I build agent systems.
Most of what I care about lives around agents: harness engineering, tool use, post-training, RL-flavored data loops, and the stuff that makes models hold up once they leave the demo stage.
I like working on both sides of the problem: shaping model behavior, and building the engineering around it so that behavior actually survives contact with real tasks.
Agents are just models with tools, memory, and consequences.
Currently spending a lot of time on coding agents, agent workflows, trajectory filtering, evaluation harnesses, and fine-tuning pipelines.
Still learning. Still building. Still trying to make agents a little less fake-smart and a little more useful.
