Reinforcement Learning for Reasoning · LLMs · Structured AI Systems
I am a researcher focusing on reinforcement learning, reasoning systems, and structured representations for large language models. I completed my M.S. in Machine Learning and Data Science at UC San Diego, where I worked with Prof. Zhiting Hu on offline RL for multi-step reasoning and credit assignment.
My work examines how intermediate reasoning states can be modeled, evaluated, and optimized. I have developed methods integrating soft Bellman consistency, MCTS-guided trajectory sampling, and operator-level value modeling to improve stability and generalization in long-horizon reasoning. I also study verifiable reasoning mechanisms for aligning intermediate semantic states through coherence, consistency, completeness, and grounding criteria.
My research interests center on building reliable and interpretable reasoning systems through reinforcement learning, structured representations, and controlled architectural evolution. I am applying to PhD programs for Fall 2026 to further pursue these directions.
I am a machine learning researcher interested in reinforcement learning, multi-step reasoning, and structured representations for large language models. I received my M.S. in Machine Learning and Data Science from UC San Diego, where I worked in Prof. Zhiting Hu’s MixLab on offline RL, soft Bellman consistency, and MCTS-guided reasoning policy learning.
My recent work focuses on modeling and evaluating intermediate reasoning states, including coherence, consistency, completeness, and grounding. I also design system-level methods for reliable LLM deployment, including retrieval-integrated workflows, value modeling for operator sequences, and scalable serving pipelines. In parallel, I collaborate with ShelteredAI to build applied LLM systems for real-world social-service environments.
My broader research goal is to develop structured, interpretable, and self-improving reasoning frameworks that unify reinforcement learning, verifiable intermediate states, and controlled structural evolution in agentic LLMs. I am currently applying to PhD programs for Fall 2026 to further pursue these directions.
MixLab (Machine Learning & Reasoning Group), University of California, San Diego
Advisor: Prof. Zhiting Hu
Studied reinforcement learning for multi-step reasoning in large language models. Developed offline RL methods with soft Bellman consistency, MCTS-guided trajectory sampling, and operator-level value modeling. Achieved substantial improvements on MATH, GSM8K, and ALFWorld generalization benchmarks.
ShelteredAI (Non-profit), Charlotte & New York
Advisor: Lead ML Team
Built AI systems for social-service delivery, focusing on domain adaptation, few-shot learning, and structured reasoning alignment. Designed evaluation metrics for intermediate reasoning states including coherence, consistency, completeness, and grounding. Developed Dockerized LLM microservices with vLLM, FAISS retrieval, and Twilio-based phone workflows.
NEAT Labs (Neural Engineering & Translation Labs), University of California, San Diego
Advisor: Prof. Mariam S. Thomas (Lab PI)
Conducted multimodal neural data analysis involving EEG, ERP, and behavioral signals. Implemented statistical evaluation pipelines to study cognitive and neural modulation, and automated psychometric scoring workflows with REDCap integration.
Oceanic AI & Remote Sensing Group, Nanjing University of Information Science and Technology
Advisor: Prof. Wenjin Sun
Investigated significant wave height forecasting using LSTM and classical numerical modeling. Processed long-term buoy datasets, handled large missing-value patterns, and evaluated climate-dependent model behavior across Caribbean and Atlantic regions.
AI Oceanography Collaboration, NUIST & UCLA
Advisor: Prof. Wenjin Sun and UCLA collaborators
Implemented PSPNet for multi-layer satellite feature extraction. Benchmarked AI-based oceanic eddy detection against geometric baseline methods and analyzed regional morphology differences.
* denotes equal contribution
Huaijie Wang, Shibo Hao, Hanze Dong, Shenao Zhang, Yilin Bao, Ziran Yang, Yi Wu
ACL(Presented)
Brandon J. Bethel, Changming Dong, Shuyi Zhou, Wenjin Sun, Yilin Bao
SSRN Working Paper(Published)
I am currently seeking PhD opportunities for Fall 2026 in Computer Science and Electrical & Computer Engineering departments. If you are interested in my work or would like to discuss potential research collaborations, feel free to get in touch.
