Kehan Guo

LLM Post-Training, Agents, and Verifiable Evaluation

Email · Google Scholar · GitHub · X · LinkedIn · CV

Oreo, my dog Portrait of Kehan Guo

hover to meet Oreo

I am a Ph.D. candidate in Computer Science at the University of Notre Dame, advised by Prof. Xiangliang Zhang. I am currently a research intern at Apple, working on post-training for large language models, following research internships at Amazon AWS AI and ByteDance.

I build reliable LLM agents and design the feedback that trains and evaluates them. My research asks what models actually learn when optimized against imperfect proxies: do they acquire the intended capability, exploit the evaluator, or overfit the environment? I work on reward and verifier design for post-training, tool-using agents grounded in external evidence, and evaluations that expose failures under real constraints. Scientific reasoning, especially chemistry, is a central testbed because the evidence is concrete and the outcomes are checkable.

More broadly, I study interfaces for steering generative models, including how noise-data coupling in flow matching can encode property control without an inference-time reward model.

Selected Publications

A selection; * denotes equal contribution. Full list on Google Scholar.

Lead and Equal-Contribution Work
Reward Transport: Property Control in Flow Matching via Noise-Space Alignment
Kehan Guo, Yili Shen, Yujun Zhou, Yue Huang, Chujie Gao, Shiyi Du, Xiangliang Zhang
Under review, 2026. [Code]
Can LLMs Solve Molecule Puzzles? A Multimodal Benchmark for Molecular Structure Elucidation
Kehan Guo, Bozhao Nan, Yujun Zhou, Taicheng Guo, Zhichun Guo, Mihir Surve, Zhenwen Liang, Nitesh V. Chawla, Olaf Wiest, Xiangliang Zhang
NeurIPS, 2024. Spotlight. [Paper] [Code]
What Can Large Language Models Do in Chemistry? A Comprehensive Benchmark on Eight Tasks
Taicheng Guo*, Kehan Guo*, Bozhao Nan, Zhenwen Liang, Zhichun Guo, Nitesh V. Chawla, Olaf Wiest, Xiangliang Zhang
NeurIPS, 2023. [Paper] [Code]
Evaluating Large Language Models in Scientific Discovery
Zhangde Song*, Jieyu Lu*, Yuanqi Du*, Botao Yu*, Thomas M. Pruyn*, Yue Huang*, Kehan Guo*, Xiuzhe Luo*, Yuanhao Qu*, and others
arXiv preprint, 2025. [Paper] [Code] [Dataset] [Oracle]
Contribution: Led chemistry question collection and evaluation.
Selected Collaborations
Getting Better at Working With You: Compiling User Corrections into Runtime Enforcement for Coding Agents
Yujun Zhou, Kehan Guo, Haomin Zhuang, Xiangqi Wang, Yue Huang, Zhenwen Liang, Pin-Yu Chen, Tian Gao, Nuno Moniz, Nitesh V. Chawla, Xiangliang Zhang
arXiv preprint, 2026. [Paper] [Code]
AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking
Xiangqi Wang, Yue Huang, Yanbo Wang, Xiaonan Luo, Kehan Guo, Yujun Zhou, Xiangliang Zhang
NeurIPS, 2025. Spotlight. [Paper] [Code]
ChemOrch: Empowering LLMs with Chemical Intelligence via Synthetic Instructions
Yue Huang, Zhengzhe Jiang, Xiaonan Luo, Kehan Guo, Haomin Zhuang, Yujun Zhou, Zhengqing Yuan, Xiaoqi Sun, Jules Schleinitz, Yanbo Wang, Shuhao Zhang, Mihir Surve, Nitesh V. Chawla, Olaf Wiest, Xiangliang Zhang
NeurIPS, 2025. [Paper] [Code]
Causally-Enhanced Reinforcement Policy Optimization
Xiangqi Wang, Yue Huang, Yujun Zhou, Xiaonan Luo, Kehan Guo, Xiangliang Zhang
arXiv preprint, 2025. [Paper] [Code]

News

More news
  • 2026.04Two papers accepted at ICML 2026: ProbeLLM and Capability-Oriented Training Induced Alignment Risk.
  • 2026.04Research Scientist Intern at ByteDance (San Jose), Apr–Jun 2026.
  • 2026.04PolicyLLM accepted at ACL 2026 (Findings).
  • 2026.03Passed Ph.D. candidacy exam.
  • 2026.01TrustGen accepted at ICLR 2026.
  • 2025.08Completed Applied Scientist internship, Amazon AWS AI, NYC.

Experience

Education

Awards

Service

Reviewer: NeurIPS (2024–25), ICLR (2025–26), ICML, AAAI, IJCAI, KDD, WWW, ACL Rolling Review (ACL/EMNLP).

Beyond Research

Most days outside the lab I'm with Oreo, my dog. He's the one in the photo if you hovered.