Testing a model on different devices is manual and error-prone
Device constraints aren’t obvious from model exports
Rewriting deployment code for multiple platforms wastes days or weeks
I built Refactor AI, an infrastructure tool that:
Analyzes a trained ML model and flags ops that won’t run on the target device
Refactors the model where possible
Generates deployment-ready code for CoreML, ONNX Runtime, ONNX.js, and TFLite
This reduces deployment time from days to minutes, and lets teams run inference natively on-device, saving on cloud GPU costs.
Open to feedback, testing on real models, or ideas for improvement.