curtburk/consent-agent-reachy

Medical Triage Voice Agent for Reachy Mini

An always-listening voice agent that runs entirely on local hardware (HP ZGX Nano) — no cloud APIs needed. Designed for medical office triage assistance.

What it does

1. 1.Listens continuously via the robot's built-in microphone
2. 2.Detects speech using Silero VAD (voice activity detection)
3. 3.Transcribes speech with faster-whisper (running locally on CPU)
4. 4.Generates a medical triage response with Llama-3.1-8B-UltraMedical via vLLM (running locally on GPU)
5. 5.Synthesizes speech with Piper TTS (running locally)
6. 6.Plays the response through the robot's built-in speaker

Requirements

• —HP ZGX Nano (or any NVIDIA GPU with 24GB+ VRAM) running the ZGX AI API container
• —Reachy Mini (Wireless) with SDK installed

Architecture

Reachy Mini (mic/speaker)
    │
    │  HTTP (network)
    │
ZGX Nano Docker Container
    ├── faster-whisper STT (with Silero VAD)
    ├── vLLM + Llama-3.1-8B-UltraMedical
    └── Piper TTS