README.md

---
title: Multi-Agent Medical Assistant
emoji: 🩺
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
suggested_hardware: t4-small
short_description: Multi-agent clinical assistant with RAG and imaging analysis
---

# Multi-Agent Medical Assistant

Multi-agent clinical assistant: diagnosis, retrieval, reasoning, and imaging
agents behind a FastAPI backend with a React frontend, deployed as a single
Docker Space.

## Architecture on Spaces

Everything runs in one container on port 7860:

- **FastAPI** — single entrypoint, serves both the API and the built React app.
- **React** — built to static files at image-build time, served by FastAPI.
- **Embedded Qdrant** — local path, no separate server. Rebuilt on each boot.
- **SQLite** — file-based, no Postgres server.
- **Imaging models** — loaded lazily on the T4 GPU on first request.
- **LLM agents** — call a hosted API; keys come from Space secrets.

## Configuration

Set these under **Settings → Variables & secrets**:

- `LLM_API_KEY` — your hosted LLM provider key
- `ELEVENLABS_API_KEY` — for voice output (optional)

## Notes

- Free tier disk is ephemeral: the vector store is rebuilt from seed docs on
  every restart. User uploads do not persist across restarts.
- Requires a paid GPU tier (t4-small) for imaging inference. Pause the Space
  when not in use to control cost.