← back to notebook
TrustMed-AI screenshot
2024 · MEDICAL AI

TrustMed-AI

TrustMed-AI is a production multi-agent RAG system for clinical question answering. Parallelized retrieval agents pull from verified medical sources, a FAISS cross-encoder reranks results to cut hallucinations, and a Next.js frontend with a Three.js avatar delivers the answer back as natural speech.

Stack
PythonLangChainFAISSGPT-4oFastAPIChromaDBThree.jsDockerAWSNext.jsSnowflakeTableau
Highlights
  • Multi-agent RAG with parallelized retrieval lifted precision 40% over a single-agent baseline while cutting response latency from 12s to 4s
  • FAISS cross-encoder reranking on a privacy-preserving pipeline cut hallucinations 65%, reaching 91% accuracy on a MedQA-style benchmark across 5 medical domains
  • 100-prompt evaluation suite logged to Snowflake and visualized in Tableau tracks precision and faithfulness over time