MedAsk — Medical RAG Intelligence

MedAsk: Multimodal Medical RAG Assistant

MedAsk is a web app that lets medical professionals ask questions about clinical guideline PDFs, such as the ADA Standards of Care. It answers only from the uploaded documents and cites the page number and recommendation for each answer. It also reads tables, drug dosage flowcharts, and diagrams, using Docling and PyMuPDF to extract them and Google Gemini Vision to describe them. Retrieval is hybrid: FAISS semantic search (all-MiniLM-L6-v2) and BM25 keyword search, merged with Reciprocal Rank Fusion, and it runs in about 20 ms from an in-memory index. The FastAPI backend streams answers token by token to a React, Vite, and TailwindCSS dark-mode chat interface. It supports multiple PDFs, conversation memory, and automatic English/Arabic detection.

Tech stack: Python, FastAPI, Uvicorn, Pydantic, FAISS, Sentence-Transformers, Rank-BM25, Docling, PyMuPDF, Google Gemini API, React 18, Vite, TailwindCSS

I dropped the “zero hallucination” and “100% accuracy” wording from your original text. Those claims invite hard questions from reviewers unless you can back them with an evaluation. If you have numbers, such as citation accuracy on a test question set, add one line with them. A measured result is more convincing than a absolute claim.