The Problem
Medical AI can produce plausible answers without making the evidence behind them visible. This makes it hard for users to know what to trust.
HEALTHCARE · AGENTIC RAG
TrustMed-AI
Source-grounded medical intelligence with transparent retrieval, citations, and voice accessibility.
GitHub · soonLive demo · soon
Stack
AI / RETRIEVAL
ChromaDBFAISS (optional)Cross-Encoder (optional)
BACKEND
PythonFastAPIUvicorn
FRONTEND
Next.jsReactTypeScript
VOICE / VISUALS
ElevenLabsThree.jsReact Three Fiber
What I Built
A full-stack medical QA assistant that routes questions across specialized retrieval agents and returns source-grounded answers with citations and response metadata.
The Difference
Specialized retrieval · reranking · cited answers · explainable routing · voice accessibility · repeatable evaluation.
Numbers That Matter
9K+medical chunks indexed
3specialized retrieval agents
100evaluation prompts
5medical domains
Engineering Notes ✎
Retrieval Architecture
Queries are routed across symptoms, diseases, and medicines collections. Agent outputs are aggregated into the final response.
Ranking Strategy
ChromaDB handles vector retrieval, with optional FAISS refinement and cross-encoder reranking to improve evidence selection.
Evaluation First
A 100-prompt benchmark evaluates retrieval precision, faithfulness, keyword coverage, and hallucination/grounding risk across 5 medical domains.