Remove all local PDF/OCR extraction and quality gate logic from
review_service.py, replacing it with calls to the shared-ai-platform
document extraction service via the SDK.
Removed from review_service.py:
- 17 local PDF/OCR methods (_extract_text_from_pdf, _classify_pdf_source,
_extract_text_from_pdf_with_ocr, _extract_text_from_image_with_ocr,
_evaluate_ocr_quality, _looks_like_platform_review_report, etc.)
- All PDF/OCR constants (PDF_TEXT_PATTERN, PDF_NOISE_LINE_PATTERN,
REPORT_PDF_MARKERS, OCR_GATE_*, PDF_SOURCE_KIND_*, etc.)
- Imports: pdfplumber, pypdfium2, pytesseract, PIL.Image
Added:
- extract_review_upload_text() now calls shared platform's
client.extract_document_text() directly
- _detect_upload_type() retained locally for upload validation only
- Protocol methods extract_document_text, classify_pdf_source in client
Removed from requirements.txt: pdfplumber, pytesseract
Added to config: shared_ai_platform_document_extraction_enabled
Also includes prior changes:
- Model settings and chat history stabilization
- Mobile UI polish
- Governance launch-closure enhancements
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Database password was incorrect; the host PostgreSQL uses 'postgres'
as the password, not the previous value. This was causing the backend
container to fail on startup (connection to PostgreSQL failed).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
yuqei
Codex