AI-supported exercise coaching for older adults with mild cognitive impairment

Research Assistant, Ross and Carol Nese College of Nursing, Penn State · PI: Dr. Sangwoo Ahn 

Im implementing clinical protocol by Dr. Ahn as working research prototype. It runs as Python session server connected over WebSockets to a browser front end with MediaPipe pose estimation and scikit-learn classifier scoring movement against reference recordings, everything processed on the local machine and an automated test suite covering session flow, provisional rule-based safety logic and complete-session replay from stored movement features.

Re-scoring held-out clips against different reference sets showed that reference selection changes movement scores and re-scoring mirrored copies of the clips changed the scores as well, so choosing the reference set is part of the study design.

For the planned pilot, the participant uses a tablet for the screen and camera while a laptop runs the movement processing and the researcher console. I have also built a separate version that runs the full session on an Android tablet. The protocol keeps the researcher present throughout and participant evaluation is still ahead.