CarePathAI Patient Intake Automation (Demo)

This repo includes a LiveKit voice agent plus a LeanMCP backend and Next.js UI to demo a patient intake automation pipeline.

Demo disclaimer

Demo only. Do not enter real PHI.
This project is not HIPAA compliant and does not provide medical advice.

Monorepo layout

• services/agent: LiveKit Agents Python app (src/agent.py)
• services/mcp: LeanMCP backend (TypeScript) + REST + WebSocket + SQLite
• services/web: Next.js UI

Environment configuration

Create a .env at the repo root with:

LIVEKIT_URL=...
LIVEKIT_API_KEY=...
LIVEKIT_API_SECRET=...

LLM_PROVIDER=openai
OPENAI_API_KEY=...

MCP_PORT=4000
DATABASE_PATH=./data/intake.db

MCP_BACKEND_URL=http://localhost:4000
NEXT_PUBLIC_BACKEND_URL=http://localhost:4000

Note: Creating .env.example is blocked in this workspace due to ignore rules. Use the snippet above as a template.

Local dev (manual)

1) LeanMCP backend

cd services/mcp
npm install
npm run dev

2) Web UI

cd services/web
npm install
npm run dev

Open http://localhost:3000.

3) LiveKit agent

cd services/agent
uv sync
MCP_BACKEND_URL=http://localhost:4000 uv run src/agent.py start

Docker Compose

docker compose up --build

The web UI runs at http://localhost:3000 and the backend at http://localhost:4000.

End-to-end smoke test

1. Open the web UI and click “Start patient intake”.
2. Allow microphone access, speak naturally.
3. Confirm transcript is updating and structured fields fill in.
4. Click “Finalize intake” to generate summary and export.
5. Open /staff/[session_id] to view the clinician summary and export link.