AI-powered break optimization platform for factory workers. Detects fatigue in real time via computer vision and recommends optimal break schedules to reduce errors and workplace accidents.

---

Architecture

Camera (ml/camera.py)
       │  POST /update
       ▼
FastAPI Backend (back/server.py)
       ├── GET  /telemetry
       ├── POST /update
       └── WS   /ws
                │
                ▼
Next.js Frontend (frontend/)
       ├── /              → Shift overview dashboard
       ├── /analysis      → Per-station fatigue analysis
       └── /current-shift → Live fatigue signal chart

---

Computer Vision (ml/camera.py)

Uses MediaPipe Face Landmarker at 30+ FPS to detect two fatigue signals:

• Eye closure — both eyes must be closed for 8 consecutive frames (EYE_CLOSED_THRESHOLD = 0.012)
• Yawning — mouth open for 20 consecutive frames (MOUTH_OPEN_THRESHOLD = 0.04)

On a new fatigue event, the camera POSTs to the backend:

{ "status": "UNFOCUSED (EYES CLOSED & YAWNING)", "fatigue_level": 90.0, "timestamp": "14:35:22" }

---

Backend (back/server.py)

FastAPI w/ three endpoints:

Method	Route	Description
GET	/telemetry	Current worker state
POST	/update	Receive fatigue event from camera; broadcast via WebSocket
WS	/ws	Real-time WebSocket channel for the frontend

Only UNFOCUSED events update the internal state. Each WebSocket message has the shape:

{ "type": "webhook_update", "received_at": 1710000000.123, "accepted": true, "payload": { ... } }

---

Frontend (frontend/)

Next.js 16, React 19, TypeScript, Tailwind CSS, Plotly.js.

• / — Overview of all active shifts and stations. Expand any station to see its fatigue curve and AI-optimized break schedule.
• /analysis — Select a shift to view an interactive floor map and per-station AI diagnostics (risk summary, detection logic, recommendation, expected impact).
• /current-shift — Live Plotly line chart updated every second via WebSocket. Signal weights: yawn = 1, eyes closed = 2, both = 3.