Jarvis

Jarvis is a voice-first AI assistant with multiple conversation modes such as pote, coach, psychologue, prof, medecin, and traducteur.

Architecture

• src/ contains the Vite + React frontend.
• Jarvis_Backend/ contains the Express + WebSocket backend.
• The frontend records the user's voice and sends audio to the backend.
• The backend sends that audio to Gemini for understanding and response generation.
• The backend then sends the Gemini reply to Gemini TTS and returns spoken audio to the frontend.
• The UI does not display the transcription or model text reply.

Local setup

Backend

1. Create Jarvis_Backend/.env from Jarvis_Backend/.env.example.
2. Add your Gemini API key as GEMINI_API_KEY.
3. Start the backend:

cd Jarvis_Backend
npm start

Frontend

1. Start the frontend:

npm run dev

2. The frontend expects the backend WebSocket at ws://localhost:3000 by default.
3. For deployment, set VITE_BACKEND_WS_URL to your production WebSocket URL.

Notes

• Conversation memory is kept in backend memory per session and trimmed to the last 20 messages.
• psychologue and medecin modes include guardrails so Jarvis stays supportive without pretending to be a licensed professional.
• The current implementation is optimized for a strong local demo and can later be upgraded to Gemini Live API for lower-latency streaming speech.