As part of my Master's coursework, I worked on a project in a team of two where we leverage on-device efficient model training techniques for an athletic use case. We collect live images from members of MIT’s rowing team, and use TinyTraining techniques to classify these images as ‘good pose’ vs ‘bad pose’. This application would be useful for coaches to assess athletes. We implemented a pose identifier running on an Android application that evaluates athletes’ physical form, to be deployed during the Para High Performance Group competitions. We also trained a Temporal Shift Module (TSM) model on images of rowers with different proper and incorrect poses, and evaluated this model on live images of rowers.