Caesar Li L&S Math & Physical Sciences
LLM-Driven Therapeutic Exercise for Outpatient Stroke Rehabilitation
Most stroke rehabilitation occurs at home, where patients perform exercises without direct therapist supervision. However, clinicians lack tools to remotely monitor performance or dynamically adjust treatment. This project investigates the use of large language models (LLMs) for code generation to support therapist-driven, personalized rehabilitation. We developed a prototype system that enables occupational therapists to author custom exercise programs via natural language, which are translated into executable code and delivered through augmented reality (AR) headsets. These programs guide patients step by step, deliver real-time feedback, monitor task completion quality, and generate structured clinical summaries. A proof-of-concept study with 20 therapists demonstrated the system’s usability, clinical relevance, and potential safety advantages over conventional home programs. This work opens new possibilities for AI-powered, therapist-directed care in outpatient rehabilitation.
Message To Sponsor
Anselm MPS Fund, I am deeply grateful for the generous support of my SURF fellowship. This experience allowed me to grow as both a researcher and a person, giving me the confidence to pursue independent inquiry and the skills to contribute meaningfully to my field.