Zhengxu Yan
About Me
I'm a Computer Science M.S. student at Stanford University and a recent UC Berkeley graduate in Computer Science and Data Science. My work spans verifiable code generation, diffusion transformer systems, and AI for social impact, and I love building collaborative teams that turn research into real-world impact.
Stanford, CA
Publications
Under Review
- Ye, Z., Yan, Z., He, J., Kasriel, T., Yang, K., & Song, D. VERINA: Benchmarking Verifiable Code Generation. Under review at ICLR 2026. https://arxiv.org/abs/2505.23135
- Luo, M., Hao, A., Yan, Z., Cao, C., & Nguyen, Q. L. N. DiT-Serve: An Efficient Serving Engine for Diffusion Transformers. Under review at ICLR 2026.
Published
- Yan, Z., Dube, V., Heselton, J., Johnson, K., Yan, C., Jones, V., Blaskewicz Boron, J., & Shade, M. (2024). Understanding older people’s voice interactions with smart voice assistants: a new modified rule-based natural language processing model with human input. Frontiers in Digital Health, 6, 1329910. https://doi.org/10.3389/fdgth.2024.1329910
- Jones, V. K., Yan, C., Shade, M. Y., Boron, J. B., Yan, Z., Heselton, H. J., Johnson, K., & Dube, V. (2024). Reducing Loneliness and Improving Social Support among Older Adults through Different Modalities of Personal Voice Assistants. Geriatrics (Basel, Switzerland), 9(2), 22. https://doi.org/10.3390/geriatrics9020022
Working Papers
VeriAgentBench: Benchmarking Project-Level Verifiable Code Agents in the Wild
- Introduced a benchmark of 50 real-world repositories supporting repository-level verification of existing projects and verifiable generation from scratch.
- Supervised by Dr. Dawn Song and Dr. Kaiyu Yang, I developed an automated evaluation pipeline measuring proof validity and functional correctness, revealing that state-of-the-art code agents achieve single-digit success rates on VeriAgentBench.
Adaptive Operations Management in Buildings: A Reinforcement Learning Approach for Operational Adaptability in Healthcare Facilities
- Introduced a reinforcement learning-based framework to enable adaptive and integrated operations management in healthcare facilities, optimizing spatial, social, and operational performance through coordinated resource sharing.
- Supervised by Dr. Yehuda Kalay and Dr. Davide Schaumann, I utilized deep reinforcement learning (RL) and simulation to develop a smart building management system for the Cardiac Catheterization Lab at St. Bernardine Medical Center, significantly enhancing facility adaptability and operational efficiency. The manuscript is in preparation for journal review.
Work Experience
Teaching Assistant for Large Language Model Agents (CS 194-196)
August 2024 - PresentDepartment of Electrical Engineering and Computer Sciences at UC Berkeley
- Coordinated course logistics, managed lecture recordings, enhanced assignment materials, and provided academic support by addressing student inquiries.
- Assisted in organizing and coordinating the LLM Agents Hackathon, hosted by Berkeley RDI in conjunction with the LLM Agents MOOC, designed to foster innovation, expand the AI agent community, and advance LLM agent technology.
Research Assistant
February 2024 - PresentUC Berkeley Sky Computing Lab
- Competitively selected through UC Berkeley EECS Diversifying Access to Research in Engineering (DARE) and Sky Summer Undergraduate Programs to join the DiT-Serving Project as a research assistant.
- Led a team of 5 research assistants in integrating Ring Attention and Brick Attention into Diffusion Transformer (DiT) models, pioneering scalable video generation techniques.
- Directed continuous batching strategies to optimize system throughput, enhancing performance efficiency across video processing requests.
Research Assistant
February 2024 - PresentUC Berkeley College of Environmental Design
- Competitively selected to join the Smart Hospital Project through UC Berkeley Undergraduate Research Apprentice Program.
- Led a team of 3 research assistants in the development and implementation of a deep reinforcement learning-based smart building management system for the Cardiac Catheterization Lab at St. Bernardine Medical Center.
Machine Learning (ML) Engineer Lead
January 2024 - PresentUC Berkeley School of Public Health
- Selected as lead computer scientist for Rapid Reviews: Infectious Diseases via the UC Berkeley CDSS Discovery Program.
- Independently leveraged a fine-tuned Large Language Model (LLM) to efficiently categorize and identify preprints within the RR\ID domain.
- Led a team of 3 UC Berkeley CS students in utilizing LLM APIs to analyze and provide insights on medical preprints.
- Directed a team of 5 UC Berkeley CS students in engineering scripts to automate the collection of medical preprints.
- Collaborated with the Dean of the UC Berkeley School of Public Health and researchers from UCSF to develop automated systems for the academic review of infectious disease papers, set to be published by MIT Press.
Research Assistant
April 2023 – September 2023University of Nebraska
- Collaborated with gerontology and nursing researchers to develop a natural language processing (NLP) model designed to automate the processing of speech-to-text data from user interactions with AI-enabled smart voice assistants.
Backend Engineer Lead
October 2022 – December 2022Coffee Tea, Inc.
- Led a team of 3 UC Berkeley computer science students in backend development for a social platform.
- Directed the design and development of backend REST APIs using FastAPI, Poetry, Alembic, and PostgreSQL.
Projects
Machine Learning
Multi-Agent LLM Trading System
August 2024 – PresentPython, Pytorch, Tensorflow, AutoGen
- Developed a multi-agent trading system integrating LLMs with specialized agents for multimodal data processing, layered memory, and Retrieval-Augmented Generation, aiming to exceed standard prediction models by 5-7% in market accuracy.
IM2SPAIN Project
January 2024 – May 2024Python
- Employed nearest neighbors (k-NN) to predict geographic coordinates based on CLIP embeddings of geo-tagged images from Flickr, capturing diverse landscape features across Spain.
MNIST Competition
January 2024 – May 2024Python, Pytorch
- Engaged in the Kaggle MNIST classification challenge, leveraging Linear and Quadratic Discriminant Analysis, Logistic Regression, Multi-layer Perceptron, Support Vector Machines, and Convolutional Neural Networks to optimize prediction accuracy.
Language Identification RNN
August 2023 – December 2023Python, Tensorflow
- Engineered an RNN to identify word languages, achieving over 81% test set accuracy.
Graph Partitioning for Unsupervised Learning
August 2022 – December 2022Golang, AWS, Google Cloud
- Developed graph partitioning algorithms using the Kernighan–Lin method to support unsupervised machine learning workflows.
- Implemented a relational database and Golang backend deployed on AWS and Google Cloud to manage outputs and orchestrate large-scale experiments.
Systems Programming and Software Development
Pintos Operating System
January 2023 – May 2023C, Rust
- Led the development of a comprehensive Pintos operating system, focusing on systems programming, memory allocation, resource management, file systems, networking, and security.
Secure Client Application
January 2023 – May 2023Golang
- Developed a secure client Golang application incorporating cryptographic primitives for authentication and file management.
Gitlet Version Control System
January 2021 – May 2022Java
- Created "Gitlet," a Git-like Version Control System, to streamline tracking and management of code changes across projects.
Education
Stanford University
Expected Graduation: April 2027Master of Science in Computer Science
University of California, Berkeley
August 2021 – May 2025Bachelor of Arts in Computer Science and Data Science
GPA: 3.95/4.00, High Distinction
Selected Coursework:
Optimization Models (EECS 127), Probability for Data Science (Data C140), Computer Security (CS 161), Operating Systems (CS 162), Efficient Algorithm (CS 170), Computer Vision (CS 180), Artificial Intelligence (CS 188), Machine Learning (CS 189), LLM Agents (CS 194-196).