CV
Tong Zhu
Summary
UCLA PhD student focused on LLM alignment, multi-agent systems, and mechanism design, with a paper accepted at NeurIPS 2025. Industry experience in AWS software engineering at Amazon and applied machine learning at ByteDance.
Education
- Doctor of Philosophy (PhD) in BiostatisticsUniversity of California, Los AngelesCourses: Uncertainty in LLMs, Agent-based AI
- Master of Science in Computer Science2024-05-01Northeastern UniversityGPA: 4.0Courses: Algorithm, Distributed Database
- Master of Science in Statistics and Operations Research2020-07-01University of North Carolina at Chapel HillCourses: Applied Statistics, Machine Learning, Time Series Forecasting
Work Experience
- Software Engineer Intern2024-06-01 - 2024-08-01Amazon
- Built features for Amazon B2B on AWS to automate event-driven workflows using EventBridge
- Enhanced the Visibility Service using Java, JavaScript, and TypeScript to support real-time monitoring of business transactions
- Developed and executed integration tests for outbound services and events to validate service behavior before deployment
- Designed and validated CloudWatch dashboards to monitor key service performance metrics
- Data Scientist2021-02-01 - 2022-08-01ByteDance Ltd.
- Applied XGBoost to evaluate approximately 30 candidate product features, identifying approximately 5 key features associated with order volume and daily user activity to guide product development
- Partnered with cross-functional teams to deploy a dynamic subscription tool to improve customer experience
- Applied interrupted time-series analysis to estimate potential revenue impact and assess risks of restrictions on creators' quoting strategies
- Used attribution analysis to evaluate approximately 10 proposed marketing campaigns, estimating expected revenue to prioritize approximately 3 campaigns for launch
- Data Scientist2020-08-01 - 2020-12-01Blingby
- Built data ETL pipelines through Apache Spark to transform raw data into features by combining business sense and statistical knowledge
- Developed, maintained web-based dashboards with Tableau to update daily data analysis report, which increased 20% daily work efficiency
- Machine Learning Intern2020-06-01 - 2020-08-01TouchSuite
- Queried and cleaned terabyte-sized order data from Azure SQL using pyodbc
- Conducted online analytical processing (OLAP) to display critical sales performance from different dimensions
- Developed item-based approaches to handle cold-start problems and tuned the model hyper-parameters through SparkML cross-evaluation toolbox which reduced root mean square errors by 10%
Skills
Programming Languages
- Python (pandas, NumPy, scikit-learn)
- Java
- JavaScript
- TypeScript
- SQL
- R
Tools and Platforms
- AWS (EventBridge, CloudWatch)
- Spark
- XGBoost
- MySQL
- Tableau
Publications
- Multi-Layer Kernel Machines: Fast and Optimal Nonparametric Regression with Uncertainty Quantification2024Under R&R at Journal of Machine Learning Research (JMLR)Wenlu Xu†, Tong Zhu†, Huiying Zhong, and Xiaowu Dai. [Code]
- Common-Agency Games for Multi-Objective Test-Time Alignment2026PreprintBaiting Chen†, Tong Zhu†, Rui Yu†, and Xiaowu Dai. A game-theoretic framework for multi-objective test-time alignment of language models using common-agency games. [Code]
- Mechanism Design Meets Large Language Models: Foundations and Frontiers2026PreprintBaiting Chen, Tong Zhu, Xuanang Li, Yichi Zhang, and Xiaowu Dai. A survey of the intersection of mechanism design theory and large language models. [Code]
- Incentivizing Truthful Language Models via Peer Elicitation Games2025NeurIPS 2025Tong Zhu†, Baiting Chen†, Jiale Han, Lexin Li, Gang Li, and Xiaowu Dai. Introduces PEG, a training-free framework that uses game-theoretic incentives to elicit truthful LLM behavior without ground-truth labels or fine-tuning.
- Competitive Multi-Agent Delegation For LLM ReasoningWorking paperThis paper introduces COMMAND, a competitive multi-agent delegation framework that treats LLM reasoning as a principal-agent delegation process, using competition and alignment incentives to elicit higher-quality answers.
- LightAgent: Production-level Open-source Agentic AI Framework2025arXiv preprintWeige Cai, Tong Zhu, Jinyi Niu, Ruiqi Hu, Lingyao Li, Tenglong Wang, Xin Guo, Zhaowei Liu, Dongpo Cheng, Xiaowu Dai, Weining Shen, and Liwen Zhang. A lightweight, open-source agentic AI framework integrating memory, tool use, and Tree of Thought.
Teaching
- Biostat M2572026University of California, Los Angeles, Biostatistics DepartmentRole: TAStatistical Computing — Computational algorithms for research in (bio)statistics, including numerical linear algebra and optimization.
- Biostat 4062026University of California, Los Angeles, Biostatistics DepartmentRole: TAApplied Multivariate Biostatistics — Use of multiple regression, principal components, factor analysis, discriminant function analysis, logistic regression, and canonical correlation in biomedical data analysis.
- Biostat 2312026University of California, Los Angeles, Biostatistics DepartmentRole: TAStatistical Power and Sample Size Methods for Health Research — Sample size and power analysis methods for common study designs, including comparisons of means and proportions, ANOVA, time-to-event data, group sequential trials, linear regression, cluster randomized trials and multilevel data.
- Biostat 2852026University of California, Los Angeles, Biostatistics DepartmentRole: TAAdvanced Topics: Recent Developments — Advanced topics and developments in biostatistics not covered in other courses. Possible topics include time-series analysis, classification procedures, correspondence analysis, etc.
- Biostat 203A2025University of California, Los Angeles, Biostatistics DepartmentRole: TAIntroduction to Data Management and Statistical Computing
- Biostat 2012025University of California, Los Angeles, Biostatistics DepartmentRole: TATopics in Applied Regression
Portfolio
- Portfolio item number 1
