Anatomy_View

Boilerplate monorepo that pairs a Vite-powered React frontend with a minimal Flask backend.

Project Structure

.
├── backend
│   ├── requirements.txt
│   └── server.py
├── frontend
│   ├── .gitignore
│   ├── index.html
│   ├── package.json
│   ├── src
│   │   ├── App.jsx
│   │   └── main.jsx
│   └── vite.config.js
└── README.md

• The frontend directory contains the Vite React application configured with routing and Three.js dependencies.
• The backend directory hosts the Flask API server.

Running the Applications

1. Install frontend dependencies

   cd frontend
   npm install
   npm run dev

2. Install backend dependencies

   cd backend
   python -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt
   python server.py

With both servers running, API requests from the frontend to /api/* will be proxied to the Flask backend.

Gemini + Dedalus Labs Integration

The backend exposes a /api/chat endpoint that streams conversation history to Google Gemini so the model can issue tool calls. The React frontend executes those tools via the Dedalus Labs controller to rotate, highlight and annotate the Three.js heart scene.

1. Create a Google AI Studio API key and make it available to the backend:

   export GEMINI_API_KEY="your_api_key"
   # Optional: choose a different model variant
   export GEMINI_MODEL="models/gemini-1.5-pro-latest"

2. When developing locally, point the Vite dev server to the Flask backend:

   # from the frontend directory
   export VITE_API_BASE_URL="http://localhost:5000"
   npm run dev

3. Provide Dedalus Labs credentials so semantic medical questions can be answered with the Dedalus SDK. Set the credentials required by the SDK (for example DEDALUS_API_KEY) and optionally choose a model variant:

   export DEDALUS_ENABLED=true
   export DEDALUS_MODEL="openai/gpt-5"

4. Start the Flask server with the same environment variables so the /api/chat route can contact Gemini and Dedalus Labs:

   cd backend
   flask --app server run

If the Gemini API key is missing, the assistant will return a friendly error message instead of contacting Gemini. When Dedalus Labs is enabled, the backend will automatically route purely semantic medical questions to Dedalus while continuing to use Gemini for anatomy-specific instructions and tool calls. Successful responses include tool call metadata that the frontend uses to animate the model, along with a replySource field indicating whether Gemini or Dedalus generated the textual explanation.