Hi, I'm Kehan Guo.
I'm a fourth-year Ph.D. student in the MINE Lab at the University of Notre Dame, advised by Prof. Xiangliang Zhang.
I build reflective scientific agents — LLM‑based systems that reason about science, check themselves, and improve via tools. I began in AI‑for‑Science (chemistry) and now focus on world‑modeling for LMs (rule‑/constraint‑aware reasoning that mirrors how the world works) and reflective loops (critique → verification → refinement) to make agents robust and auditable. Methodologically I work on RL post‑training/alignment, uncertainty calibration, and agentic tool use; application‑wise I turn messy, literature‑grounded chemistry problems (extraction, retrosynthesis, data curation) into structured, provenance‑clear tasks and benchmarks. My goal is to bridge scientific reasoning and reflective agent design, delivering systems that ship as products while advancing methods.
Research
I build large language models and generative AI systems that reason reliably in complex scientific domains.
My current focus is world modeling for large language models, where models learn and are constrained by rules so they mirror how the world works, and self-evolving LMs that critique, verify, and iteratively improve via tools and interaction loops.
Methodologically I work on RL post-training and alignment, uncertainty calibration, and agentic tool use. Application-wise I emphasize AI for Science with a focus on chemistry, including literature-grounded extraction, retrosynthesis planning, and data curation. I also design complex, cross-domain benchmarks and datasets, turning messy real-world data into structured, auditable tasks with clear provenance and rigorous scoring.
What's New
- 2025.09 Two papers accepted to NeurIPS 2025: ChemOrch (LLMs + chemical tools) and AdaReasoner — Adaptive Reasoning Enables More Flexible Thinking (Spotlight).
- 2025.09 New preprint: Causally‑Enhanced Reinforcement Policy Optimization (RL + causal coherence for LLMs). [arXiv](https://arxiv.org/abs/2509.23095)
- 2025.08 Uncertainty‑aware Yield Prediction accepted to CIKM 2025 — see you in Seoul!
- 2025.04 Survey “Artificial Intelligence in Spectroscopy” accepted to IJCAI 2025.
- 2025.02 Check out latest preprint: AI in Spectroscopy, HHH principle
- 2025.02 I will join Amazon AWS as an Applied Scientist Intern in 2025 Summer. See you in NYC!
Selected Publications
See full list in Publications.
Contact
Email: kguo2@nd.edu
If my research interests you, feel free to reach out via email.