EdgeTutor: a 200MB UX model for rapid A/B testing on-device

Tech · 4 min read

EdgeTutor is a tiny generative model that produces layout and copy alternatives tailored to short-run A/B experiments. At 200MB, it sits comfortably on modern phones and mid-tier laptops, enabling designers and product managers to generate variations without leaving the device. The model outputs JSON variants ready for fast swap-in during user tests.

Because EdgeTutor runs on-device, it supports privacy-sensitive research where captured data must remain local. Instrumentation hooks let researchers collect anonymized metrics and user feedback without transmitting raw screens. This is particularly valuable in regulated industries or on trials with sensitive content.

Integrations with analytics SDKs make it straightforward to compare predicted engagement against real results, and a small dashboard app helps teams manage variant rollouts. Early users appreciate the speed and the ability to prototype many more permutations before committing to server-side experiments.