CaliForma

About This Project

CaliForma is an iPhone app that counts reps and grades form live from the camera. Pose detection runs entirely on the device: joint angles are smoothed, judged against per-exercise rules with hysteresis, and turned into per-rep scores, faults and spoken cues while you train. It also times static holds like planche leans and L-sits, tracks progress through skill ladders from push-ups to planche, and puts a generated or coach-assigned plan on your calendar.

I'm the sole developer across three products in one repo: the iOS app, a macOS app for coaches, and a TypeScript backend. Coaches review client videos with the same rep engine, run a structured intake interview, and generate a periodised plan from it using tables adapted from a personal-trainer course, then edit and assign it; clients see the plan and message their coach from the phone. The rep engine lives in its own Swift package with synthetic-pose tests and a fixture corpus of real clips, and a parity suite keeps the live and offline paths agreeing.

The hard part has been making the live rep counter honest about what the camera actually saw: reps too fast to resolve, bottoms hidden by occlusion, and joints that disappear for a few frames are counted with depth marked as unmeasured rather than guessed. Videos never leave the phone unless a client shares them with their coach; the backend is self-hosted in the EU with GDPR export and deletion built in.