Reliable reasoning
I design reasoning frameworks that make language models more consistent, more controllable, and more trustworthy on complex tasks.
Reliable reasoning and decision-making
with LLM post-training, reinforcement learning, and agents.
I am a Ph.D. candidate at Tianjin University and a member of the TJU DRL Lab. I work with Jianye Hao, Yan Zheng, and Hongyao Tang on reinforcement learning, LLM post-training, and agentic systems that make language models reason more reliably, act more effectively, and support scientific discovery.
Signature Themes
Research Pillars
Three directions that define how I think about reliable language-model systems and their real-world use.
LLM post-training
I study reinforcement-learning-based post-training methods that improve capability while reducing cost and instability.
Agentic systems for science
I build agent systems that translate language-model reasoning into useful workflows for research, analysis, and discovery.
Research
Selected Publications
A curated set of papers that best represent my current work across reasoning, agent systems, and AI for science.
Featured
Reasoning
From Chaos to Order: The Atomic Reasoner Framework for Fine-grained Reasoning in Large Language Models
Breaks complex reasoning into small atomic steps so language models can reason more reliably without relying on heavyweight search or tool orchestration.
CellAgent: LLM-Driven Multi-Agent Framework for Natural Language-Based Single-Cell Analysis
Lets researchers run end-to-end single-cell and spatial transcriptomics analysis through natural language, reducing programming overhead without sacrificing result quality.
Ovd-explorer: Optimism should not be the sole pursuit of exploration in noisy environments
Improves exploration in noisy reinforcement learning settings by separating genuinely useful uncertainty from stochastic noise that would otherwise mislead the agent.
Systems
Selected Projects
Selected systems and learning artifacts that turn research ideas into tools people can actually use.
LLM Agent Tutorial
Project page for the LLM Agent Tutorial website, highlighting the public learning resource, companion materials, and entry points for readers.
Updates
Recent News
Recent milestones across papers, systems, tutorials, and community work.
- 📝 2026-01 Three papers accepted by ICLR 2026: ReMix, CellAgent, and From Seeing to Doing.
- 🧠 2026-01 Two workshop papers accepted at ICLR 2026: AgentMemoryBench and DistRLVR.
- 📝 2026-01 Paper accepted by ACM TheWebConf 2026 Industry: AFE-Master.
- 🔥 2025-12 Released our beginner-friendly LLM Agent tutorial (website & PDF).
- 🎤 2025-12 The 137th RLCHINA Paper Seminar hosted!
Trajectory
Experience Snapshot
A brief view of research training, industry collaboration, and selected recognition.
Experience
Algorithm Research Intern
Shanghai AI Lab (advised by Shuyue Hu)
Algorithm Research Intern (Project Collaboration)
Kuaishou (advised by Hangyu Mao)
Algorithm Research Intern
NetEase (advised by Yujing Hu)
Selected Honors
CSIG Science and Technology Progress Award, First Prize (2025年度CSIG科技进步奖一等奖)
CSIG
Distinguished PC Members in AAMAS 2025
AAMAS 2025
Academic First-class Scholarship (Top 10%)
Tianjin University