Partnered with a global manufacturer to modernise quality control across five plants. We combined edge-ready computer vision models with explainable decision tooling so operators could trust automated calls.

Results delivered:

• 27% reduction in false positives after deploying active-learning feedback loops.
• Sub-200ms inference budgets achieved through model distillation and TensorRT optimisation.
• Operations playbook that trained plant managers to monitor drift and recalibrate in under an hour.