WOX: an AI personal trainer for a staff-free studio

A personal training studio in Nyíregyháza wanted every session to be fully automated and personal. Clients trained in front of the studio's monitors and headphones, and the trainer picked the right workout for each of them by hand, from more than 300 different styles.

Both the choice of workout and the coaching during the session depended on one person. The task was to hand them to an AI system that talks with clients in real time.

What we built

We built the system end to end: the voice interface, a proprietary model, the workout selection logic, client profiles, admin screens and a self-service platform. We also integrated the studio's hardware. The work ran from July to September 2025.

After the first week the client rethought how the program should work. We took the change on board and still met every deadline, with daily progress updates throughout.

How a WOX session works

• Self-service: clients use the studio without staff on site.

• Profile: each client's parameters and goals are stored in their profile.

• Workout choice: a decision-tree logic layer picks the most suitable workout from the studio's library.

• Voice coaching: speech-to-text and text-to-speech let the system talk with the client in real time, and it motivates them by name.

• End of session: feedback and information on how the workout went.

• Admin: screens for managing clients and the system.

Two decisions that shaped the system

Rules choose the workout, the voice does the coaching

Workout selection runs on a decision-tree logic layer, and the conversation runs on the voice side. In our view the choice of workout should follow rules the studio can trace, while the coaching should sound natural. Keeping the two apart lets each do its own job.

A natural voice through ElevenLabs

The voice output comes from ElevenLabs, connected through our own connectors. Since the voice is the only coach in the room, we wanted it to sound like a person.