Developer Builds Privacy-Safe AI System for Circular Manufacturing Supply Chains

A software developer built a machine learning system combining federated learning, active learning, and quantum-inspired optimization to address data privacy and labeling challenges in circular manufacturing supply chains. The project was prompted by a real pilot involving thermal imaging data from refurbished electric vehicle batteries, where OEM partners refused to share raw sensor data due to strict legal agreements. Traditional supervised models trained on the fragmented, sparse data achieved only a 68% F1-score on defect classification, highlighting the need for a new approach. The solution allows a shared predictive maintenance model to learn across distributed facilities without exposing raw data, while also reducing the burden of manual expert annotation. Energy constraints at edge deployment sites — many running on solar or battery power — were an additional design consideration driving the low-power architecture.