User behavior modeling for design optimization

Products fail not because of bad engineering, but because human behavior is hard to predict. I develop computational models of human cognition and behavior — grounded in machine learning, active inference, and naturalistic data — that let you anticipate how users will respond before you build, and diagnose why they struggle after. My lab's work spans drivers interacting with automated vehicles, clinicians making high-stakes decisions, and workers managing complex systems: in each case, the output is a validated behavioral model or data-driven insight that feeds directly into design recommendations.

Engagements typically move through three milestones: (1) characterize — rapid analysis of existing data or a targeted user study (driving simulation, wearables, instrumented tasks) to identify where behavior breaks down; (2) model — a prototype computational or ML model that predicts user actions, errors, or workload under realistic conditions; and (3) validate and translate — testing the model against ground-truth outcomes and delivering actionable design criteria or evaluation benchmarks. Past partners have included Waymo, Toyota, State Farm, and the U.S. DOT, with project timelines typically ranging from three months to five years depending on scope.