Homepage
I am a final-year Ph.D. candidate in Computer Science at the University of California, Los Angeles.
UniversityI build AI systems that reason over scientific and structured data. My recent focus is on LLM agents for search, planning, optimization, and autonomous discovery over long horizons.
I am a recipient of the Amazon Fellowship, the D. E. Shaw Research Doctoral and Postdoctoral Fellowship, and the UCLA Graduate Dean’s Scholar Award. My publications have received more than 10,000 citations.
Recent Work
RetroAgent COLM 2026
LLM agents for search and planning: an LLM searches over structured memory to plan multi-step chemical syntheses, trained with reinforcement learning from verifiable rewards.
FormalFlow
Long-horizon autoformalization agents that coordinate through a shared repository, continuous integration, and code review. In 63 days they produced a complete, machine-checked Lean 4 proof of a core theorem underlying MIP* = RE, about 126K lines of Lean, and repaired two errors in the published theorem statement.
Also ARLArena (ICML 2026), a framework for stable agentic RL training, and benchmarks and models for AI for science: SciBench (ICML 2024), MARCEL (ICLR 2024), SPiCE (NeurIPS 2025), MatSciBench (KDD 2026), CrystalSTAR (COLM 2026), and SLOT-IR (NeurIPS 2026).
Recent Updates
- 9/24/2026 SLOT-IR accepted to NeurIPS 2026. Congrats to Jingru and team!
- 9/16/2026 Our work FormalFlow on long-horizon autoformalization is out! Read the blog post.
- 7/8/2026 RetroAgent accepted to COLM 2026. See you in 🌉 San Francisco!
- 6/17/2026 MatSciBench accepted to KDD 2026. Congrats to Junkai and Jingru!
- 4/30/2026 ARLArena accepted to ICML 2026. Congrats to the team!