---
title: Med
emoji: 🏆
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.20.1
app_file: app.py
pinned: false
short_description: Medical field with next‑gen technology
---

# AI-Powered Medical Knowledge Graph Assistant

This project uses:
- **BioGPT-Large-PubMedQA** for specialized biomedical question-answering
- **PubMed API** + **Chroma** for retrieval-augmented generation
- **Pyvis** to visualize relevant docs and medical terms
- **Gradio** + **FastAPI** for an interactive UI and robust routing

## Setup

1. **Install dependencies**:

   ```bash
   pip install -r requirements.txt
Set your PubMed API key (optional, but recommended for higher rate limits):

bash
Copy
export PUBMED_API_KEY="YOUR_NCBI_API_KEY"
Run locally:

bash
Copy
python app.py
Then open http://localhost:7860 in your browser.

Hugging Face Spaces:

Select FastAPI or Gradio + FastAPI as the SDK.
Push this repo to your Space. The app object in app.py should be auto-detected.
Wait for the build, then test it out!
How It Works
User enters a query (e.g., "Fever and cough treatment").
Retrieval: The app fetches PubMed abstracts via E-utilities, indexes them in Chroma, then pulls the top 3 relevant docs.
BioGPT QA: The retrieved docs are appended as context to the LLM prompt. BioGPT then generates a specialized medical answer.
Graph Visualization: The docs and extracted key terms are turned into a Pyvis network, served by a separate /graph endpoint.
Display: Gradio shows the textual answer, plus an <iframe> that loads the interactive graph from /graph?query=....
Notes
For production or large usage, consider more robust caching/persistence for Chroma data.
Replace the naive key-term extraction with a real medical NER for better graph detail.
Explore domain-specific transformers for improved embedding and QA performance.
yaml
Copy

---

# How This Fixes the Graph Issue

- **No Inline Scripts**: The Pyvis HTML is served from `/graph?query=...` as a standalone page, so the JavaScript runs without being sanitized or stripped by Gradio.  
- **Gradio + FastAPI**: We embed that route in an `<iframe>`, so you get a live, interactive graph.  
- **BioGPT + PubMed**: You still benefit from domain‑specific QA plus real medical data retrieval from PubMed.  

Deploy or run this project, and you should see a **fully functional** AI medical assistant with an interactive Pyvis graph at the bottom of the Gradio interface!