99.2% recall across 38 production lines, deployed on-edge with sub-200ms inference.
11 weeks · 2025 · Squad of 5
99.2%
Defect recall
<200ms
On-edge inference latency
38
Production lines covered
11 wks
Time to production
Manual visual inspection on 38 production lines was missing ~8% of defects, leading to costly downstream recalls. Installing cloud-connected cameras was not viable due to factory network constraints.
We trained a custom YOLO-based defect detection model on client-provided image datasets, quantized to ONNX for edge deployment on NVIDIA Jetson hardware, and integrated with the existing MES system for real-time flagging.
Want results like these?
Tell us the problem. We'll tell you if we can help.