Scanner AI
Machine-learning visual quality inspection for assembly lines.
The challenge
In-line visual inspection relies on the human eye: it is inconsistent, gets tired and leaves no evidence of why a part passed or failed.
The solution
An inspection station captures the image, runs vision models on the same server and returns a pass/fail verdict in seconds, with history, analytics and a review workflow.
What I built
- .NET 9 API with PostgreSQL and C# inference on ONNX Runtime — no Python needed in production.
- React web app: KPI analytics, a full-screen inspection station with barcode scanner support, part catalogs and assembly workflows.
- Training module: dataset capture, box and polygon annotation, auto-suggestions, and defect, parts and pass-inspector models.
- Expo Android app that turns a phone into an inspection camera kiosk.
Technical details
- Local inference: no cloud, no network latency
- Same camera stream for training and production
- Traceability for every inspected part
- Works with any kind of assembly