← Selected work

Computer Vision · Founding team · 2022

HomeSweatHome

An AI virtual trainer that reads posture and movement from a laptop webcam and coaches workouts in real time, entirely on-device.

Python Computer Vision Pose estimation On-device

Problem

People training at home have no one watching their form, and cloud round-trips are too slow to give useful feedback mid-exercise.

Approach

As part of the founding team I designed the architecture and built the perception core: real-time posture and movement analysis and an on-device action recognition model, both light enough to run smoothly on consumer hardware.

Result

Reached a working MVP presented at the demo day of Bocconi University's IdeaBoosterLab accelerator.