ManufacturingComputer vision
Task. At a production site, safety violations (no helmet, entering danger zones) were recorded manually and selectively — an inspector physically could not cover all workshops, and reporting took days.
Solution
We deployed video analytics on existing cameras: the model detects PPE violations in real time, instantly notifies the responsible person and generates automatic violation reports. On-premise deployment — video streams never leave the enterprise.
Results
15×faster average violation detection
30×faster report generation
-25%defect rate after extension to quality control
Technologies: OpenCV, Ultralytics YOLO, Python, Docker, on-premise GPU.
Project stages
- Site survey (1 week) — camera layout audit and safety-scenario definition.
- Model training (3–5 weeks) — computer-vision models for PPE and hazard-zone detection.
- Deployment (2–3 weeks) — edge or cloud inference, alert integration with safety officers.
- Monitoring — accuracy tracking and periodic retraining.
Where this applies
Safety-violation detection is in demand on construction sites in Almaty and Astana, industrial plants in Uzbekistan, and mining and manufacturing facilities across Central Asia. See AI for construction.