2026
Industrial Anomaly DetectionFeatured
A model that finds defects on factory parts after seeing only good ones. It is PatchCore-style and trained on MVTec AD. Training and evaluation run as a Prefect pipeline, and the model is live as a FastAPI service on Google Cloud Run, where you can upload an image and try it.
- Image AUROC 0.977 and pixel AUROC 0.977 across all 15 categories
- Every run saves its source commit, dataset checksum and working-tree state, so each result can be traced back to the exact code
- Packaged with Docker, tested in CI, and checked for data drift by a scheduled job on Cloud Run
- PyTorch
- Prefect
- FastAPI
- Docker
- Cloud Run
- GitHub Actions
