PostureBuddy (MVP)

Lightweight AI-powered posture coach that works with your webcam.

Tech

• Frontend: React (Vite)
• Backend: FastAPI (Python)
• ML: MediaPipe Pose + heuristic classifier (no images stored)
• Storage: SQLite (aggregate posture events only with explicit consent)

Features (MVP)

• Real-time posture detection: good vs slouched
• Visual/audio nudges with streak detection
• Daily/weekly summary (if consented)
• Safe failure modes and clear messages for edge cases

Privacy

• Frames are processed in-memory only; never stored
• Aggregated posture event labels are stored only when the user explicitly enables consent in the UI

---

Local Development

1) Backend

Requirements: Python 3.10+

cd /workspace/backend
python3 -m venv venv && /workspace/backend/venv/bin/pip install --upgrade pip
/workspace/backend/venv/bin/pip install -r requirements.txt

# Run API
/workspace/backend/venv/bin/uvicorn app:app --host 0.0.0.0 --port 8000 --reload

API docs: http://localhost:8000/docs

2) Frontend

Requirements: Node 18+

cd /workspace/frontend
npm install
npm run dev

App: http://localhost:5173

---

API Overview

• POST /api/classify (multipart/form-data with field image): returns posture classification

  • Response: { label: "good" | "slouched" | "no_person" | "uncertain", confidence: number }
  • Query params: store=true|false to store aggregate label (requires user consent)

• GET /api/summary?period=daily|weekly: percent upright for the selected period

  • Response: { uprightRatio: number (0..1), totalEvents: number }

• GET /api/consent: { consent: boolean }

• POST /api/consent with JSON { consent: boolean }

---

Notes

• The classifier uses MediaPipe Pose landmarks and simple geometric heuristics around head/shoulder/hip alignment. It focuses on sitting contexts (remote workers, students, gamers).
• The system fails safe: if confidence is low or person not detected, it returns uncertain/no_person and does not nudge.