Fitness App with AI Coach

01The challenge:

Programs went out as PDFs, with no way to see who followed them

FitPro started as two personal-training studios and grew an online program on the side. Remote clients received a four-week PDF, logged workouts in whatever notes app they liked and checked in by email on Sundays. Each coach could keep up with about 40 of them before check-ins started slipping.

The PDF did not change when a client missed sessions or stalled on a lift, so the next week's plan was often wrong until a coach caught it in the Sunday email. Most online clients cancelled in their second month, while the waitlist for a coach kept growing.

02What we built:

An app that writes the next workout from the last one

The React Native app runs on iPhone and Android from one codebase. Clients log sets, reps, weight and how hard each set felt, and a TensorFlow model trained on FitPro's own programming history uses that log, missed days and the client's goal to set the next session's exercises and loads.

Coaches see every recommendation in a Node.js review queue and can approve or change it before it reaches the client. Progress charts and photos sit in the app, and small groups share a feed with weekly challenges. Firebase handles sign-in and push reminders, and MongoDB stores the training history.

03What changed:

4.8 stars, 25,000 downloads and room for more clients per coach

The app holds a 4.8 rating on the App Store and passed 25,000 downloads in its first year, most of them outside FitPro's home market. Each coach now handles about 110 online clients, up from 40, because the model drafts the week and the coach reviews it.

Fewer online clients leave in their second month, since the plan adjusts when they miss a week. We ship an update every month and retrain the model each quarter on fresh logs, and FitPro's coaching team picks what goes into each release.