Latest work
Behavioral Self-Rewarding (BSR)
Learning from debugging behavior to improve software engineering agents beyond binary test rewards.
Preprint, 2026
About me
I am a PhD candidate at Penn State’s College of Information Sciences and Technology, where I have been advised by Prof. Qingyun Wu since 2022. Before joining Penn State, I received my master’s degree in AI & ML from Imperial College London and my bachelor’s degree in Computer Science from the University of California, Davis.
My research focuses on large language models, especially multi-agent systems and reinforcement learning for language agents.
I was a research intern at Microsoft Research, Redmond in 2024 and 2025.
I’m looking for full-time research scientist positions, preferably in North America. Résumé · Contact me
Selected projects
Agent-native computers
Unified environments for digital agents, connecting graphical and text interaction.
ICML 2026 · Position Paper Track
ExCyTIn-Bench
A benchmark for evaluating LLM agents on cyber threat investigation tasks.
Creator · ICML 2026
Absolute Zero
Reinforced self-play for reasoning, with no human-curated training data.
Co-creator & maintainer · NeurIPS 2025
AutoGen / AG2
An open-source framework for building LLM applications through multi-agent conversation.
Co-creator & maintainer
Explore by topic
Jump to year
No selected papers in this topic yet. Switch to Full to explore all publications.
Beyond Binary Verifiable Rewards: Behavioral Self-Rewarding for SWE Agents
Preprint 2026
Position: Digital Agents Require Unified Agent-Native Computers
ExCyTIn-Bench: Evaluating LLM Agents on Cyber Threat Investigation
SimpleDoc: Multi-Modal Document Understanding with Dual-Cue Page Retrieval and Iterative Refinement
Absolute Zero: Reinforced Self-play Reasoning with Zero Data
StateFlow: Enhancing LLM Task-Solving through State-Driven Workflows
AutoDefense: Multi-Agent LLM Defense against Jailbreak Attacks
AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
MathChat: Converse to Tackle Challenging Math Problems with LLM Agents
2026
Beyond Binary Verifiable Rewards: Behavioral Self-Rewarding for SWE Agents
Preprint 2026
When Does Multi-Agent RL Improve LLM Workflows? Workflow, Scale, and Policy-Sharing Tradeoffs
MetaAgent-X: Breaking the Ceiling of Automatic Multi-Agent Systems via End-to-End Reinforcement Learning
Position: Digital Agents Require Unified Agent-Native Computers
DispatchMAS: Fusing Taxonomy and Artificial Intelligence Agents for Emergency Medical Services
MemCollab: Cross-Agent Memory Collaboration via Contrastive Trajectory Distillation
Live-Evo: Online Evolution of Agentic Memory from Continuous Feedback
ExCyTIn-Bench: Evaluating LLM Agents on Cyber Threat Investigation
2025
A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
SimpleDoc: Multi-Modal Document Understanding with Dual-Cue Page Retrieval and Iterative Refinement
Absolute Zero: Reinforced Self-play Reasoning with Zero Data
2024
StateFlow: Enhancing LLM Task-Solving through State-Driven Workflows
AutoDefense: Multi-Agent LLM Defense against Jailbreak Attacks
AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
MathChat: Converse to Tackle Challenging Math Problems with LLM Agents
HyperTime: Hyperparameter Optimization for Combating Temporal Distribution Shifts
2023
Unified Off-Policy Learning to Rank: A Reinforcement Learning Perspective
2022
Automated Object Detection in Experimental Data Using a Combination of Unsupervised and Supervised Methods
* Equal contribution.
Back to top