source: DataSource | None = None
@ -131,7 +127,6 @@ class YouTubeAudit(CamelModel):
avg_video_length: str
upload_frequency: str
channel_created_date: str
subscriber_rank: str
channel_description: str
linked_urls: list[LinkedUrl]
playlists: list[str]
@ -156,8 +151,8 @@ class InstagramAccount(CamelModel):
class InstagramAudit(CamelModel):
accounts: list[InstagramAccount]
diagnosis: list[DiagnosisItem]
accounts: list[InstagramAccount] = []
diagnosis: list[DiagnosisItem] = []
class BrandInconsistencyValue(CamelModel):
@ -188,17 +183,17 @@ class FacebookPage(CamelModel):
linked_domain: str
reviews: int
recent_post_age: str
has_whatsapp: bool
post_frequency: str | None = None
has_whatsapp: bool | None = None
post_frequency: str
top_content_type: str | None = None
engagement: str | None = None
engagement: str
class FacebookAudit(CamelModel):
pages: list[FacebookPage]
diagnosis: list[DiagnosisItem]
brand_inconsistencies: list[BrandInconsistency]
consolidation_recommendation: str
pages: list[FacebookPage] = []
diagnosis: list[DiagnosisItem] = []
brand_inconsistencies: list[BrandInconsistency] = []
consolidation_recommendation: str | None = None
class OtherChannel(CamelModel):

View File

@ -36,9 +36,24 @@ class DataSource(StrEnum):
SCRAPE = "scrape"
class SourceType(StrEnum):
MAINPAGE = "mainpage"
INSTAGRAM = "instagram"
FACEBOOK = "facebook"
NAVER_BLOG = "naver_blog"
YOUTUBE = "youtube"
TIKTOK = "tiktok"
GANGNAM_UNNI = "gangnam_unni"
KAKAOTALK = "kakaotalk"
NAVER_CAFE = "naver_cafe"
# 부가 수집/분석 (HTML/CSS 재크롤 + Vision 로고 매칭) — 한 raw_info entry 에 brandAssets/channelLogos 같이 보관.
BRANDING = "branding"
class Language(StrEnum):
KR = "KR"
EN = "EN"
WW = "WW"
class VideoType(StrEnum):

View File

@ -1,29 +1,28 @@
import json
import logging
from common.db import fetchone, execute, fetch_raw, get_analysis_raw_data, save_analysis_report, get_market_analysis
import re
from datetime import datetime
from urllib.parse import urlparse
from common.db.run import update_run_report, update_run_plan, select_run_report_data
from common.db.source import select_run_raw_data, select_mainpage_logo_url
from common.db.market import select_market
from integrations.llm.llm_service import LLMService
from integrations.llm.prompt import report_prompt, plan_prompt
from integrations.llm.schemas.report import ReportOutput
from services.instagram_audit import build_instagram_accounts
from integrations.llm.prompt import report_prompt, plan_prompt, youtube_diagnosis_prompt
from integrations.llm.schemas.report import ReportOutput, ClinicSnapshot, YouTubeAudit
from services.branding import analyze_branding
from services.instagram_audit import build_instagram_audit
from services.facebook_audit import build_facebook_audit
from services.kpi_dashboard import build_kpi_dashboard
from integrations.llm.schemas.plan import PlanOutput
from models.status import AnalysisStatus
logger = logging.getLogger(__name__)
async def generate_report(analysis_run_id: str) -> ReportOutput:
run = await fetchone(
"SELECT hospital_id FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
clinic_row = await fetchone(
"SELECT raw_data FROM hospital_baseinfo WHERE hospital_id = %s",
(run["hospital_id"],),
)
raw_data = clinic_row["raw_data"] if clinic_row else None
clinic = json.loads(raw_data) if isinstance(raw_data, str) else (raw_data or {})
raw = await get_analysis_raw_data(analysis_run_id)
market = await get_market_analysis(analysis_run_id)
raw = await select_run_raw_data(analysis_run_id)
clinic = raw.get("mainpage") or {}
branding = raw.get("branding") or {}
market = await select_market(analysis_run_id)
def _json(v) -> str | None:
return json.dumps(v, ensure_ascii=False) if v else None
@ -40,34 +39,36 @@ async def generate_report(analysis_run_id: str) -> ReportOutput:
"market_keywords": _json(market.get("keywords")),
"market_trend": _json(market.get("trend")),
"market_target_audience": _json(market.get("target_audience")),
# firecrawl 이 mainpage 에서 뽑은 branding 메타(logoUrl/ogImage/faviconUrl) + Vision/CSS 산출물
"branding": _json(clinic.get("branding")),
"brand_assets": _json(clinic.get("brandAssets")),
"tiktok": _json(clinic.get("tiktok")),
"instagram_en": _json(clinic.get("instagramEn")),
"facebook_en": _json(clinic.get("facebookEn")),
"channel_logos": _json(clinic.get("channelLogos")),
"brand_assets": _json(branding.get("brandAssets")),
"channel_logos": _json(branding.get("channelLogos")),
# 부가 채널 (raw_info entry) — raw dict 의 한국식 key 그대로
"tiktok": _json(raw.get("tiktok")),
"instagram_en": _json(raw.get("instagram_en")),
"facebook_en": _json(raw.get("facebook_en")),
"kakao_talk": _json(raw.get("kakaotalk")),
"naver_cafe": _json(raw.get("naver_cafe")),
# 메인 5채널은 raw dict 그대로 펼쳐서 prompt placeholder 와 매칭
**{
channel: _json(data)
for channel, data in raw.items()
source_type: _json(data)
for source_type, data in raw.items()
if source_type not in {
"mainpage", "branding",
"tiktok", "instagram_en", "facebook_en", "kakaotalk", "naver_cafe",
}
},
}
return await LLMService(provider="perplexity").generate(report_prompt, input_data)
async def generate_plan(analysis_run_id: str) -> PlanOutput:
run = await fetchone(
"SELECT hospital_id, report_data FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
clinic_row = await fetchone(
"SELECT raw_data FROM hospital_baseinfo WHERE hospital_id = %s",
(run["hospital_id"],),
)
raw_data = clinic_row["raw_data"] if clinic_row else None
clinic = json.loads(raw_data) if isinstance(raw_data, str) else (raw_data or {})
report_data = run["report_data"]
report = json.loads(report_data) if isinstance(report_data, str) else report_data
market = await get_market_analysis(analysis_run_id)
raw = await select_run_raw_data(analysis_run_id)
clinic = raw.get("mainpage") or {}
branding = raw.get("branding") or {}
report = await select_run_report_data(analysis_run_id)
market = await select_market(analysis_run_id)
def _json(v) -> str | None:
return json.dumps(v, ensure_ascii=False) if v else None
@ -85,47 +86,32 @@ async def generate_plan(analysis_run_id: str) -> PlanOutput:
"market_keywords": _json(market.get("keywords")),
"market_trend": _json(market.get("trend")),
"market_target_audience": _json(market.get("target_audience")),
"tiktok": _json(clinic.get("tiktok")),
"instagram_en": _json(clinic.get("instagramEn")),
"facebook_en": _json(clinic.get("facebookEn")),
"channel_logos": _json(clinic.get("channelLogos")),
"brand_assets": _json(clinic.get("brandAssets")),
"tiktok": _json(raw.get("tiktok")),
"instagram_en": _json(raw.get("instagram_en")),
"facebook_en": _json(raw.get("facebook_en")),
"naver_blog": _json(_naver_blog_summary(raw.get("naver_blog"))),
"naver_cafe": _json(raw.get("naver_cafe")),
"kakao_talk": _json(raw.get("kakaotalk")),
"channel_logos": _json(branding.get("channelLogos")),
"brand_assets": _json(branding.get("brandAssets")),
}
return await LLMService(provider="perplexity").generate(plan_prompt, input_data)
async def _build_overrides(analysis_run_id: str) -> dict:
run = await fetchone(
"SELECT hospital_id, instagram_data_id, facebook_data_id,"
" naver_blog_data_id, youtube_data_id, gangnam_unni_data_id"
" FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
if not run:
return {}
hospital_row = await fetchone(
"SELECT raw_data FROM hospital_baseinfo WHERE hospital_id = %s",
(run["hospital_id"],),
)
hospital = json.loads(hospital_row["raw_data"]) if hospital_row and isinstance(hospital_row.get("raw_data"), str) else (hospital_row or {}).get("raw_data") or {}
instagram = await fetch_raw("instagram_data", run["instagram_data_id"]) or {}
facebook = await fetch_raw("facebook_data", run["facebook_data_id"]) or {}
naver_blog = await fetch_raw("naver_blog_data", run["naver_blog_data_id"]) or {}
youtube = await fetch_raw("youtube_data", run["youtube_data_id"]) or {}
gangnam_unni = await fetch_raw("gangnam_unni_data", run["gangnam_unni_data_id"]) or {}
def _build_clinic_snapshot(gangnam_unni: dict, mainpage: dict, brand_assets: dict, logo_url: str | None) -> dict:
snapshot: dict = {}
# ── gangnam_unni ──────────────────────────────────────────────────────────
doctors = gangnam_unni.get("doctors", [])
lead = max(doctors, key=lambda d: d.get("reviews", 0)) if doctors else None
if gangnam_unni.get("name"): snapshot["name"] = gangnam_unni["name"]
if gangnam_unni.get("rating"): snapshot["overall_rating"] = gangnam_unni["rating"]
if gangnam_unni.get("totalReviews"): snapshot["total_reviews"] = gangnam_unni["totalReviews"]
if gangnam_unni.get("address"): snapshot["location"] = gangnam_unni["address"]
if gangnam_unni.get("badges"): snapshot["certifications"] = gangnam_unni["badges"]
if gangnam_unni.get("name"): snapshot["name"] = gangnam_unni["name"]
if mainpage.get("clinicNameEn"): snapshot["name_en"] = mainpage["clinicNameEn"]
if mainpage.get("phone"): snapshot["phone"] = mainpage["phone"]
domain = mainpage.get("domain") or urlparse(mainpage.get("sourceUrl") or "").netloc
if domain: snapshot["domain"] = domain
if gangnam_unni.get("rating"): snapshot["overall_rating"] = gangnam_unni["rating"]
if gangnam_unni.get("totalReviews"): snapshot["total_reviews"] = gangnam_unni["totalReviews"]
if gangnam_unni.get("address"): snapshot["location"] = gangnam_unni["address"]
if gangnam_unni.get("badges"): snapshot["certifications"] = gangnam_unni["badges"]
if gangnam_unni.get("totalMajorStaffs"): snapshot["staff_count"] = gangnam_unni["totalMajorStaffs"]
if lead:
snapshot["lead_doctor"] = {
@ -134,87 +120,216 @@ async def _build_overrides(analysis_run_id: str) -> dict:
"rating": lead.get("rating"),
"review_count": lead.get("reviews"),
}
# logo URL 은 raw_info.logo_url 컬럼에서, brand_colors 는 JSON 에서 강제 주입. LLM 의 null 처리 차단.
if logo_url:
snapshot["logo_images"] = {"circle": None, "horizontal": logo_url, "korean": None}
if brand_assets.get("brand_colors"): snapshot["brand_colors"] = brand_assets["brand_colors"]
return ClinicSnapshot.model_validate(snapshot).model_dump()
# ── instagram (KR·EN 계정을 코드에서 구성 → LLM 출력 무시하고 교체) ──────────────
ig_patch = build_instagram_accounts(
instagram, hospital.get("instagramEn") or {}, hospital.get("channelLogos") or {},
def _naver_blog_summary(blog: dict | None) -> dict | None:
"""plan 카드 한 장에 들어가는 건 전체 포스트 수와 최근 활동 시점뿐. 그 외(본문·링크·제목)는
던져봐야 토큰만 늘고 LLM 무관 정보로 hallucinate ."""
if not blog:
return None
posts = blog.get("posts") or []
return {
"totalPosts": blog.get("totalResults"),
"latestPostDate": posts[0].get("postDate") if posts else None,
}
def _parse_iso_duration_seconds(iso: str) -> int:
m = re.match(r"PT(?:(\d+)H)?(?:(\d+)M)?(?:(\d+)S)?", iso or "")
if not m:
return 0
h, mins, s = (int(x or 0) for x in m.groups())
return h * 3600 + mins * 60 + s
def _format_seconds(seconds: int) -> str:
m, s = divmod(seconds, 60)
h, m = divmod(m, 60)
return f"{h}시간 {m}" if h else f"{m}{s}"
def _format_clock(seconds: int) -> str:
m, s = divmod(seconds, 60)
h, m = divmod(m, 60)
return f"{h}:{m:02d}:{s:02d}" if h else f"{m}:{s:02d}"
def _calc_avg_video_length(videos: list[dict]) -> str:
durations = [_parse_iso_duration_seconds(v.get("duration", "")) for v in videos]
durations = [d for d in durations if d > 0]
if not durations:
return ""
return _format_seconds(sum(durations) // len(durations))
def _relative_date(date_str: str) -> str:
if not date_str:
return ""
try:
past = datetime.fromisoformat(date_str[:10])
except ValueError:
return ""
days = (datetime.now() - past).days
if days < 1:
return "오늘"
if days < 30:
return f"{days}일 전"
if days < 365:
return f"{days // 30}개월 전"
return f"{days // 365}년 전"
def _calc_upload_frequency(videos: list[dict]) -> str:
dates = sorted(
[v["date"][:10] for v in videos if v.get("date")],
reverse=True,
)
if len(dates) < 2:
return ""
gaps = [
(datetime.fromisoformat(dates[i]) - datetime.fromisoformat(dates[i + 1])).days
for i in range(len(dates) - 1)
]
avg_days = sum(gaps) // len(gaps)
if avg_days <= 7:
return f"{7 // max(avg_days, 1)}"
if avg_days <= 30:
return f"{30 // avg_days}"
return f"{avg_days}일에 1회"
# ── facebook ──────────────────────────────────────────────────────────────
fb_patch: dict = {}
if facebook.get("pageUrl"): fb_patch["url"] = facebook["pageUrl"]
if facebook.get("pageUrl"): fb_patch["link"] = facebook["pageUrl"]
if facebook.get("pageName"): fb_patch["page_name"] = facebook["pageName"]
if facebook.get("followers"): fb_patch["followers"] = facebook["followers"]
if facebook.get("intro"): fb_patch["bio"] = facebook["intro"]
if facebook.get("categories"): fb_patch["category"] = ", ".join(facebook["categories"])
if facebook.get("website"): fb_patch["linked_domain"] = facebook["website"]
# ── youtube ───────────────────────────────────────────────────────────────
yt_patch: dict = {}
if youtube.get("channelName"): yt_patch["channel_name"] = youtube["channelName"]
if youtube.get("handle"): yt_patch["handle"] = youtube["handle"]
if youtube.get("subscribers"): yt_patch["subscribers"] = youtube["subscribers"]
if youtube.get("totalVideos"): yt_patch["total_videos"] = youtube["totalVideos"]
if youtube.get("totalViews"): yt_patch["total_views"] = youtube["totalViews"]
async def _build_youtube_audit(youtube: dict) -> dict:
videos = youtube.get("videos", [])
yt_patch: dict = {
"weekly_view_growth": {"absolute": 0, "percentage": 0.0},
"estimated_monthly_revenue": {"min": 0, "max": 0},
"linked_urls": [],
"avg_video_length": _calc_avg_video_length(videos),
"upload_frequency": _calc_upload_frequency(videos),
}
if youtube.get("channelName"): yt_patch["channel_name"] = youtube["channelName"]
if youtube.get("handle"): yt_patch["handle"] = youtube["handle"]
if youtube.get("subscribers"): yt_patch["subscribers"] = youtube["subscribers"]
if youtube.get("totalVideos"): yt_patch["total_videos"] = youtube["totalVideos"]
if youtube.get("totalViews"): yt_patch["total_views"] = youtube["totalViews"]
if youtube.get("publishedAt"): yt_patch["channel_created_date"] = youtube["publishedAt"][:10]
if youtube.get("description"): yt_patch["channel_description"] = youtube["description"]
if youtube.get("videos"):
if youtube.get("description"): yt_patch["channel_description"] = youtube["description"]
if youtube.get("playlists"): yt_patch["playlists"] = youtube["playlists"]
if videos:
yt_patch["top_videos"] = [
{
"title": v["title"],
"views": v["views"],
"duration": v.get("duration"),
"duration": _format_clock(_parse_iso_duration_seconds(v.get("duration", ""))),
"type": "Short" if "M" not in v.get("duration", "") else "Long",
"uploaded_ago": v.get("date", "")[:10],
"uploaded_ago": _relative_date(v.get("date", "")),
}
for v in youtube["videos"]
for v in videos
]
overrides: dict = {}
if snapshot:
overrides["clinic_snapshot"] = snapshot
if ig_patch:
overrides["instagram_audit"] = {"accounts": ig_patch}
if fb_patch:
overrides["facebook_audit"] = {"pages": [fb_patch]}
if yt_patch:
overrides["youtube_audit"] = yt_patch
return overrides
diagnosis_result = await LLMService(provider="perplexity").generate(
youtube_diagnosis_prompt,
{
"channel_name": yt_patch.get("channel_name"),
"subscribers": yt_patch.get("subscribers"),
"total_videos": yt_patch.get("total_videos"),
"total_views": yt_patch.get("total_views"),
"avg_video_length": yt_patch.get("avg_video_length"),
"upload_frequency": yt_patch.get("upload_frequency"),
"top_videos": json.dumps(yt_patch.get("top_videos", []), ensure_ascii=False),
"playlists": json.dumps(yt_patch.get("playlists", []), ensure_ascii=False),
},
)
yt_patch["diagnosis"] = [item.model_dump() for item in diagnosis_result.diagnosis]
return YouTubeAudit.model_validate(yt_patch).model_dump()
def _deep_merge(base: dict, overrides: dict) -> dict:
"""dict 끼리 만나면 재귀로 안쪽까지 합치고, 그 외(list/scalar/None) 는 override 값으로 통째 치환."""
for k, v in overrides.items():
if isinstance(v, dict) and isinstance(base.get(k), dict):
_deep_merge(base[k], v)
elif isinstance(v, list) and isinstance(base.get(k), list):
for i, item in enumerate(v):
if i < len(base[k]) and isinstance(item, dict) and isinstance(base[k][i], dict):
_deep_merge(base[k][i], item)
else:
base[k] = v
return base
def _patch_report(result: ReportOutput, overrides: dict) -> ReportOutput:
async def _build_overrides(analysis_run_id: str, result: ReportOutput) -> ReportOutput:
raw = await select_run_raw_data(analysis_run_id)
if not raw:
return result
mainpage = raw.get("mainpage", {}) or {}
branding = raw.get("branding", {}) or {}
instagram = raw.get("instagram", {}) or {}
facebook = raw.get("facebook", {}) or {}
youtube = raw.get("youtube", {}) or {}
gangnam_unni = raw.get("gangnam_unni", {}) or {}
naver_blog = raw.get("naver_blog", {}) or {}
instagram_en = raw.get("instagram_en", {}) or {}
facebook_en = raw.get("facebook_en", {}) or {}
tiktok = raw.get("tiktok", {}) or {}
naver_cafe = raw.get("naver_cafe", {}) or {}
brand_assets = branding.get("brandAssets") or {}
channel_logos = branding.get("channelLogos") or {}
logo_url = await select_mainpage_logo_url(analysis_run_id)
llm_fb_pages = result.model_dump().get("facebook_audit", {}).get("pages", [])
snapshot: dict = _build_clinic_snapshot(gangnam_unni, mainpage, brand_assets, logo_url)
yt_patch: dict = await _build_youtube_audit(youtube)
ig_patch = build_instagram_audit(instagram, instagram_en, channel_logos)
fb_patch = build_facebook_audit(facebook, facebook_en, llm_fb_pages)
kpi_extras = {
"instagramEn": instagram_en,
"facebookEn": facebook_en,
"tiktok": tiktok,
"naverCafe": naver_cafe,
}
kpi = build_kpi_dashboard(instagram, facebook, youtube, gangnam_unni, kpi_extras, naver_blog)
overrides: dict = {}
if snapshot: overrides["clinic_snapshot"] = snapshot
if ig_patch: overrides["instagram_audit"] = ig_patch
if fb_patch: overrides["facebook_audit"] = fb_patch
if yt_patch: overrides["youtube_audit"] = yt_patch
if kpi: overrides["kpi_dashboard"] = kpi
merged = _deep_merge(result.model_dump(), overrides)
# 인스타 계정은 프롬프트에서 LLM이 []로 두게 했고, 코드가 수집 데이터로 채운다 (데이터 없으면 빈 리스트)
merged.setdefault("instagram_audit", {})["accounts"] = (overrides.get("instagram_audit") or {}).get("accounts") or []
return ReportOutput(**merged)
async def run_report_task(analysis_run_id: str) -> None:
logger.info("[report] start run=%s", analysis_run_id)
await analyze_branding(analysis_run_id)
result = await generate_report(analysis_run_id)
result = _patch_report(result, await _build_overrides(analysis_run_id))
await save_analysis_report(analysis_run_id, result.model_dump())
result = await _build_overrides(analysis_run_id, result)
await update_run_report(analysis_run_id, result.model_dump())
logger.info("[report] done run=%s", analysis_run_id)
def _patch_plan(result: PlanOutput, logo_desc: str) -> PlanOutput:
"""brand_guide.channel_branding[].profile_photo 는 LLM 안 맡기고 코드가 박는다
(모든 채널 동일값 = brand_assets.logo_description). LLM fallback 문구 hallucinate 방지."""
p = result.model_dump()
for ch in (p.get("brand_guide") or {}).get("channel_branding") or []:
ch["profile_photo"] = logo_desc
return PlanOutput(**p)
async def run_plan_task(analysis_run_id: str) -> None:
logger.info("[plan] start run=%s", analysis_run_id)
result = await generate_plan(analysis_run_id)
await execute(
"UPDATE analysis_runs SET plan_data = %s WHERE analysis_run_id = %s",
(json.dumps(result.model_dump(), ensure_ascii=False), analysis_run_id),
)
# profile_photo 는 brand_assets.logo_description 으로 코드가 박음 (LLM "(가이드 미보유)" 같은 hallucination 차단).
raw = await select_run_raw_data(analysis_run_id)
branding = raw.get("branding") or {}
logo_desc = ((branding.get("brandAssets") or {}).get("logo_description")) or ""
result = _patch_plan(result, logo_desc)
await update_run_plan(analysis_run_id, result.model_dump())
logger.info("[plan] done run=%s", analysis_run_id)

View File

@ -0,0 +1,172 @@
"""collect 단계 - HTML/CSS 텍스트에서 brand 로고 URL + 색상 추출"""
import logging
import re
from collections import Counter
from urllib.parse import urljoin
logger = logging.getLogger(__name__)
# ── 로고 URL 추출 ─────────────────────────────────────────────────────────────
LOGO_IMG_PATTERNS = [
re.compile(r'<img[^>]*\bclass=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
re.compile(r'<img[^>]*\bsrc=["\']([^"\']+)["\'][^>]*\bclass=["\'][^"\']*\blogo\b[^"\']*["\']', re.IGNORECASE),
re.compile(r'<img[^>]*\bid=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
re.compile(r'<img[^>]*\balt=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE),
re.compile(r'<(?:a|h[1-6]|div|span)[^>]*\b(?:class|id)=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+)["\']', re.IGNORECASE | re.DOTALL),
re.compile(r'<(?:a|div|span|h[1-6])[^>]*\b(?:class|id)=["\'][^"\']*\blogo\b[^"\']*["\'][^>]*\bstyle=["\'][^"\']*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)', re.IGNORECASE),
re.compile(r'<(?:a|div|span|h[1-6])[^>]*\bstyle=["\'][^"\']*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)[^"\']*["\'][^>]*\b(?:class|id)=["\'][^"\']*\blogo\b', re.IGNORECASE),
re.compile(r'<img[^>]*\bsrc=["\']([^"\']*\blogo\b[^"\']*\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE),
re.compile(r'<header\b[^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE | re.DOTALL),
re.compile(r'<nav\b[^>]*>(?:[^<]|<(?!img))*<img[^>]*\bsrc=["\']([^"\']+\.(?:png|svg|jpe?g|webp)[^"\']*)["\']', re.IGNORECASE | re.DOTALL),
re.compile(r'<meta[^>]*\bproperty=["\']og:image["\'][^>]*\bcontent=["\']([^"\']+)["\']', re.IGNORECASE),
re.compile(r'<meta[^>]*\bcontent=["\']([^"\']+)["\'][^>]*\bproperty=["\']og:image["\']', re.IGNORECASE),
]
LOGO_CSS_PATTERN = re.compile(
r'\.[\w-]*\blogo\b[\w-]*\s*(?:,\s*\.[\w-]+\s*)*\{[^}]*background(?:-image)?\s*:\s*url\(\s*["\']?([^"\')\s]+)',
re.IGNORECASE | re.DOTALL,
)
def find_logo_url_in_html(html: str, base_url: str, css_texts: list[str] | None = None) -> str | None:
"""HTML 에서 logo URL 찾기. 우선순위: 1) class/id/alt 명시 img 2) 외부 CSS .logo bg 3) header/nav 첫 img."""
def _is_noise(src: str) -> bool:
if not src or src.startswith("data:"):
return True
if re.search(r"(blank|spacer|pixel|transparent|1x1)\b", src, re.IGNORECASE):
return True
if re.search(r"(lang[-_]?(kor|eng|chn|jpn|rus|jp|en|ko|cn|ar|in)|flag|country|icon-|btn-|arrow|prev|next|search)\b", src, re.IGNORECASE):
return True
return False
for pat in LOGO_IMG_PATTERNS[:8]:
for m in pat.finditer(html):
src = m.group(1)
if _is_noise(src):
continue
return urljoin(base_url, src)
for css in (css_texts or []):
m = LOGO_CSS_PATTERN.search(css)
if m:
src = m.group(1)
if not _is_noise(src):
return urljoin(base_url, src)
for pat in LOGO_IMG_PATTERNS[8:]:
for m in pat.finditer(html):
src = m.group(1)
if _is_noise(src):
continue
return urljoin(base_url, src)
return None
# ── 색상 추출 ────────────────────────────────────────────────────────────────
HEX6 = re.compile(r"#([0-9a-fA-F]{6})\b")
HEX3 = re.compile(r"#([0-9a-fA-F]{3})\b(?![0-9a-fA-F])")
RGB = re.compile(r"rgba?\(\s*(\d{1,3})\s*,\s*(\d{1,3})\s*,\s*(\d{1,3})\s*(?:,\s*[\d.]+\s*)?\)")
STYLE_BLOCK = re.compile(r"<style[^>]*>(.*?)</style>", re.IGNORECASE | re.DOTALL)
NOISE = {
"#ffffff", "#000000", "#fff", "#000",
"#333", "#222", "#111", "#444", "#555", "#666", "#777", "#888", "#999",
"#aaa", "#bbb", "#ccc", "#ddd", "#eee", "#f0f0f0", "#f5f5f5", "#fafafa",
}
def _normalize(hex_str: str) -> str:
h = hex_str.lstrip("#").lower()
if len(h) == 3:
h = "".join(c * 2 for c in h)
if len(h) == 8:
h = h[:6]
return f"#{h}"
def _rgb_to_hex(r: int, g: int, b: int) -> str:
return f"#{r:02x}{g:02x}{b:02x}"
def _hex_to_rgb(h: str) -> tuple[int, int, int]:
h = h.lstrip("#")
return int(h[0:2], 16), int(h[2:4], 16), int(h[4:6], 16)
def _distance(a: str, b: str) -> float:
ar, ag, ab = _hex_to_rgb(a)
br, bg, bb = _hex_to_rgb(b)
return ((ar - br) ** 2 + (ag - bg) ** 2 + (ab - bb) ** 2) ** 0.5
def _is_grayscale(h: str, tol: int = 12) -> bool:
r, g, b = _hex_to_rgb(h)
return max(r, g, b) - min(r, g, b) < tol
def _extract_hex(text: str) -> list[str]:
out: list[str] = []
out.extend(_normalize(m.group(0)) for m in HEX6.finditer(text))
out.extend(_normalize(m.group(0)) for m in HEX3.finditer(text))
for m in RGB.finditer(text):
r, g, b = int(m.group(1)), int(m.group(2)), int(m.group(3))
if 0 <= r <= 255 and 0 <= g <= 255 and 0 <= b <= 255:
out.append(_rgb_to_hex(r, g, b))
return out
def _cluster(colors: Counter, threshold: float = 25.0) -> list[tuple[str, int]]:
ranked = colors.most_common()
clusters: list[tuple[str, int]] = []
for color, count in ranked:
merged = False
for i, (rep, rep_count) in enumerate(clusters):
if _distance(color, rep) < threshold:
clusters[i] = (rep, rep_count + count)
merged = True
break
if not merged:
clusters.append((color, count))
return clusters
def extract_brand_colors_from_text(html: str, css_texts: list[str], source_url: str = "") -> dict:
"""HTML + CSS 텍스트에서 hex 빈도 분석 → primary/accent/text + palette. (fetch 없음)"""
all_text_chunks: list[str] = list(STYLE_BLOCK.findall(html))
all_text_chunks.append(html)
all_text_chunks.extend(css_texts)
counter: Counter = Counter()
for text in all_text_chunks:
for color in _extract_hex(text):
if color in NOISE:
continue
counter[color] += 1
if not counter:
logger.info("[brand_parser] no colors extracted from %s", source_url)
return {}
clustered = _cluster(counter)
chromatic = [c for c, _ in clustered if not _is_grayscale(c)]
grayscale = [c for c, _ in clustered if _is_grayscale(c)]
palette_top = clustered[:8]
palette = [{"name": f"색상 {i+1}", "hex": h, "usage": f"빈도 {n}"} for i, (h, n) in enumerate(palette_top)]
return {
"brand_colors": {
"primary": chromatic[0] if chromatic else None,
"accent": chromatic[1] if len(chromatic) > 1 else None,
"text": grayscale[0] if grayscale else None,
},
"color_palette": palette,
"extracted_from": "html+css",
}

86
app/services/branding.py Normal file
View File

@ -0,0 +1,86 @@
"""report 단계 - Gemini Vision 으로 로고 묘사 + 채널 로고 매칭."""
import logging
import os
from urllib.parse import urlparse
from common.db.source import (
select_run_raw_data, update_raw_info_merge,
select_branding_info_id, select_mainpage_logo_url,
)
from common.utils import _run_optional_step
from integrations.llm.gemini_vision import VisionClient
logger = logging.getLogger(__name__)
async def _describe_logo(analysis_run_id: str, info_id: int, vc: VisionClient) -> None:
"""공식 로고 정성 묘사. branding raw_info["brandAssets"] 머지.
호출 우선순위: raw_info.logo_url 컬럼 (HTML parser canonical) firecrawl 메타 fallback."""
raw = await select_run_raw_data(analysis_run_id)
mainpage = raw.get("mainpage") or {}
homepage_url = mainpage.get("sourceUrl") or ""
branding_meta = mainpage.get("branding") or {}
column_logo = await select_mainpage_logo_url(analysis_run_id)
candidates = [u for u in [
column_logo,
branding_meta.get("logoUrl"),
branding_meta.get("faviconUrl"),
] if u]
if homepage_url:
parsed = urlparse(homepage_url)
if parsed.scheme and parsed.netloc:
candidates.append(f"{parsed.scheme}://{parsed.netloc}/favicon.ico")
if not candidates:
logger.info("[brand_logo] skip — no candidates")
return
logger.info("[brand_logo] start run=%s candidates=%d", analysis_run_id, len(candidates))
result: dict = {}
for cand in candidates:
result = await vc.analyze_brand_assets(logo_url=cand, homepage_url=homepage_url)
if result:
break
if result:
await update_raw_info_merge(info_id, {"brandAssets": result})
logger.info("[brand_logo] done keys=%s", list(result.keys()) if result else None)
async def _describe_channel_logos(analysis_run_id: str, info_id: int, vc: VisionClient) -> None:
"""채널 프로필 로고를 공식 로고와 비교. branding raw_info["channelLogos"] 머지."""
raw = await select_run_raw_data(analysis_run_id)
official = await select_mainpage_logo_url(analysis_run_id)
_label = {
"instagram": "Instagram",
"facebook": "Facebook",
"youtube": "YouTube",
"instagram_en": "Instagram EN",
"facebook_en": "Facebook EN",
"tiktok": "TikTok",
}
logos = [{"channel": label, "url": img}
for key, label in _label.items()
if (img := (raw.get(key) or {}).get("_logo_url"))]
if not logos:
logger.info("[channel_logos] skip — no channel profileImages")
return
logger.info("[channel_logos] start run=%s channels=%s official=%s",
analysis_run_id, [l["channel"] for l in logos], bool(official))
result = await vc.describe_channel_logos(official, logos)
if result:
await update_raw_info_merge(info_id, {"channelLogos": result})
logger.info("[channel_logos] done keys=%s", list(result.keys()) if result else None)
async def analyze_branding(analysis_run_id: str) -> None:
"""report build 직전 호출 — 로고 묘사 + 채널 로고 매칭 (Gemini). 둘 다 격리."""
api_key = os.getenv("GEMINI_API_KEY")
if not api_key:
logger.info("[branding] skip — GEMINI_API_KEY 없음")
return
branding_info_id = await select_branding_info_id(analysis_run_id)
if branding_info_id is None:
logger.info("[branding] skip — branding source 없음 run=%s", analysis_run_id)
return
vc = VisionClient(api_key)
logger.info("[branding] start run=%s", analysis_run_id)
await _run_optional_step(_describe_logo(analysis_run_id, branding_info_id, vc), "brand_logo")
await _run_optional_step(_describe_channel_logos(analysis_run_id, branding_info_id, vc), "channel_logos")
logger.info("[branding] done run=%s", analysis_run_id)

View File

@ -1,115 +1,199 @@
import asyncio
import logging
from common.db import (
fetchone,
set_instagram_status, save_instagram_raw_data,
set_facebook_status, save_facebook_raw_data,
set_naver_blog_status, save_naver_blog_raw_data,
set_youtube_status, save_youtube_raw_data,
set_gangnam_unni_status, save_gangnam_unni_raw_data,
execute, save_hospital_raw_data,
)
from common.db.hospital import update_hospital_status, update_hospital
from common.db.source import select_run_sources, update_raw_info_status, update_raw_info
from common.utils import get_env, _run_optional_step
from integrations.apify import ApifyClient
from integrations.naver import NaverClient
from integrations.youtube import YouTubeClient
from integrations.firecrawl import FirecrawlClient
from services.enrichment import collect_brand_assets, collect_extra_channels, collect_channel_logos
from models.status import SourceType
from integrations.site_fetcher import fetch_html_and_css
from services.brand_parser import find_logo_url_in_html, extract_brand_colors_from_text
from common.db.source import update_raw_info_merge, update_raw_info_logo_url, select_run_raw_data
from common.db.base import fetchone
from services.facebook_audit import transform_for_storage as transform_facebook
logger = logging.getLogger(__name__)
async def collect_instagram(analysis_run_id: str, row_id: int, url: str) -> None:
async def _save_with_logo(info_id: int, data: dict) -> None:
await update_raw_info(info_id, data)
if data.get("profileImage"):
await update_raw_info_logo_url(info_id, data["profileImage"])
async def collect_instagram(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[instagram] start run=%s url=%s", analysis_run_id, url)
await set_instagram_status(row_id, "processing")
await update_raw_info_status(info_id, "processing")
data = await ApifyClient(get_env("APIFY_API_TOKEN")).get_instagram_profile(url)
await save_instagram_raw_data(row_id, data)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[instagram] failed run=%s", analysis_run_id)
return
await _save_with_logo(info_id, data)
logger.info("[instagram] done run=%s", analysis_run_id)
async def collect_facebook(analysis_run_id: str, row_id: int, url: str) -> None:
async def collect_facebook(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[facebook] start run=%s url=%s", analysis_run_id, url)
await set_facebook_status(row_id, "processing")
await update_raw_info_status(info_id, "processing")
data = await ApifyClient(get_env("APIFY_API_TOKEN")).get_facebook_page(url)
await save_facebook_raw_data(row_id, data)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[facebook] failed run=%s", analysis_run_id)
return
data = transform_facebook(data)
await _save_with_logo(info_id, data)
logger.info("[facebook] done run=%s", analysis_run_id)
async def collect_naver_blog(analysis_run_id: str, row_id: int, url: str) -> None:
async def collect_naver_blog(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[naver_blog] start run=%s url=%s", analysis_run_id, url)
await set_naver_blog_status(row_id, "processing")
await update_raw_info_status(info_id, "processing")
data = await NaverClient(get_env("NAVER_CLIENT_ID"), get_env("NAVER_CLIENT_SECRET")).get_blog_rss(url)
await save_naver_blog_raw_data(row_id, data)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[naver_blog] failed run=%s", analysis_run_id)
return
await update_raw_info(info_id, data)
logger.info("[naver_blog] done run=%s", analysis_run_id)
async def collect_youtube(analysis_run_id: str, row_id: int, url: str) -> None:
async def collect_youtube(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[youtube] start run=%s url=%s", analysis_run_id, url)
await set_youtube_status(row_id, "processing")
await update_raw_info_status(info_id, "processing")
data = await YouTubeClient(get_env("YOUTUBE_API_KEY")).get_channel(url)
await save_youtube_raw_data(row_id, data)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[youtube] failed run=%s", analysis_run_id)
return
await _save_with_logo(info_id, data)
logger.info("[youtube] done run=%s", analysis_run_id)
async def collect_gangnam_unni(analysis_run_id: str, row_id: int, url: str) -> None:
async def collect_gangnam_unni(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[gangnam_unni] start run=%s url=%s", analysis_run_id, url)
await set_gangnam_unni_status(row_id, "processing")
await update_raw_info_status(info_id, "processing")
data = await FirecrawlClient(get_env("FIRECRAWL_API_KEY")).get_gangnam_unni(url)
await save_gangnam_unni_raw_data(row_id, data)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[gangnam_unni] failed run=%s", analysis_run_id)
return
await update_raw_info(info_id, data)
logger.info("[gangnam_unni] done run=%s", analysis_run_id)
async def collect_clinic_info(analysis_run_id: str, hospital_id: str, url: str) -> None:
logger.info("[clinic] start run=%s url=%s", analysis_run_id, url)
await execute("UPDATE hospital_baseinfo SET status = 'processing' WHERE hospital_id = %s", (hospital_id,))
async def collect_mainpage(analysis_run_id: str, info_id: int, hospital_id: str, url: str) -> None:
logger.info("[mainpage] start run=%s url=%s", analysis_run_id, url)
await update_raw_info_status(info_id, "processing")
await update_hospital_status(hospital_id, "processing")
data = await FirecrawlClient(get_env("FIRECRAWL_API_KEY")).fetch_clinic_info(url)
await save_hospital_raw_data(hospital_id, data, analysis_run_id=analysis_run_id)
logger.info("[clinic] done run=%s", analysis_run_id)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[mainpage] failed run=%s", analysis_run_id)
return
# 홈페이지 URL 자체도 raw_data 에 박아둬야 brand_assets / 분석 단계에서 mainpage URL 재조회 없이 사용 가능.
data = {**data, "sourceUrl": url}
await update_raw_info(info_id, data)
await update_hospital(hospital_id, data, analysis_run_id=analysis_run_id)
logger.info("[mainpage] done run=%s", analysis_run_id)
async def collect_all(
analysis_run_id: str,
hospital_id: str,
instagram_id: int | None = None,
facebook_id: int | None = None,
naver_blog_id: int | None = None,
youtube_id: int | None = None,
gangnam_unni_id: int | None = None,
tiktok_url: str | None = None,
instagram_en_url: str | None = None,
facebook_en_url: str | None = None,
) -> None:
async def _url(table: str, row_id: int) -> str:
row = await fetchone(f"SELECT url FROM {table} WHERE id = %s", (row_id,))
return row["url"] if row else ""
async def collect_tiktok(analysis_run_id: str, info_id: int, url: str) -> None:
logger.info("[tiktok] start run=%s url=%s", analysis_run_id, url)
await update_raw_info_status(info_id, "processing")
data = await ApifyClient(get_env("APIFY_API_TOKEN")).get_tiktok_profile(url)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[tiktok] failed run=%s", analysis_run_id)
return
await _save_with_logo(info_id, data)
logger.info("[tiktok] done run=%s", analysis_run_id)
hospital = await fetchone("SELECT url FROM hospital_baseinfo WHERE hospital_id = %s", (hospital_id,))
tasks = [collect_clinic_info(analysis_run_id, hospital_id, hospital["url"])]
if instagram_id:
tasks.append(collect_instagram(analysis_run_id, instagram_id, await _url("instagram_data", instagram_id)))
if facebook_id:
tasks.append(collect_facebook(analysis_run_id, facebook_id, await _url("facebook_data", facebook_id)))
if naver_blog_id:
tasks.append(collect_naver_blog(analysis_run_id, naver_blog_id, await _url("naver_blog_data", naver_blog_id)))
if youtube_id:
tasks.append(collect_youtube(analysis_run_id, youtube_id, await _url("youtube_data", youtube_id)))
if gangnam_unni_id:
tasks.append(collect_gangnam_unni(analysis_run_id, gangnam_unni_id, await _url("gangnam_unni_data", gangnam_unni_id)))
async def collect_naver_cafe(analysis_run_id: str, info_id: int, url: str) -> None:
"""카페는 로그인 필요라 본문 못 봄. URL 활성·cafeId·이름 언급수만 신호로 수집."""
logger.info("[naver_cafe] start run=%s url=%s", analysis_run_id, url)
await update_raw_info_status(info_id, "processing")
data = await NaverClient(get_env("NAVER_CLIENT_ID"), get_env("NAVER_CLIENT_SECRET")).get_cafe_info(url)
if data is None:
await update_raw_info_status(info_id, "failed")
logger.warning("[naver_cafe] failed run=%s", analysis_run_id)
return
await update_raw_info(info_id, data)
logger.info("[naver_cafe] done run=%s", analysis_run_id)
async def collect_kakaotalk(analysis_run_id: str, info_id: int, url: str) -> None:
"""카카오톡은 수집 X — URL 보관만. LLM이 채널 존재 신호로만 사용."""
logger.info("[kakaotalk] url-only run=%s url=%s", analysis_run_id, url)
await update_raw_info(info_id, {"url": url})
async def collect_brand_basics(analysis_run_id: str, info_id: int) -> None:
logger.info("[brand_basics] start run=%s info=%s", analysis_run_id, info_id)
raw = await select_run_raw_data(analysis_run_id)
mainpage = raw.get("mainpage") or {}
homepage_url = mainpage.get("sourceUrl") or ""
branding_meta = mainpage.get("branding") or {}
html, css_texts = await fetch_html_and_css(homepage_url) if homepage_url else ("", [])
html_logo_url = find_logo_url_in_html(html, homepage_url, css_texts) if html else None
css_colors = extract_brand_colors_from_text(html, css_texts, homepage_url) if html else {}
logo_url = html_logo_url or branding_meta.get("logoUrl") or branding_meta.get("ogImage")
if logo_url:
mainpage_row = await fetchone(
"SELECT ri.info_id FROM raw_info ri JOIN remote_source rs USING (source_id)"
" WHERE ri.analysis_run_id = %s AND rs.source_type = 'mainpage' LIMIT 1",
(analysis_run_id,),
)
if mainpage_row:
await update_raw_info_logo_url(mainpage_row["info_id"], logo_url)
payload: dict = {}
if css_colors:
if css_colors.get("brand_colors"): payload["brand_colors"] = css_colors["brand_colors"]
if css_colors.get("color_palette"): payload["color_palette"] = css_colors["color_palette"]
payload["color_source"] = "html+css"
if payload:
await update_raw_info_merge(info_id, {"brandAssets": payload})
logger.info("[brand_basics] done logo_url=%s colors=%s", bool(logo_url), bool(payload))
async def collect_all(analysis_run_id: str, hospital_id: str) -> None:
rows = await select_run_sources(analysis_run_id)
# source_type → collector. KR/EN 구분은 collector 입장에서 동일, language 컬럼만 다름.
_collectors = {
SourceType.INSTAGRAM: collect_instagram,
SourceType.FACEBOOK: collect_facebook,
SourceType.NAVER_BLOG: collect_naver_blog,
SourceType.YOUTUBE: collect_youtube,
SourceType.GANGNAM_UNNI: collect_gangnam_unni,
SourceType.TIKTOK: collect_tiktok,
SourceType.NAVER_CAFE: collect_naver_cafe,
SourceType.KAKAOTALK: collect_kakaotalk,
}
tasks = []
branding_info_id: int | None = None
for row in rows:
info_id = row["info_id"]
source_type = row["source_type"]
url = row["url"]
if source_type == SourceType.BRANDING:
branding_info_id = info_id # mainpage·채널 수집 끝난 뒤 2단계에서 사용
continue
if source_type == SourceType.MAINPAGE:
tasks.append(collect_mainpage(analysis_run_id, info_id, hospital_id, url))
elif source_type in _collectors:
tasks.append(_collectors[source_type](analysis_run_id, info_id, url))
await asyncio.gather(*tasks, return_exceptions=True)
# 아래 3단계는 모두 hospital raw_data를 read-modify-write 하므로 race 방지 위해 순차.
# brand_assets : clinic_info가 채운 branding.logoUrl로 공식 로고/hex 추출
# extra_channels: 틱톡/인스타EN/페북EN 수집
# channel_logos : 공식 로고(brand_assets)+채널 profileImage(extra_channels) 채워진 뒤 Vision 비교
# 부가 기능이라 실패해도 리포트는 나와야 하므로 _run_optional_step으로 각각 격리.
await _run_optional_step(collect_brand_assets(analysis_run_id, hospital_id), "brand_assets")
await _run_optional_step(
collect_extra_channels(
analysis_run_id, hospital_id,
tiktok_url=tiktok_url, instagram_en_url=instagram_en_url, facebook_en_url=facebook_en_url,
),
"extra_channels",
)
await _run_optional_step(collect_channel_logos(analysis_run_id, hospital_id), "channel_logos")
# 2단계: branding (brandAssets → channelLogos 한 raw_info 안에 머지). mainpage·채널 raw_data 의존이라 순차.
# 부가 기능이라 실패해도 리포트는 나와야 하므로 _run_optional_step 으로 격리.
if branding_info_id is not None:
await _run_optional_step(collect_brand_basics(analysis_run_id, branding_info_id), "brand_basics")

View File

@ -1,175 +0,0 @@
import asyncio
import json
import logging
import os
from urllib.parse import urlparse
from common.db import fetchone, fetch_raw, merge_hospital_raw_data
from common.utils import get_env
from integrations.apify import ApifyClient
from integrations.vision import VisionClient
from integrations.color_extractor import extract_brand_assets_from_site
logger = logging.getLogger(__name__)
async def collect_brand_assets(analysis_run_id: str, hospital_id: str) -> None:
"""홈페이지에서 로고 URL + brand hex 색상을 뽑아 raw_data["brandAssets"]에 저장.
- 로고 URL/hex: HTML·CSS 정규식 (color_extractor) Vision 의존 X, 사이트 전체 컬러 시스템이 정확.
- 로고 정성 묘사(심볼/워드마크/): Gemini Vision (GEMINI_API_KEY 없으면 색상만 저장하고 skip).
"""
logger.info("[brand_assets] start run=%s", analysis_run_id)
row = await fetchone(
"SELECT raw_data, url FROM hospital_baseinfo WHERE hospital_id = %s",
(hospital_id,),
)
if not row:
return
raw = row["raw_data"]
raw_data = json.loads(raw) if isinstance(raw, str) else (raw or {})
branding = raw_data.get("branding") or {}
homepage_url = row["url"]
# 0~1. 사이트 1회 fetch로 logo URL + brand hex 동시 추출 (img/background-image/CSS .logo, Vision 의존 X)
site = await extract_brand_assets_from_site(homepage_url) if homepage_url else {}
html_logo_url = site.get("logo_url")
css_colors = site.get("colors") or {}
if html_logo_url:
logger.info("[brand_assets] HTML logo found: %s", html_logo_url)
if css_colors:
logger.info("[brand_assets] css colors: %s", css_colors.get("brand_colors"))
# 2. 로고/대표 이미지 후보 (logo → og:image → favicon 순)
logo_url = html_logo_url or branding.get("logoUrl")
og_image = branding.get("ogImage")
favicon = branding.get("faviconUrl")
candidates: list[tuple[str, str]] = []
if logo_url: candidates.append(("logo", logo_url))
if og_image: candidates.append(("og", og_image))
if favicon: candidates.append(("favicon", favicon))
if homepage_url:
parsed = urlparse(homepage_url)
if parsed.scheme and parsed.netloc:
candidates.append(("favicon", f"{parsed.scheme}://{parsed.netloc}/favicon.ico"))
if not candidates and not css_colors:
logger.info("[brand_assets] skip — no logo/og/favicon candidates and no CSS colors")
return
# 3. Vision은 로고 정성 묘사만 (hex는 CSS 추출이 더 정확). 키 없으면 색상만 저장.
result: dict = {}
used_kind: str | None = None
api_key = os.getenv("GEMINI_API_KEY")
if api_key and candidates:
vc = VisionClient(api_key)
for kind, cand in candidates:
result = await vc.analyze_brand_assets(logo_url=cand, homepage_url=homepage_url)
if result:
used_kind = kind
break
# favicon으로만 분석된 경우 진짜 로고가 아니므로 logo URL은 박지 않음 (묘사는 OK)
if result and used_kind == "favicon" and result.get("logo_images"):
result["logo_images"] = {"circle": None, "horizontal": None, "korean": None}
elif not api_key:
logger.info("[brand_assets] GEMINI_API_KEY not set — 색상만 저장, Vision 묘사 skip")
# 4. CSS에서 추출한 brand_colors/palette를 Vision보다 우선 사용
if css_colors:
if css_colors.get("brand_colors"):
result["brand_colors"] = css_colors["brand_colors"]
if css_colors.get("color_palette"):
result["color_palette"] = css_colors["color_palette"]
result["color_source"] = "html+css"
elif result:
result["color_source"] = "vision"
if result:
result["logo_source"] = used_kind or "none"
await merge_hospital_raw_data(hospital_id, {"brandAssets": result})
logger.info("[brand_assets] done keys=%s", list(result.keys()) if result else None)
async def collect_extra_channels(
analysis_run_id: str,
hospital_id: str,
tiktok_url: str | None = None,
instagram_en_url: str | None = None,
facebook_en_url: str | None = None,
) -> None:
"""틱톡 / 인스타 EN / 페북 EN 수집 → hospital raw_data에 저장 (별도 테이블 없이).
인스타EN·페북EN은 기존 Apify 수집기 재사용, 틱톡은 신규 액터."""
apify = ApifyClient(get_env("APIFY_API_TOKEN"))
jobs: dict = {}
if instagram_en_url:
jobs["instagramEn"] = apify.get_instagram_profile(instagram_en_url)
if facebook_en_url:
jobs["facebookEn"] = apify.get_facebook_page(facebook_en_url)
if tiktok_url:
jobs["tiktok"] = apify.get_tiktok_profile(tiktok_url)
if not jobs:
return
logger.info("[extra_channels] start run=%s channels=%s", analysis_run_id, list(jobs))
done = await asyncio.gather(*jobs.values(), return_exceptions=True)
results: dict = {}
for key, res in zip(jobs.keys(), done):
if isinstance(res, Exception):
logger.warning("[extra_channels] %s 수집 실패: %s", key, res)
elif res:
results[key] = res
if not results:
logger.info("[extra_channels] 수집 결과 없음 run=%s", analysis_run_id)
return
await merge_hospital_raw_data(hospital_id, results)
logger.info("[extra_channels] done run=%s keys=%s", analysis_run_id, list(results))
async def collect_channel_logos(analysis_run_id: str, hospital_id: str) -> None:
"""채널별 프로필 이미지(로고)를 모아 Gemini Vision으로 설명 + 공식 로고 일치 여부 평가.
hospital raw_data["channelLogos"] 저장. GEMINI_API_KEY 없으면 skip.
brand_assets(공식 로고)·extra_channels(틱톡/EN profileImage) 다음에 실행돼야 ."""
api_key = os.getenv("GEMINI_API_KEY")
if not api_key:
logger.info("[channel_logos] skip — GEMINI_API_KEY 없음")
return
hrow = await fetchone("SELECT raw_data FROM hospital_baseinfo WHERE hospital_id = %s", (hospital_id,))
raw = hrow["raw_data"] if hrow else None
raw_data = json.loads(raw) if isinstance(raw, str) else (raw or {})
official = ((raw_data.get("brandAssets") or {}).get("logo_images") or {}).get("horizontal")
run = await fetchone(
"SELECT instagram_data_id, facebook_data_id, youtube_data_id"
" FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
logos: list[dict] = []
# 전용 테이블 채널 (KR)
for ch, table, col in [
("Instagram", "instagram_data", "instagram_data_id"),
("Facebook", "facebook_data", "facebook_data_id"),
("YouTube", "youtube_data", "youtube_data_id"),
]:
rid = (run or {}).get(col)
if rid:
d = await fetch_raw(table, rid) or {}
if d.get("profileImage"):
logos.append({"channel": ch, "url": d["profileImage"]})
# 추가 채널 (hospital raw_data)
for ch, key in [("Instagram EN", "instagramEn"), ("Facebook EN", "facebookEn"), ("TikTok", "tiktok")]:
img = (raw_data.get(key) or {}).get("profileImage")
if img:
logos.append({"channel": ch, "url": img})
if not logos:
logger.info("[channel_logos] skip — 채널 프로필 이미지 없음")
return
logger.info("[channel_logos] start run=%s channels=%s official=%s", analysis_run_id,
[l["channel"] for l in logos], bool(official))
result = await VisionClient(api_key).describe_channel_logos(official, logos)
if result:
# Vision이 못 본 채널도 url은 채워둠 (프론트에서 이미지 표시용)
result["logos"] = logos
await merge_hospital_raw_data(hospital_id, {"channelLogos": result})
logger.info("[channel_logos] done run=%s keys=%s", analysis_run_id, list(result.keys()) if result else None)

View File

@ -0,0 +1,106 @@
"""Facebook audit 페이지(KR·EN)를 수집 데이터로 구성.
수치 지표(최근 게시일·게시 빈도·참여율) **수집 시점에** 결정적으로 산출해 DB에 박는다 (transform_for_storage).
콘텐츠 주제(top_content_type) 캡션 본문 이해가 필요해 LLM이 채운다 (리포트 프롬프트 지시)."""
from datetime import datetime, timezone
from common.utils import parse_ts
from integrations.llm.schemas.report import FacebookAudit
def _humanize_age(days: int) -> str:
days = max(days, 0)
if days == 0: return "오늘"
if days < 7: return f"{days}일 전"
if days < 30: return f"{days // 7}주 전"
if days < 365: return f"{days // 30}개월 전"
return f"{days // 365}년 전"
def _frequency_label(avg_gap_days: float) -> str:
"""게시물 사이 평균 간격(일) → 빈도 라벨."""
if avg_gap_days <= 1.5: return "거의 매일"
if avg_gap_days <= 10: return f"{7 / avg_gap_days:.1f}"
if avg_gap_days <= 45: return f"{30 / avg_gap_days:.1f}"
return "비정기 (분기 이상 간격)"
def _engagement_text(posts: list[dict]) -> str:
"""게시물당 좋아요/반응/공유/조회를 min~max 범위로. 전부 0인 지표는 제외.
댓글은 posts actor가 줘서 '댓글 거의 없음' 고정 부가 (FB 페이지는 댓글 희박이 일반적)."""
def _rng(vals: list[int], label: str, unit: str) -> str | None:
lo, hi = min(vals), max(vals)
if hi == 0:
return None
return f"{label} {lo}{unit}" if lo == hi else f"{label} {lo}~{hi}{unit}"
parts = [
_rng([p.get("likes", 0) for p in posts], "좋아요", ""),
_rng([p.get("reactions", 0) for p in posts], "반응", ""),
_rng([p.get("shares", 0) for p in posts], "공유", ""),
]
vid_views = [p.get("views", 0) for p in posts if p.get("isVideo")]
if vid_views:
parts.append(_rng(vid_views, "영상 조회", ""))
parts = [x for x in parts if x]
if not parts:
return "게시물당 참여 거의 없음"
return "게시물당 " + " · ".join(parts) + " · 댓글 거의 없음"
def transform_for_storage(fb: dict | None) -> dict | None:
"""apify 원본 → DB에 저장할 최종 형태.
- 수치 지표(recent_post_age·post_frequency·engagement) 자리에서 계산해 박음.
- 게시물은 캡션·타입만 남김 (raw 숫자/timestamp는 어차피 재계산 하므로 버림).
수집 시점에 계산 리포트 생성 때는 그대로 갖다 박기만 ."""
if not isinstance(fb, dict):
return fb
posts = fb.get("latestPosts") or []
out = {k: v for k, v in fb.items() if k != "latestPosts"}
if posts:
dts = sorted((d for d in (parse_ts(p.get("timestamp")) for p in posts) if d), reverse=True)
if dts:
out["recent_post_age"] = _humanize_age((datetime.now(timezone.utc) - dts[0]).days)
if len(dts) > 1:
avg_gap = ((dts[0] - dts[-1]).days or 1) / (len(dts) - 1)
out["post_frequency"] = _frequency_label(avg_gap)
out["engagement"] = _engagement_text(posts)
out["latestPosts"] = [
{"caption": (p.get("text") or "")[:160],
"type": "video" if p.get("isVideo") else "image"}
for p in posts
]
else:
out["latestPosts"] = []
return out
def _page_patch(fb: dict, language: str, label: str) -> dict:
"""저장된 페북 페이지 → FacebookPage 스키마 필드 패치. 수치 지표는 수집 시점에 박혀있어 그대로 복사.
language/label 데이터 있을 때만 명시적으로 박음 template-copy KR 값을 EN 슬롯에 잘못 상속시키는 방지."""
p: dict = {}
if fb.get("pageUrl"): p["url"] = p["link"] = fb["pageUrl"]
if fb.get("pageName"): p["page_name"] = fb["pageName"]
if fb.get("followers"): p["followers"] = fb["followers"]
if fb.get("intro"): p["bio"] = fb["intro"]
if fb.get("categories"): p["category"] = ", ".join(fb["categories"])
if fb.get("website"): p["linked_domain"] = fb["website"]
if fb.get("reviews") is not None: p["reviews"] = fb["reviews"]
if fb.get("following") is not None: p["following"] = fb["following"]
for key in ("recent_post_age", "post_frequency", "engagement"):
if fb.get(key): p[key] = fb[key]
if p:
p["language"] = language
p["label"] = label
return p
def build_facebook_audit(facebook: dict, facebook_en: dict, llm_pages: list[dict] | None = None) -> dict:
"""KR·EN 페북 페이지 구성. logo/logo_description 은 LLM Vision 결과(첫 페이지) 모든 페이지에 공통 적용,
나머지 필드는 코드가 수집 데이터로 계산."""
llm_logo = {k: v for k, v in ((llm_pages or [{}])[0]).items() if k in {"logo", "logo_description"} and v}
pages = [{**llm_logo, **p} for p in (
_page_patch(facebook, "KR", "페이스북 KR"),
_page_patch(facebook_en, "EN", "페이스북 EN"),
) if p]
return FacebookAudit.model_validate({"pages": pages}).model_dump(exclude_unset=True)

View File

@ -2,7 +2,8 @@ import logging
from fastapi import HTTPException, UploadFile
from common.db import execute, fetchall, fetchone, insert_file_row
from common.db.run import select_run
from common.db.file_data import insert_file, select_run_files, select_file, delete_file
from integrations.azure_blob import AzureBlobUploader
from models.file import FileListItem, FileType, FileUploadResponse
@ -31,10 +32,7 @@ async def upload_analysis_file(
content_type: str | None = None,
) -> tuple[int, str]:
"""analysis_run에 딸린 파일 업로드. Blob 업로드 + file_data row 생성. (file_id, url) 반환."""
run = await fetchone(
"SELECT hospital_id FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
run = await select_run(analysis_run_id)
if not run:
raise HTTPException(status_code=404, detail="analysis_run not found")
hospital_id = run["hospital_id"]
@ -47,7 +45,7 @@ async def upload_analysis_file(
content_type=content_type,
)
file_id = await insert_file_row(
file_id = await insert_file(
analysis_run_id=analysis_run_id,
hospital_id=hospital_id,
file_type=file_type,
@ -61,12 +59,7 @@ async def upload_analysis_file(
async def list_analysis_files(analysis_run_id: str) -> list[dict]:
"""analysis_run에 딸린 (삭제 안 된) 파일 목록."""
return await fetchall(
"SELECT id, file_type, file_name, file_url, size_bytes, created_at FROM file_data"
" WHERE analysis_run_id = %s AND is_deleted = FALSE"
" ORDER BY created_at DESC",
(analysis_run_id,),
)
return await select_run_files(analysis_run_id)
async def handle_analysis_file_upload(
@ -102,7 +95,7 @@ async def handle_analysis_file_upload(
async def get_analysis_files_response(analysis_run_id: str) -> list[FileListItem]:
"""run 존재 확인 + 응답 모델 생성."""
if not await fetchone("SELECT 1 FROM analysis_runs WHERE analysis_run_id = %s", (analysis_run_id,)):
if not await select_run(analysis_run_id):
raise HTTPException(status_code=404, detail="analysis_run not found")
rows = await list_analysis_files(analysis_run_id)
return [FileListItem(**{**r, "created_at": str(r["created_at"])}) for r in rows]
@ -110,14 +103,8 @@ async def get_analysis_files_response(analysis_run_id: str) -> list[FileListItem
async def soft_delete_analysis_file(analysis_run_id: str, file_id: int) -> None:
"""analysis_run에 딸린 파일을 소프트 삭제. 멱등성 보장."""
row = await fetchone(
"SELECT id FROM file_data WHERE id = %s AND analysis_run_id = %s",
(file_id, analysis_run_id),
)
row = await select_file(file_id, analysis_run_id)
if not row:
raise HTTPException(status_code=404, detail="file not found")
await execute(
"UPDATE file_data SET is_deleted = TRUE WHERE id = %s AND is_deleted = FALSE",
(file_id,),
)
await delete_file(file_id)
logger.info("soft-deleted analysis file run=%s file_id=%s", analysis_run_id, file_id)

View File

@ -1,6 +1,8 @@
"""Instagram audit 계정(KR·EN)을 수집 데이터로 구성.
fix (handle/followers/highlights/content_format ) 전부 코드에서 박는다 LLM 출력 무시."""
from integrations.llm.schemas.report import InstagramAudit
_MEDIA = {"GraphImage": "이미지", "GraphSidecar": "카드뉴스", "GraphVideo": "영상/릴스"}
@ -38,11 +40,11 @@ def _account(data: dict, language: str, label: str, channel: str, channel_logos:
}
def build_instagram_accounts(instagram: dict, instagram_en: dict, channel_logos: dict) -> list[dict]:
def build_instagram_audit(instagram: dict, instagram_en: dict, channel_logos: dict) -> dict:
"""KR·EN 인스타 계정 리스트 구성 (username 있는 것만)."""
accounts: list[dict] = []
if instagram.get("username"):
accounts.append(_account(instagram, "KR", "인스타그램 KR", "Instagram", channel_logos))
if instagram_en.get("username"):
accounts.append(_account(instagram_en, "EN", "인스타그램 EN", "Instagram EN", channel_logos))
return accounts
return InstagramAudit.model_validate({"accounts": accounts}).model_dump()

View File

@ -0,0 +1,96 @@
"""mockup 7개 역분석 — 채널 규모별 3개월/12개월 target 성장률 공식."""
from integrations.llm.schemas.report import KPIMetric
def _round_clean(n: int) -> int:
if n < 100: return n
if n < 1000: return round(n / 100) * 100
if n < 10_000: return round(n / 500) * 500
if n < 100_000: return round(n / 1000) * 1000
if n < 1_000_000: return round(n / 5000) * 5000
return round(n / 50_000) * 50_000
def _target_multiplier(current: int) -> tuple[float, float]:
if current < 1_000: return (2.5, 9.0)
if current < 5_000: return (1.7, 4.0)
if current < 25_000: return (1.5, 2.5)
if current < 50_000: return (1.3, 2.2)
return (1.1, 1.9)
def _follower_kpi(metric: str, val: int | None, unit: str = "") -> dict | None:
if not val: return None
m3, m12 = _target_multiplier(val)
return {
"metric": metric,
"current": f"{val:,}{unit}",
"target_3_month": f"{_round_clean(int(val * m3)):,}{unit}",
"target_12_month": f"{_round_clean(int(val * m12)):,}{unit}",
}
def _blog_frequency(posts: list) -> tuple[str, str, str] | None:
"""RSS posts timestamp로 (current, target_3m, target_12m) 라벨 반환. target은 절대 downgrade 안 함."""
from common.utils import parse_ts
dts = sorted((d for d in (parse_ts(p.get("postDate")) for p in posts) if d), reverse=True)
if len(dts) < 2: return None
avg_gap = (dts[0] - dts[-1]).days / (len(dts) - 1)
if avg_gap > 90: current = f"방치 ({dts[0].strftime('%Y-%m')})"
elif avg_gap <= 1: current = f"{7 // max(int(avg_gap), 1)}"
elif avg_gap <= 3: current = "주 2~3회"
elif avg_gap <= 14: current = "주 1~2회"
elif avg_gap <= 30: current = f"{max(30 // int(avg_gap), 1)}"
else: current = "월 1회 미만"
if avg_gap > 3: return current, "주 2회", "주 3회"
if avg_gap > 2: return current, "주 3회", "주 5회"
if avg_gap > 1: return current, "주 5회", "주 7회"
return current, f"{current} 유지", f"{current} 유지"
def build_kpi_dashboard(
instagram: dict, facebook: dict, youtube: dict, gangnam_unni: dict, hospital: dict,
naver_blog: dict | None = None,
) -> list[dict]:
ig_en = hospital.get("instagramEn") or {}
fb_en = hospital.get("facebookEn") or {}
tiktok = hospital.get("tiktok") or {}
cafe = hospital.get("naverCafe") or {}
kpis: list[dict] = []
for k in [
_follower_kpi("YouTube 구독자", youtube.get("subscribers")),
_follower_kpi("Instagram KR 팔로워", instagram.get("followers")),
_follower_kpi("Instagram EN 팔로워", ig_en.get("followers")),
_follower_kpi("Facebook KR 팔로워", facebook.get("followers")),
_follower_kpi("Facebook EN 팔로워", fb_en.get("followers")),
_follower_kpi("TikTok 팔로워", tiktok.get("followers")),
_follower_kpi("Naver Cafe 회원 수", cafe.get("memberCount")),
]:
if k: kpis.append(k)
if naver_blog:
freq = _blog_frequency(naver_blog.get("posts") or [])
if freq:
cur, t3, t12 = freq
kpis.append({
"metric": "네이버 블로그 포스팅 빈도",
"current": cur,
"target_3_month": t3,
"target_12_month": t12,
})
gu_reviews = gangnam_unni.get("totalReviews")
if gu_reviews:
if gu_reviews < 1000: rm3, rm12 = 2.0, 6.0
elif gu_reviews < 5000: rm3, rm12 = 1.10, 1.50
else: rm3, rm12 = 1.07, 1.27
kpis.append({
"metric": "강남언니 리뷰",
"current": f"{gu_reviews:,}",
"target_3_month": f"{_round_clean(int(gu_reviews * rm3)):,}",
"target_12_month": f"{_round_clean(int(gu_reviews * rm12)):,}",
})
return [KPIMetric.model_validate(k).model_dump() for k in kpis]

View File

@ -1,7 +1,9 @@
import asyncio
import json
import logging
from common.db import fetchone, execute
from common.db.run import select_run
from common.db.hospital import select_hospital
from common.db.market import upsert_market_status, upsert_market_result
from common.db.source import select_run_raw_data
from integrations.llm.llm_service import LLMService
from integrations.llm.prompt import (
market_competitors_prompt,
@ -18,49 +20,27 @@ _TYPES = ["competitors", "keywords", "trend", "target_audience"]
async def _save(analysis_run_id: str, analysis_type: str, result, exc: Exception | None) -> None:
if exc:
logger.warning("[market] %s failed run=%s: %s", analysis_type, analysis_run_id, exc)
await execute(
"INSERT INTO market_analysis (analysis_run_id, analysis_type, status)"
" VALUES (%s, %s, 'failed')"
" ON DUPLICATE KEY UPDATE status = 'failed'",
(analysis_run_id, analysis_type),
)
await upsert_market_status(analysis_run_id, analysis_type, "failed")
else:
await execute(
"INSERT INTO market_analysis (analysis_run_id, analysis_type, status, data)"
" VALUES (%s, %s, 'done', %s)"
" ON DUPLICATE KEY UPDATE status = 'done', data = VALUES(data)",
(analysis_run_id, analysis_type, json.dumps(result.model_dump(), ensure_ascii=False)),
)
await upsert_market_result(analysis_run_id, analysis_type, result.model_dump())
async def run_market_analysis(analysis_run_id: str) -> None:
logger.info("[market] start run=%s", analysis_run_id)
run = await fetchone(
"SELECT hospital_id FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
clinic = await fetchone(
"SELECT hospital_name, road_address, raw_data FROM hospital_baseinfo WHERE hospital_id = %s",
(run["hospital_id"],),
)
run = await select_run(analysis_run_id)
clinic = await select_hospital(run["hospital_id"])
raw = await select_run_raw_data(analysis_run_id)
mainpage = raw.get("mainpage") or {}
raw_data = clinic["raw_data"]
clinic_data = json.loads(raw_data) if isinstance(raw_data, str) else (raw_data or {})
clinic_name = clinic["hospital_name"] or ""
address = clinic["road_address"] or ""
services = clinic_data.get("services", [])
clinic_name = (clinic or {}).get("hospital_name") or ""
address = (clinic or {}).get("road_address") or ""
services = mainpage.get("services", [])
services_str = ", ".join(services[:3])
primary_service = services[0] if services else ""
for analysis_type in _TYPES:
await execute(
"INSERT INTO market_analysis (analysis_run_id, analysis_type, status)"
" VALUES (%s, %s, 'processing')"
" ON DUPLICATE KEY UPDATE status = 'processing'",
(analysis_run_id, analysis_type),
)
await upsert_market_status(analysis_run_id, analysis_type, "processing")
llm = LLMService(provider="perplexity")
results = await asyncio.gather(

View File

@ -1,5 +1,5 @@
import logging
from common.db import fetchone, execute
from common.db.run import select_run, update_run_status
from models.status import AnalysisStatus
from services.collect import collect_all
from services.market import run_market_analysis
@ -8,49 +8,23 @@ from services.analysis import run_report_task, run_plan_task
logger = logging.getLogger(__name__)
async def run_pipeline(analysis_run_id: str, extra_channels: dict | None = None) -> None:
async def run_pipeline(analysis_run_id: str) -> None:
logger.info("[pipeline] start run=%s", analysis_run_id)
extra_channels = extra_channels or {}
# ── 1. Collect ──────────────────────────────────────────────────────────
run = await fetchone(
"SELECT hospital_id, instagram_data_id, facebook_data_id,"
" naver_blog_data_id, youtube_data_id, gangnam_unni_data_id"
" FROM analysis_runs WHERE analysis_run_id = %s",
(analysis_run_id,),
)
await collect_all(
analysis_run_id,
hospital_id=run["hospital_id"],
instagram_id=run["instagram_data_id"],
facebook_id=run["facebook_data_id"],
naver_blog_id=run["naver_blog_data_id"],
youtube_id=run["youtube_data_id"],
gangnam_unni_id=run["gangnam_unni_data_id"],
tiktok_url=extra_channels.get("tiktok"),
instagram_en_url=extra_channels.get("instagram_en"),
facebook_en_url=extra_channels.get("facebook_en"),
)
run = await select_run(analysis_run_id)
await collect_all(analysis_run_id, hospital_id=run["hospital_id"])
# ── 2. Market ────────────────────────────────────────────────────────────
await execute(
"UPDATE analysis_runs SET status = %s WHERE analysis_run_id = %s",
(AnalysisStatus.ANALYZING, analysis_run_id),
)
await update_run_status(analysis_run_id, AnalysisStatus.ANALYZING)
await run_market_analysis(analysis_run_id)
# ── 3. Report ────────────────────────────────────────────────────────────
await run_report_task(analysis_run_id)
# ── 4. Plan ──────────────────────────────────────────────────────────────
await execute(
"UPDATE analysis_runs SET status = %s WHERE analysis_run_id = %s",
(AnalysisStatus.PLANNING, analysis_run_id),
)
await update_run_status(analysis_run_id, AnalysisStatus.PLANNING)
await run_plan_task(analysis_run_id)
await execute(
"UPDATE analysis_runs SET status = %s WHERE analysis_run_id = %s",
(AnalysisStatus.COMPLETED, analysis_run_id),
)
await update_run_status(analysis_run_id, AnalysisStatus.COMPLETED)
logger.info("[pipeline] done run=%s", analysis_run_id)

View File

@ -10,3 +10,4 @@ passlib[bcrypt]==1.7.4
python-multipart==0.0.26
uuid6==2025.0.1
aiomysql==0.3.2
resvg-py==0.3.2