Pick a physiotherapy exercise from the University of Melbourne CHESM video library (knee and hip osteoarthritis), open camera + chat, and get live guidance from an AI coach (Gemini Live). No exercise detection — you choose the exercise; the coach guides you. Targeted at elderly people with knee and/or hip OA.
→ Features and requirements: docs/FEATURES.md.
(Start virtual environment)
# 1. Backend
pip install -r requirements-api.txt
# Set GEMINI_API_KEY in a .env file (copy .env.example)
uvicorn api.main:app --reload --port 8000
# Or: npm run api (with venv activated), or scripts\run_api.bat
# 2. Frontend (other terminal)
npm i && npm run devOpen https://localhost:5173 → pick an exercise → Start → camera + coach session.
Exercises are loaded from data/exercises.json, which contains the University of Melbourne CHESM knee and hip osteoarthritis set (e.g. seated knee extension, sit to stand, step ups, partial wall squats). “View in video library” in the app links to that page.
| Path | Purpose |
|---|---|
api/ |
FastAPI: /api/exercises, /api/exercises/{slug}, /api/coach (WS), /api/health |
src/ |
Frontend: exercise picker + camera + coach chat (multimodal) |
data/ |
exercises.json (CHESM exercise list and coach context) |