In the manufacturing industry, particularly within Top-Level Assembly (TLA), ensuring consistent quality remains a challenge. TLA involves a single operator assembling complex, multi-layer products ranging from mechanical to electronic components, where human error caused by fatigue or cognitive overload can result in costly rework, material wastage, and reduced operational  efficiency. Traditional inspection techniques often fail to detect defects early, while existing AI-based CV systems lack the flexibility needed to handle the complexity and variability of TLA.

To address this problem, we developed VOTÉ (Vision on the Edge), a cost-effective, AI-powered edge computer vision (CV) inspection system that enables real-time, step-by-step assembly guidance. The solution empowers quality control (QC) engineers to train and deploy custom CV models through a no-code desktop application, while operators receive guided inspection feedback via an  intuitive edge-based interface. The system integrates Poka-Yoke principles to catch and prevent defects.

We designed and iterated on a complete end-to-end solution comprising a compact mechanical design for a stable, non-intrusive,  and customisable setup, alongside user-friendly interfaces for both model training and live inference. Through extensive user testing with our industry partners, VOTÉ has evolved into a robust and responsive system that helps improve quality control measures in the TLA industry.

Key insights that we have gained in this project include the importance of model interpretability, balanced datasets, and a no-code UI. Ultimately, VOTÉ has demonstrated its ability to reduce misassemblies and false positives, while providing actionable feedback to both engineers and operators. By bridging the gap between advanced CV technologies and non-technical users, VOTÉ paves the way for scalable, accessible AI-based CV inspection across small and large-scale manufacturing environments.