Reego John-Annaselvan

As subteam lead, I help direct an atomistic-to-electronic modeling workflow for gallium nitride within Georgia Tech's partnership with Sandia National Laboratories. The team is connecting controlled lattice disturbances and molecular-dynamics trajectories to Quantum ESPRESSO calculations. The current roadmap covers model validation, representative snapshot selection, input conversion, and convergence checks before interpreting changes in electronic response.

My research with Dr. Michael Bloodgood examined when active-learning systems should stop requesting labeled examples. This question connects model performance with the cost of annotation. A detailed case study will explain the experimental setup, stopping criteria, and lessons from the work, with supporting visualizations added when ready.

Across three years at Nano PV Solar, my experience connects photovoltaic technology with computational analysis. This section is reserved for a detailed account of device-characterization methods, model comparisons, and measurement interpretation, supported by material suitable for public release.

STOPLAB is a planned benchmark for understanding when active-learning stopping rules fail. The project will compare prediction-stability criteria under label noise, class imbalance, and changes in stop-set composition. Repeated trials and controlled comparisons will separate premature stopping from unnecessary labeling. The interactive view will eventually expose those trade-offs directly.

SENSORLAB is a planned study of how sensor placement affects the recovery of unknown thermal properties. A verified heat solver will generate controlled observations, allowing adaptive layouts to be compared with uniform and random placements under the same measurement budget. The emphasis is on parameter recovery, uncertainty, and the conditions under which the parameters can be identified.

LUMEN is a planned CPU/GPU path tracer for physically based lighting simulation. Development will begin with a CPU reference renderer, then explore a GPU wavefront pipeline. The eventual demo will connect sampling choices to visible image quality, while benchmarks compare convergence, throughput, and memory under matched conditions.