I lead the simulation subteam in Georgia Tech's Compound Semiconductor Products VIP partnership with Sandia National Laboratories. We study how atomic motion and distortion in gallium nitride change its electrical properties, helping evaluate ideas for neutron detection. My work connects atomic-motion simulations with calculations of the material's electrical response and checks how sensitive the results are to numerical choices.

During my freshman year at The College of New Jersey, I worked with Dr. Michael Bloodgood on stopping criteria for active learning using small-text. The research examined when a learner can stop requesting annotations while retaining predictive performance. This experience motivated LastLabel, my subsequent independent study of published stopping methods under annotation noise and alternative stop-set constructions.

Across three years at Nano PV Solar, I combined team leadership with hands-on development of Python tools to analyze solar-cell measurements and compare them with device models. I examined how test conditions and measurement variability affect the extracted parameters, helping distinguish changes in device behavior from testing artifacts.

LastLabel is my independent study of when a learning system should stop requesting human labels. Building on methods by Dr. Michael Bloodgood and his coauthors, I compared labeling effort with performance on unseen examples across 15,360 training runs. The experiments examine incorrect labels and different ways of checking whether learning has settled. Saving labels can also mean giving up performance, and the best stopping point depends on the task. Study, methods, and full results ↗

SENSORLAB is a research concept centered on the relationship between sensor placement and thermal-property recovery. Its proposed experimental design pairs a differentiable heat model with adaptive, uniform, and random measurement layouts under equal budgets. The central questions concern parameter error, uncertainty, and identifiability from sparse observations.

LUMEN builds images by following simulated light as it bounces around a scene. That produces soft shadows, reflections, and glass effects from the same lighting model. I built CPU and GPU renderers and compared their output and rendering time. The animation uses saved renders, so it plays without a powerful graphics card. Source code and rendering benchmarks ↗