Context & Problem
Standard pain reporting scales are highly subjective, making chronic condition tracking unreliable over the long term. A patient's self-reported "5 out of 10" pain level may not represent the same physiological experience from one year to the next.
Additionally, health tracking systems often ask for input when the user is actively experiencing symptoms (like a migraine), leading to high abandonment rates if the interface causes cognitive load or visual strain.
Architecture & Approach
On the modeling side, I implemented a backend system utilizing Gradient Boosting Decision Trees (GBDT) and Hurdle Models. By separating the problem into two probabilistic stages — whether an event will occur, and if so, how severe it will be — the model mirrors true clinical outcomes.
On the frontend, I engineered an accessible, low-friction UI specifically designed for users experiencing neurological discomfort. The interface relies on tap-based inputs like symptom pill buttons and segmented controls, minimizing typing. I also implemented usage-based smart sorting to dynamically elevate the user's most frequent triggers and medications to the top of the interface.
Key Results
- Created an accessible, separation-of-concerns architecture for the frontend, passing 100% of UI threshold tests.
- Reduced cognitive load for active users through intelligent, personalized sorting algorithms.
- Developed an analytical model that systematically accounts for longitudinal data drift and the subjectivity inherent in self-reported pain scores.