Ian Chen

Principal Applied Scientist @ Microsoft

Research Interests

  • LLM Agent Evaluation — rubric-based quality assessment for conversational AI agents
  • Rubric Quality & Generation — automated rubric optimization via perturbation-driven HPO
  • Reward Modeling — rubrics as reward signals for RL fine-tuning (GRPO/DRO)
  • Reinforcement Learning for LLM — reward hacking detection, training signal quality

Background

Traditional ML → Deep Learning → Reinforcement Learning → LLM Agents

I work on making LLM agents more reliable through better evaluation. My current focus is on rubric quality — ensuring that the checklists we use to grade AI outputs actually measure what they claim to measure.

Contact