Quality inspection

Industrial Anomaly Detection

A two-stage visual inspection system across 15 industrial product categories.

defect detection accuracy: 96%defect detection accuracy
false positives: -15%false positives

The problem

On a production line, every false alarm stops a good part and costs time, while a missed defect costs far more.

What I built

1
Designed a two-stage system on the MVTec dataset: product classification first, then defect detection per category.
2
Served the model through a FastAPI endpoint.

Outcome

92% classification accuracy over 15 product categories and 96% defect detection accuracy.
False positives reduced by 15%.

Have a similar problem?

Tell me about it. I reply within 48 hours with a first read of the problem and an honest opinion on fit.