arXiv preprint
I developed a task-aware framework for comparing humanoid actuation with human biomechanics and locating weak joint-task pairs.
MD Sunbeam
Research engineer at the U.S. Army Research Laboratory
I study how learning systems fail when data, compute, or feedback are limited. I build evaluations, demonstration data, and research software for models that act in games, simulators, and robots.
Selected work
KnowFM @ AAAI 2025 · Poster
I built the evaluation harness and ran the in-context learning experiments for a closed-loop benchmark of multimodal models as Atari policies.
Interactive Learning with Implicit Human Feedback Workshop · ICML 2023
I created the gaze-augmented demonstrations, designed the evaluation protocol, and ran experiments for a visual imitation-learning model.
TensorBloom
2026
Personal project
A visual editor that propagates tensor shapes through PyTorch graphs and reports structural errors before training.
Writing
What an Evaluation Harness Decides
The model is only one part of a closed-loop benchmark. Parsing, recovery, timing, and rollout accounting determine what its score means.