education

University of California, Los Angeles Graduating Spring 2027

B.S. in Physics, Data Science Engineering Minor

Relevant Coursework: Data Structures & Algorithms, Computer Architecture, Multivariable Calculus, Linear Algebra, Differential Equations, Modern Physics

experience

Scale AI • Generative AI Intern January 2025 - present

Contribute to training and evals for SOTA reasoning and agentic models across 5+ clients and 100+ tasks, ranging from abstract visual reasoning, to deep research studies, to next-gen SWE agents solving real world GitHub issues.
Execute rigorous quality assurance reviews for critical datasets powering SWE agents in Java, C++, Python, Go, and Rust, evaluating intern deliverables against customer specs to ensure data integrity. Improve data quality rating by organizing and leading in-person project workshops for contributing interns.
Construct robust Docker testing environments and write comprehensive rubrics for industry standard AI benchmarks such as the Aider LLM Leaderboards.

ACM AI @ UCLA • Projects Officer January 2025 - present

Design and lead student projects exploring advanced AI topics.

skills

Languages: C++, Java, Python, Swift, SQL, HTML/CSS, R
Frameworks: PyTorch, Hugging Face, React, Jekyll
Developer Tools: Git, Cursor, VS Code, XCode, Jupyter

projects

Kaggle S&P500 Prediction • ACM AI October 2025 - Present

Design and iterate neural networks, such as decision trees, feed-forward networks, and Long Short-Term Memory (LSTMs) models to predict forward returns of S&P 500 for Kaggle competition.
Improve data preprocessing and model evaluation by implementing KNN imputation for dataset NaN values and k-fold cross validation. Competition scoring in progress.

R1 Reasoning • ACM AI March 2025 - June 2025

Implemented reinforcement learning from human feedback (RLHF) system using Group Relative Policy Optimization (GRPO) to fine-tune Qwen2.5-7B-Instruct model for mathematical reasoning tasks. Improved out-of-the-box model accuracy on test data by 17 percentage points.
Identified and debugged issues with repetition rewards, correctness metric, and dataset parameters. Implemented custom correctness checking via regex pattern matching and repetition detection using n-gram analysis.
Optimized memory usage through gradient checkpointing, 8-bit optimizers, and automatic GPU memory management for multi-GPU training.

publications

Chen, Y., Jiao, J., & Zheng, R. (2024). Exploring changes in trip generation and impacts of built environment between regular and essential trips: A study based on the contiguous United States. Proceedings of the CICTP 2024 (pp. 3317–3326). Presented at the CICTP 2024. https://doi.org/10.1061/9780784485484.314

ryan zheng