From Rehab Robotics to GraphRAG: Why Context is King
Before I was designing Generative AI architectures, I was building explainable machine learning applications for healthcare.
Specifically, I worked on systems to help Occupational Therapists guide young people with Cerebral Palsy through home-based therapy. The challenge wasn't just "detecting a movement." It was distinguishing between a therapeutic gesture and the "noisy" neurological commands often present in CP, like spasticity or muscle synergies.
To make that work, we couldn't just throw raw data at a black box. We had to build strict calibration procedures to personalize the system to the individual’s physiology. We had to select interpretable features—like movement variability—that gave therapists actual clinical insight rather than just a binary "pass/fail".
I carried this obsession with context and calibration into my recent work with Large Language Models.