Computer Vision Engineer — PCB Defect Detection

PCB manufacturer. 4 production lines. Manual optical inspection by humans is slow, inconsistent and can't keep up with volume. Defect types we need to detect: missing components, solder bridges, lifted leads, wrong polarity, pad damage. All via 2D top-down camera images. We have 80k labelled images from the past 18 months. We need a model that runs on our existing NVIDIA hardware at line speed, outputs bounding boxes with defect type and confidence, and has a false negative rate under 0.5%. Deliverables: trained model, inference API, integration guide for our production system. Internal team handles deployment.