anupandey4/maternal-health-qa

Maternal & Child Health Q&A

A small India-focused maternal and child health Q&A endpoint built as the live evaluation target for a CeRAI evaluation-tool assignment.

Stack:

• —FastAPI
• —Groq llama-3.3-70b-versatile for generation
• —FAISS RAG over WHO / UNICEF / MoHFW maternal-health documents
• —Python-side safety pre-filter for hard-refuse and clinical-deflect cases (no external safety service required)

Endpoints:

• —POST /chat — body {"message": "..."} → {"response": "...", "sources": [...], "safety_action": "..."}
• —GET /docs — interactive Swagger UI
• —GET /health — liveness probe

Why this exists: It pairs with a reference evaluator (in the GitHub source repo) that demonstrates correct refusal handling, multilingual bias detection, and self-bias-free factuality — all three of which are observably broken in the upstream CeRAI AIEvaluationTool. See the findings document linked from the source repo.

Configuration secrets: set GROQ_API_KEY in the Space settings.