education

University of California, Los Angeles September 2024 - June 2028

B.S. in Physics, Data Science Engineering Minor; GPA: 3.875

Relevant Coursework: Probability & Statistics, Stochastic Processes, Linear Algebra, Multivariable Calculus, Differential Equations, Discrete Math, Machine Learning, Data Mining

awards & honors

USA Physics Olympiad (USAPhO) Semifinalist (2023) - top 8% of 5,173 competitors
IMC Prosperity 4 (2026) - top 8% of 18,803 teams
ACM ICPC Break the Binary Medalist (2025) - 3rd place of 25 teams
Sigma Pi Sigma Physics Honor Society
ACM AI Projects Director
Jane Street Puzzle Solver

experience

Amazon • Software Engineer Intern June 2026 - present

Seattle, WA

Architect a distributed ML experiment-orchestration platform (AWS Step Functions) that compiles declarative configs into fault-tolerant DAGs of multi-node GPU workloads, running unattended multi-day pipelines.
Design an order-independent checkpoint-selection algorithm and parallelize evaluation fan-out to overlap with in-flight training; cut end-to-end runtime by >50% and inter-stage handoff latency from hours to seconds.
Build a config compiler that deterministically deduplicates state machines, cut experiment setup from ~30 minutes to <1 minute, and streamline research iteration with agentic failure recovery.

UCLA Scalable Analytics Institute • Machine Learning Researcher January 2026 - present

Los Angeles, CA

Restructure a 3.9 TB PET/CT archive into a longitudinal panel of 80 subjects × 4 timepoints, and evaluate 4 pretrained encoders. Forecast subject-level embeddings one timestep ahead under leave-one-subject-out cross-validation, beating a random-walk benchmark on 66% of held-out transitions while avoiding look-ahead bias.
Build a vision-language model, injecting scan and forecast embeddings as image tokens into a LoRA-finetuned 8B language model with joint regression and classification heads; predict disease outcome 8 weeks ahead at 0.76 AUC / 72% vs. 0.58 / 50% scan-only under matched-budget ablation (10 LOSO runs). Submitting to ICLR.

Scale AI • Machine Learning Intern January 2025 - January 2026

San Francisco, CA (Remote)

Built statistical evaluation rubrics and Dockerized agentic RL evaluation environments, supporting industry-standard AI benchmarks (e.g., Aider LLM Leaderboards) across 200+ datapoints and 6 clients.

skills & interests

Technical: C++, CMake, Python, NumPy, statsmodels, pandas, SciPy, PyTorch, R, SQL, Linux
Languages: English (native), Mandarin (fluent)
Interests: Skiing, tennis, jazz piano, competitive archery, game speedrunning (record holder), poker

quantitative projects

Perpetual Futures Funding Carry Strategy May 2026 - present

Personal Project

Backtest a funding-rate carry strategy on BTC and SOL perpetuals; find the edge survives up to 0.04% of ADV ($210K notional) before modeled slippage exceeds funding income. Achieve 0.8 Sharpe net of costs at optimal size vs. 2.1 gross.
Build a C++ order book reconstruction and slippage simulation engine processing 400M+ historical events over a 12-month walk-forward period, decoupling market-data ingestion from simulation via a lock-free queue.
Calibrate a square-root impact model via regression; gate entry on rolling Bayesian AR(1) funding persistence.

S&P500 Return Forecasting October 2025 - December 2025

ACM AI, Hull Tactical Kaggle Competition

Ensembled daily-horizon forecasting models for S&P500 forward returns using gradient-boosted trees, feed-forward networks, and LSTMs. Achieved 1.48 Sharpe over a 6-month scoring period; top 22% of submissions.
Scaled training across GPUs via gradient checkpointing and 8-bit optimizers, cutting training time by 60%.

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