We deployed a real-time computer vision system on the production line using custom-trained YOLOv8 models running on NVIDIA Jetson edge devices. The system inspects every product at line speed, classifying 23 defect types with 99.2% accuracy.
We collected and annotated a custom dataset of 50,000+ images across lighting conditions and product variants, then trained and validated the model against ManuTech's own quality standards. Inference runs locally on edge hardware, ensuring zero dependency on internet connectivity.
A web-based quality dashboard shows real-time defect rates, trend analysis, and shift-level reports, replacing manual inspection logs entirely.