Building AI systems that are measurable, reliable, and ready to ship.

I am an AI Systems / ML Engineer focused on LLM systems, generative models, multimodal perception, and robot-learning data infrastructure. I like problems where research ideas have to survive real constraints: limited compute, noisy data, latency budgets, reproducibility requirements, and production operations.

During my software engineering internship at Abundant, I built AI-agent benchmarks, RL task environments, and evaluation infrastructure for frontier labs, focused on long-horizon DevOps/SRE tasks. I designed multi-service incidents, task families with distinct root causes, and hidden verifiers, then built the runtime and review tooling needed to evaluate them reliably. The goal was to measure whether agents could sustain an investigation across tools and services, resolve operational failures, and generalize beyond memorized fixes.

At Carnegie Mellon, my VLA / robot-learning research focuses on reliable data pipelines for manipulation datasets. Across agent evaluation and robotics, I care about the same question: how do we build environments, data, and measurements that tell us what an AI system can actually do?

My research path also includes EEG and wearable-sensing work at Khan Lab at USC, where I built experiment software, firmware support, data-processing pipelines, and machine-learning classifiers for cognitive and affective-state analysis.