Ian Chen

I explore reliable tool-using AI agents through evaluation, rubric quality, trajectory-level diagnosis and controlled improvement experiments.

Research Interests

  • LLM Agent Evaluation — trustworthy measurement of realistic agent behavior
  • Rubric Quality — whether grading criteria measure the intended behavior
  • Reward Modeling — the relationship between evaluation and useful learning signals
  • Reinforcement Learning — training-signal quality and generalization

Current Work

Agent Improvement Lab is an independent public research project. Its first milestone is the design of Evaluation Harness v0.1 for tau2 Retail.

Contact