I am an ML / Research Engineer working on LLM post-training, evaluation, and reinforcement learning. What ties my work together is a single question I have been chasing since my PhD: how do you get a learning system to improve in a way you can actually measure and trust?
Training and evaluation as one loop
At Postman I own both halves of that loop. I build multi-phase post-training pipelines (SFT, DPO, RLVR with GRPO, and RFT) on large-scale GPU infrastructure, and I design the agent evaluation benchmarks that decide whether a change actually helped. I treat a benchmark as an instrument: deterministic validators, anti-contamination, bootstrap-CI scoring, and multi-trial averaging, rebuilt into harder task suites when frontier models saturate them. I also work on separating real capability from safety-driven refusal, so a model is not judged weaker than it is just because it abstains.
Reinforcement learning, from games to LLMs
Reinforcement learning is the throughline. My CS PhD at Virginia Tech (Outstanding PhD Research Award, 2025) used game-theoretic and deep RL methods for cyber-defense, including defensive deception against advanced persistent threats and a decision-theory-guided framework that cut the cold-start cost of deep RL. Reward design, environment design, and the careful study of what a policy is really learning are the same problems whether the agent is defending a network or solving an API task.
Knowing what you do not know
A second thread is uncertainty. I built uncertainty-aware human-AI teaming systems using Vision Transformers and Evidential Deep Learning, and co-authored a survey bridging Dempster-Shafer theory, subjective logic, and neural-network uncertainty quantification (Information Fusion). Calibrated uncertainty is exactly what you need when you separate a model’s real capability from its safety-driven refusals.
Now
I am a Senior AI Software Engineer at Postman, where I am the DRI for model training and evaluation. Earlier I shipped production ML systems at Bobyard. The throughline is the same: build the thing that learns, and build the instrument that tells you the truth about it.