Problems I keep returning to
My work studies how machine learning can infer latent state from noisy, heterogeneous data. I have worked across passive sensing, clinical movement data, biomedical process variables, geospatial risk, and healthcare datasets. The common problem is measurement: how to recover useful structure from signals that are incomplete, indirect, and collected outside ideal experimental conditions.
Human behavior and physiological intelligence
Human behavior produces continuous low-friction signals: movement, sleep-wake timing, device interaction, motor control, and changes in daily rhythm. I model these signals to estimate behavioral, physiological, and clinical state. This area sits closest to my training in computer science and neuroscience, especially in passive sensing, digital phenotyping, and clinically relevant prediction.
Healthcare decision systems
Healthcare data is delayed, siloed, and usually analyzed after the most important decisions have already been made. At Avenir, I build systems that integrate claims, benefits, vendor, and population-level data so organizations can identify cost drivers, risk concentration, and intervention opportunities with more precision. I am interested in how AI can support decision-making under uncertainty, not just summarize existing reports.
AI for scientific and biological discovery
Biological and engineered systems are nonlinear, noisy, and expensive to probe experimentally. I use interpretable machine learning to model these systems, including biofabrication, biomaterials, environmental exposure, and public-health risk. The goal is to make experimentation more directed by identifying which variables appear to matter, where uncertainty remains, and what should be tested next.