Medical AI Chat Prototype

A medical AI chat application that reads each patient's structured profile from MongoDB and uses it as context to answer symptom queries, suggest diagnoses, and recommend medicines.

Stack: React + Vite frontend, FastAPI + LangGraph backend, MongoDB for persistence — all orchestrated via Docker Compose.

Disclaimer: This is an educational prototype. It does not provide real medical advice. Always consult a qualified healthcare professional.

For AI agents: Start with CLAUDE.md for navigation. For detailed reference docs, see docs/.

---

Architecture

┌──────────────────────────────────────────────┐
│  Browser — React + Vite (port 5173)          │
│  ChatPage → MessageList + ChatInput          │
└───────────────────┬──────────────────────────┘
                    │ POST /api/chat/message
                    ▼
┌──────────────────────────────────────────────┐
│  FastAPI (port 8000)                         │
│  api/chat.py → service/chat.py → ai/graph.py│
│       │                │                     │
│       ▼                ▼                     │
│    MongoDB (patient profiles + chat state)   │
└──────────────────────────────────────────────┘

LangGraph:  START → agent ←→ tools → END
                       │
                       ▼
              search_patient_records

Data flow:

1. User types a message → frontend POST /api/chat/message
2. Backend loads/creates chat session, invokes LangGraph agent
3. Agent reads the patient profile from MongoDB via search_patient_records tool
4. Agent synthesizes response → persisted to MongoDB → returned to frontend
5. Frontend renders the AI response with markdown formatting

---

Quick Start

Prerequisites

• Docker and Docker Compose
• A Google Gemini API key or OpenAI API key

Setup

# 1. Clone / navigate to project
cd prescribe-v2

# 2. Configure environment
cp .env.example .env
# Edit .env and add your API key:
#   GOOGLE_API_KEY=your-key-here
#   (or set LLM_PROVIDER=openai and OPENAI_API_KEY=your-key-here)

# 3. Start everything
docker compose up --build -d

# 4. Open in browser
open http://localhost:5173

Verify

docker compose ps                          # 3 services running
curl http://localhost:8000/health           # {"status":"ok"}
curl http://localhost:8000/api/chat/patients  # List of patients

---

Environment Variables

Variable	Default	Description
LLM_PROVIDER	google	LLM provider: google or openai
GOOGLE_API_KEY	—	Gemini API key (required if provider=google)
OPENAI_API_KEY	—	OpenAI API key (required if provider=openai)
MONGODB_URI	mongodb://mongodb:27017	MongoDB connection string
DATABASE_NAME	medical_chat	MongoDB database name

---

API Endpoints

Method	Path	Description
POST	/api/chat/message	Send a message, get AI response
GET	/api/chat/history/{chat_id}	Get chat history
GET	/api/chat/patients	List available patients
GET	/health	Health check

Example Request

curl -X POST http://localhost:8000/api/chat/message \
  -H "Content-Type: application/json" \
  -d '{"message":"What medications is John on?","patient_id":"john_doe"}'

---

Adding New Patients

New patients are created through the registration API or seed scripts — there are no files to edit. Each patient's profile in the users collection (with role: patient) is the single source of truth for demographics, conditions, medications, allergies, family history, and lifestyle. Uploaded documents live in the documents collection. The search_patient_records tool reads from those collections at request time.

---

Project Structure

prescribe-v2/
├── docker-compose.yml          # Orchestrates 3 services
├── .env.example                # Environment template
├── backend/
│   ├── Dockerfile
│   ├── requirements.txt
│   └── app/
│       ├── main.py             # FastAPI app entry
│       ├── config.py           # Settings (pydantic-settings)
│       ├── api/chat.py         # REST endpoints
│       ├── service/chat.py     # Business logic
│       ├── repository/chat.py  # MongoDB CRUD
│       ├── database/           # Motor connection
│       ├── schemas/chat.py     # Pydantic models
│       └── ai/
│           ├── graph.py        # LangGraph agent
│           ├── tools.py        # DB-backed patient profile tool
│           ├── prompts.py      # System prompt
│           ├── llm.py          # LLM factory
│           └── state.py        # Graph state
├── frontend/
│   ├── Dockerfile
│   ├── package.json
│   └── src/
│       ├── App.tsx
│       ├── api/chat.ts         # API client
│       ├── hooks/useChat.ts    # Chat state hook
│       └── components/
│           ├── ChatPage.tsx    # Main layout
│           ├── MessageList.tsx # Message display
│           ├── ChatMessage.tsx # Single message bubble
│           ├── ChatInput.tsx   # Input bar
│           └── PatientSelector.tsx
└── init-mongo.js               # DB initialization

---

Commands Reference

docker compose up --build -d      # Start everything
docker compose down               # Stop everything
docker compose down -v            # Stop + delete data
docker compose logs -f backend    # View backend logs
docker compose logs -f frontend   # View frontend logs

# Run backend tests
docker compose exec backend python -m pytest tests/ -v

# Local development (without Docker)
cd backend && pip install -r requirements.txt && uvicorn app.main:app --reload
cd frontend && npm install && npm run dev