OntarioDoctor (PoC)

A PoC of a medical symptom-checker chat application for Ontario residents, powered by RAG (Retrieval-Augmented Generation) with MedGemma-4b-it LLM.

Showcase: https://www.loom.com/share/8458c35ef3dd4c009d4a55428a1f3c98

🎯 Features

• 💬 Conversational Interface: Natural language symptom checking
• 📚 Grounded Responses: All answers backed by medical sources with citations
• 🚨 Red-Flag Detection: Automatic triage for emergency situations
• 🍁 Ontario-Focused: Localized resources (Telehealth Ontario, walk-in clinics, ER guidance)
• 🔒 Privacy-First: Session-based only, no data persistence or user accounts
• ⚡ Hybrid RAG: Vector search + BM25 with cross-encoder reranking
• 🎨 Modern UI: Vue 3 + TailwindCSS + shadcn-vue

🏗️ Architecture

┌─────────────┐
│  Frontend   │  Vue 3 + Pinia + TailwindCSS
│  (Port 5173)│
└──────┬──────┘
       │
       v
┌─────────────┐
│  Gateway    │  FastAPI (Port 8080)
│  /chat      │  Public API
│  /ingest    │
└──────┬──────┘
       │
       v
┌─────────────┐
│Orchestrator │  LangGraph (Port 8002)
│  Workflow   │  symptom_nlu → guardrails → retrieve → generate
└──────┬──────┘
       │
       ├──────> RAG Service (Port 8001)
       │        Qdrant + BM25 + Reranker
       │
       └──────> Ollama (Port 11434)
                MedGemma-4B-IT:Q6

🚀 Quick Start

Prerequisites

• Docker & Docker Compose
• NVIDIA GPU with CUDA support (for Ollama)
• Python 3.11+ (for local development)
• Node 20+ & pnpm (for frontend development)

1. Clone Repository

git clone https://github.com/pmbstyle/OntarioDoctor.git
cd OntarioDoctor

2. Start All Services

make up

This will start:

• ✅ Qdrant (vector database)
• ✅ Ollama (LLM inference)
• ✅ RAG Service
• ✅ Orchestrator
• ✅ API Gateway
• ✅ Frontend

3. Ingest Sample Data

make ingest-sample

4. Access the App

Open your browser to http://localhost:5173

5. Run Smoke Tests

make smoke-test

📦 Project Structure

OntarioDoctor/
├── backend/
│   ├── gateway/          # FastAPI: /chat, /ingest, /health
│   ├── orchestrator/     # LangGraph workflow
│   ├── rag/             # Hybrid retrieval service
│   └── shared/          # Common models & utilities
├── frontend/            # Vue 3 chat interface
├── docs/                # Documentation
├── scripts/             # Sample data & tests
├── docker-compose.yml   # Service orchestration
├── Makefile            # Helper commands
└── README.md

🛠️ Development

Backend (Local)

1. Setup virtual environment:

   make venv
   cd backend && source venv/bin/activate

2. Run services individually:

   # Terminal 1: RAG
   python -m uvicorn backend.rag.main:app --reload --port 8001
   
   # Terminal 2: Orchestrator
   python -m uvicorn backend.orchestrator.main:app --reload --port 8002
   
   # Terminal 3: Gateway
   python -m uvicorn backend.gateway.main:app --reload --port 8000

3. Start Ollama (if not using Docker):

   docker run -d --gpus=all -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
   docker exec -it ollama ollama pull amsaravi/medgemma-4b-it:q6

Frontend (Local)

cd frontend
pnpm install
pnpm dev

Access at http://localhost:5173

Environment Variables

Copy .env.example to .env and configure:

cp .env.example .env

Key variables:

• CUDA_VISIBLE_DEVICES: GPU device ID (default: 0)
• VITE_API_URL: API Gateway URL for frontend
• GATEWAY_CORS_ORIGINS: Allowed CORS origins
• *_LOG_LEVEL: Logging levels (DEBUG, INFO, WARNING, ERROR)