Challenge: Manual visual inspection on the line was slow, inconsistent and missed subtle defects.
What we built: A camera-fed vision pipeline that flags defects in real time, with an annotation loop so operators can correct and continuously improve the model. A dashboard tracks defect rates by shift and station.
Impact: Defect escape rate fell significantly while throughput rose, and the feedback loop kept accuracy climbing.
Stack: Python, PyTorch, ONNX runtime, edge inference, Laravel reporting.
Vision QA System
A computer-vision pipeline for automated manufacturing defect detection.