An AI-powered coaching app that watches your form through the phone camera, corrects it in the moment, and rebuilds your plan around what your body actually did this week.
Home workouts fail for a boring reason: nobody is watching. Users pick the wrong weight, break form on rep seven, and either hurt themselves or quietly stop opening the app. The founders had a coaching methodology that worked in person and needed it to survive the trip into a phone.
Two hard constraints shaped everything. Feedback had to arrive fast enough to fix a rep while it was happening, and video of people exercising in their bedrooms could not be shipped off to a server.
Both constraints pointed the same direction: run the model on the device. We prototyped pose estimation on mid-range hardware in week one, because if it could not hit frame rate on a three-year-old Android, the product did not exist.
A cross-platform app that coaches through the camera without a single frame of video leaving the phone. Users get live cues, a session summary showing where form drifted, and a plan that changes next week based on what actually happened this week.
The pose pipeline held frame rate on the low-end test devices, which is what made the rest of the product possible. Privacy stopped being a caveat and became a selling point.
The founders were in every Friday demo, on their own phones, doing the exercises. That feedback loop caught things no test suite would have — cues that fired too late, a voice that talked over the user's music. See our four-step process for how a build like this is run.