PERSONAL / RESEARCH
LLM Research Algorithm Intern · PhAI Labs
PKU Intern · Peking University
I work on self-improving AI agents and AI for science. My research explores how agents can learn from experience, how to evaluate that improvement, and how to build agents that collaborate with researchers on scientific tasks.
Self-improvement, from evaluation to scientific collaboration.
arXiv preprint · October 2026
A framework built on JEPA-Anything that uses predictive world models and interaction feedback to improve agents and workflows across code repair, GUI interaction, laboratory planning, reaction optimization, and immune science.
arXiv preprint · September 2026
An interactive scientific workspace that connects researcher collaboration with agent harness evolution and model reinforcement learning.
arXiv preprint · August 2026
A benchmark for testing whether persistent agents improve from retained experience, with matched controls and evidence of how improvement happens.
GET IN TOUCH
For questions about my work or research collaboration: