Problem
Manual quality checks required repeated visual comparison and lacked a fast digital workflow.
Constraints
- Needed to run with limited connectivity
- Inference had to stay lightweight for mobile hardware
- Data quality varied across real capture conditions
Approach
- Prepared dataset and labeling workflow for target classes
- Built a compact inference pipeline suitable for mobile deployment
- Designed capture and result flow for quick operator feedback
Results
- Delivered a usable prototype for offline-first quality checks
- Established a repeatable evaluation flow for future model updates
- Provided a base architecture for production hardening
Tech stack
- TensorFlow Lite
- Android
- Python
- Edge ML