Pl(AI) Lab

Led technical development of a research prototype addressing a gap in cancer rehabilitation: post-surgical patients who are chair-bound lack access to personalized exercise guidance between physical therapy sessions. Traditional PT is expensive and infrequent; existing fitness apps assume mobility and tech literacy that older adults often lack.

Built an AI-powered exercise coach using computer vision (pose estimation) and voice interaction, designed specifically for adults 65+ recovering from cancer surgery. The system guides patients through rehabilitation exercises (protocols developed in consultation with MSKCC physical therapists), adapts workouts in real-time based on patient feedback, and provides form guidance to prevent injury—all without requiring patients to touch a screen or navigate menus.

Identified that standard phone-based interfaces were unusable for chair-bound patients (camera must be 6+ feet away for full-body view, making screens unreadable). Advocated for lab to acquire Viture AR glasses, projecting the interface directly in patients' field of view regardless of body orientation during exercises. This eliminated the need for patients to face the phone directly, critical for lateral movements.

Developed Swift iOS app with real-time pose estimation to track joint positions and detect exercise completion. Built Python backend integrating OpenAI's multimodal model for voice-driven interaction—patients could request breaks, adjust rep counts, or stop exercises through natural conversation. Implemented form checking using 2D joint coordinates to flag asymmetry and range-of-motion issues (limitation: 2D pose estimation provides general guidance, not clinical-grade biomechanical analysis). Gamified exercises with on-screen targets patients "hit" with body parts, reducing cognitive load for exercise execution.

Conducted user testing with 20+ post-surgical cancer patients. 95% expressed interest in continued use if developed into a production app, validating core concept and accessibility approach. Presented findings to 3,000+ attendees at Stanford Engineering Centennial.

Demonstrated feasibility of AI-assisted rehabilitation for underserved populations (older adults, post-surgical patients) using commercially available hardware. Showed that voice interaction + AR can reduce traditional barriers to tech adoption in clinical contexts.