Zhepei Wei (魏哲培)/dʒɜːpeɪ/
Ph.D. Student·Computer Science Department·University of Virginia
About me
I am a CS Ph.D. student at the University of Virginia (UVA), advised by Prof. Yu Meng. I build efficient and self-improving agentic AI, from making LLMs cheaper to run (AdaDecode) and train (RELEX), to developing autonomous agents (WebAgent-R1) and systems that recursively improve themselves with no external supervision (Self-HPO). I have interned at 
Amazon, 
Meta, and 
Microsoft Research, and I am a recipient of the 
Capital One Ph.D. Fellowship.
On the job market. I'm open to both academic and industry positions starting in Spring/Fall 2027. Please feel free to reach out if interested!
What's New
[10/2026] Check out our latest work on self-evolving coding agents — “Self-HPO: Self-Evolving Coding Agents via Harness–Policy Optimization”!
[09/2026] Our paper “You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories” is accepted at NeurIPS 2026.
[06/2026] Thrilled to receive the Capital One Ph.D. Fellowship!
[05/2026] Check out our latest work on RLVR training dynamics — “You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories”!
[05/2026] Moved to Redmond and started my internship at Microsoft Research!
[04/2026] Our paper “TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning” is accepted at ICML 2026.
[04/2026] Our paper “Aligning Large Language Models via Fully Self-Synthetic Data” is accepted at ACL 2026.
[04/2026] Thrilled to receive the 2025-2026 UVA All-University Graduate Teaching Award (10 out of ~1,000 graduate TAs across UVA)!
[01/2026] Thrilled to receive the Tinker Research Grant from Thinking Machines ($5,000 credits)!
[10/2025] Check out our latest work on training truthful LLMs — “TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning”!
[09/2025] Our paper “The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning” is accepted at NeurIPS 2025.
[08/2025] Our paper “WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning” is accepted at EMNLP 2025.
[05/2025] Thrilled to be selected as a top reviewer at ICML 2025!
[05/2025] Moved to Redmond and started my internship at Meta!
[05/2025] Check out our latest work on LLM-based web agents — “WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning”!
[05/2025] Our paper “AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism” is accepted at ICML 2025.
[01/2025] Our paper “InstructRAG: Instructing Retrieval-Augmented Generation via Self-Synthesized Rationales” is accepted at ICLR 2025.
[01/2025] Moved to Seattle and started my internship at Amazon!
[09/2024] Thrilled to receive the Copenhaver Charitable Trust Bicentennial Fellowship ($12,000)!
[07/2024] Thrilled to be accepted to OpenAI's Researcher Access Program ($5,000 API credits)!
[06/2024] Check out our latest work on explicit denoising for RAG “InstructRAG: Instructing Retrieval-Augmented Generation via Self-Synthesized Rationales”!
[05/2024] Thrilled to be nominated for the Google PhD Fellowship (up to 3 nominees per university)!
[05/2024] Thrilled to receive the John A. Stankovic Outstanding Graduate Research Award!
[01/2024] Our paper “Incentivized Truthful Communication for Federated Bandits” is accepted at ICLR 2024.
[09/2023] Our paper “Incentivized Communication for Federated Bandits” is accepted at NeurIPS 2023.
[08/2022] Relocated to the lovely Charlottesville and began my Ph.D. journey at UVA!
Selected Publications
Preprint Agents Self-HPO: Self-Evolving Coding Agents via Harness–Policy Optimization [code]
Zhepei Wei, Wenlin Yao, Baolin Peng, Danqing Wang, Qianhui Wu, Hao Cheng, Yu Meng, Jianfeng GaoNeurIPS 2026 Efficiency You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories [code]
Zhepei Wei, Xinyu Zhu, Wei-Lin Chen, Chengsong Huang, Jiaxin Huang, Yu MengICML 2026 Trustworthiness TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning [code]
Zhepei Wei, Xiao Yang, Kai Sun, Jiaqi Wang, Rulin Shao, Sean Chen, Mohammad Kachuee, Teja Gollapudi, Tony Liao, Nicolas Scheffer, Rakesh Wanga, Anuj Kumar, Yu Meng, Wen-tau Yih, Xin Luna DongEMNLP 2025 Agents WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning [code]
Zhepei Wei, Wenlin Yao, Yao Liu, Weizhi Zhang, Qin Lu, Liang Qiu, Changlong Yu, Puyang Xu, Chao Zhang, Bing Yin, Hyokun Yun, Lihong LiICML 2025 Efficiency AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism [code]
Zhepei Wei, Wei-Lin Chen, Xinyu Zhu, Yu MengICLR 2025 Trustworthiness InstructRAG: Instructing Retrieval-Augmented Generation via Self-Synthesized Rationales [code] [webpage]
Zhepei Wei, Wei-Lin Chen, Yu MengICLR 2024 RL Theory Incentivized Truthful Communication for Federated Bandits
Zhepei Wei, Chuanhao Li, Tianze Ren, Haifeng Xu, Hongning WangNeurIPS 2023 RL Theory Incentivized Communication for Federated Bandits
Zhepei Wei, Chuanhao Li, Haifeng Xu, Hongning Wang