TrustMed AI

TrustMed AI answers medical questions about Diabetes and Cardiovascular Disease by retrieving context from two source types: authoritative clinical articles (Mayo Clinic, CDC, MedlinePlus) and real patient discussions scraped from Reddit. A hybrid retrieval pipeline (k-NN + BM25) ranks passages before passing them to Llama 3 via AWS Bedrock for answer generation.

• Scraped 201 medical articles and 1,525 Reddit threads using BeautifulSoup
• Chunked and embedded documents into Amazon OpenSearch with hybrid k-NN + BM25 indexing
• Built a LangChain retrieval chain with AWS Bedrock (Llama 3) for generation
• Evaluated with RAGAS metrics: context relevance, answer relevance, and groundedness
• Served through a Chainlit chat interface for interactive Q&A

• Hybrid retrieval (semantic + keyword) to handle both clinical terminology and colloquial patient language
• Dual-source design so answers reflect both medical authority and lived patient experience
• OpenSearch over a pure vector DB for its native BM25 support alongside k-NN
• RAGAS evaluation framework to quantify retrieval and generation quality