James
|
20c33d2fb2
|
Document processing pipeline: OpenRouter OCR + Stepfun extraction
- Switch OCR from Fireworks Qwen to Gemini 3 Flash via OpenRouter
- Switch extraction/translation to Stepfun Step 3.5 Flash via OpenRouter
- Add CallOpenRouter() to lib/llm.go (OpenAI-compatible API)
- Add OpenRouter API key support to lib/config.go
- Fix extraction preamble causing Stepfun reasoning model to return null content
(heavy user-message preamble → short system message)
- Fix timestamp fallback: no date = 0, not today
- Fix document type: use file extension instead of hardcoded "pdf"
- Fix findUploadByFilename: Value filter was silently ignored, returning all uploads
- Fix entryQuery: skip category filter when EntryID is specified
- Fix extraction prompt contamination: replace real patient examples with generic ones
- Add reprocess button (↻) for document uploads
- Add delete cascade for upload entries (removes created document/extraction entries)
- Add upload/reprocess/delete logging throughout pipeline
- Consolidated templates, translations, email templates, landing pages
- Portal: defense, genome, dossier sections, MCP, OAuth updates
- Tools: translate pipeline improvements
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
2026-03-15 05:28:05 -04:00 |
Johan Jongsma