A full-stack web app that analyzes videos of calisthenics holds using MediaPipe pose detection, geometric scoring, and LLM-generated coaching feedback.

Supported moves: planche, front lever, back lever, handstand

1. Upload a video and select the move
2. Frames are extracted at 2 FPS and run through MediaPipe pose detection
3. The longest stable-pose window is identified
4. Geometric metrics (angles, alignment, lean) are scored against calibration curves
5. An LLM generates coaching feedback from the scores
6. Results are displayed with a skeleton overlay on the best frame and a score trend chart

• Python 3.10+
• Node.js 18+
• ffmpeg (for video trimming)

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

cp .env.example .env
# Edit .env — set PROVIDER and the matching API key

python -m uvicorn main:app --reload
# Runs on http://localhost:8000

cd frontend
npm install
npm run dev
# Runs on http://localhost:5173

cd backend
source venv/bin/activate
pytest -v

1. Create backend/analyzer/moves/<move_name>.py with a METRICS list (each entry: name, label, weight, compute_fn, curve)
2. Register it in backend/analyzer/moves/__init__.py
3. Add the move to MoveType in frontend/src/api/types.ts

• Backend: FastAPI, SQLAlchemy, MediaPipe, OpenCV, NumPy
• Frontend: React, TypeScript, Recharts
• LLM: Anthropic Claude / OpenAI / Google Gemini (configurable)