Problem
The workflow needed rapid classification support in environments with unstable connectivity.
Constraints
- Inference latency needed to remain acceptable on-device
- Model behavior had to tolerate imperfect capture angles
- UI had to keep scan interactions simple for repeat use
Approach
- Built lightweight model iteration and evaluation loop
- Implemented mobile capture and inference triggers
- Tested UI flow for repeated scan-and-review usage
Results
- Delivered a demo-ready scan workflow with local inference
- Identified clear next steps for dataset and model improvement
- Documented architecture tradeoffs for deployment planning
Tech stack
- Computer Vision
- Android
- TensorFlow Lite
- Mobile UX