Research projects.

An LLM pipeline that extracts, structures, and semantically maps unstructured medical data across ODT, DOCX, PDF, PPTX, OCR'd images, and text — classifying chunks per query, converting them to structured JSON, and mapping entities to OMOP-compliant fields.

A reinforcement-learning agent (DQN) that learns to play a virtual tennis game built for stroke rehabilitation, anticipating and assisting patients by playing alongside them during therapy.

A hierarchical motor network explaining transfer of skill from the trained to the naive arm under visuomotor rotation, elucidating the computations of left and right parietal areas in unimanual and bimanual movement.

A feedforward bimanual-reaching network into which stroke is introduced, then expanded across movement types to map rehabilitation characteristics to patient characteristics — finding optimal therapy for a given patient.

Low-cost computer-vision tools that evaluate, monitor, and track a subject's motor performance — validated on healthy subjects and several patient classes.

A Unity-based VR system that recognizes patient gestures via computer vision and converts them to game events — deployed in several hospitals in India, with data currently under analysis.