I'm an Applied ML engineer in New York. I specialize in teaching machines how to learn and focus on post-training and evaluation systems that help agents evolve from experience safely and measurably. My research is centered on how we continuously evaluate, govern, and interpret agents whose behavior shifts with new data, tools, and feedback.
Before ML, I was a college football coach at Ohio State and Clemson, worked for the LA Rams, taught at NYU, and invested in education technology at Emerson Collective. My throughline and core motivation is to help people get from where they're at to where they want to go. Now I do that for agents. I studied learning design at UIUC, researching how to design optimal learning environments and the science of skill acquisition. I left to found Arc, where those concepts became the foundation for how we teach agents to learn.
ATLAS — Continual learning framework for production LLM agents | Paper
CL-Bench — Benchmark for evaluating agent continual learning in stateful environments
Active contributor to Slime (multi-turn RL training) and SGLang (inference infrastructure)
Website | LinkedIn | Twitter | jbarnes850@gmail.com
Brooklyn, NY 🗽