- Fixed TGI token limits (MAX_INPUT_TOKENS=12288, MAX_TOTAL_TOKENS=16384)
- Reduced per-step max_tokens for faster generation
- Improved clinical reasoning prompt (disease-level dx, not symptoms)
- Fixed Unicode encoding issues for Windows console
- Fixed error masking in orchestrator (failed steps now surface errors)
- Fixed endpoint URL in .env
- Added analyze_results.py for question-type categorization
- Results: 94% pipeline success, 38% top3 accuracy, 14% dx-only accuracy
- Paused endpoint to save costs
Add conflict detection step to CDS pipeline
bshepp
e684a6c
- Add ConflictType enum, ClinicalConflict, ConflictDetectionResult models to schemas
- Add conflicts field to CDSReport and conflict_detection to AgentState
- Create conflict_detection.py tool (compares guidelines vs patient data)
- Wire as Step 5 in orchestrator between guideline retrieval and synthesis
- Update synthesis to accept and format conflict data in prompt
- Add fallback to copy detected conflicts into report if LLM misses them
- Add prominent Conflicts & Gaps section to frontend CDSReport.tsx
with severity-coded cards, guideline vs patient data comparison,
and suggested resolutions
Initial commit: CDS Agent - Clinical Decision Support System
bshepp
f2c113d
- 5-step agentic pipeline: parse reason drug check RAG guidelines synthesize
- Backend: FastAPI + Python 3.10, Gemma 3 27B IT via Google AI Studio
- Frontend: Next.js 14 + React 18 + TypeScript + Tailwind CSS
- RAG: ChromaDB with 62 clinical guidelines across 14 specialties
- Drug APIs: OpenFDA + RxNorm/NLM
- Tests: E2E pipeline, 30 RAG quality queries (100% pass), 22 clinical scenarios
- Full documentation: architecture, test results, development log, writeup
