Xinyuan (Ella) Li

B.S. Data Science and Business Economics | GPA 3.99/4.0 | Provost Honors, all eligible quarters

Relevant coursework: Probabilistic Modeling and Machine Learning, Introduction to Data Visualization, Data Structures & Algorithms for Data Science, Theoretical Foundations of Data Science I & II, Practice of Data Science, Intro to Data Management, Applied Statistical Data Analysis, Econometrics A/B/C, Microeconomics A, Linear Algebra, Calculus (Multivariable)

• Built and shipped a full-stack JupyterLab extension (React/TypeScript frontend, Python/Tornado backend) that embeds an AI tutor directly in student notebooks, deployed to around 400 students across 2 quarters of DSC 10.
• Designed a context-aware retrieval layer that sanitizes live notebook state (active cell, markdown instructions, and outputs) and surfaces relevant lecture examples via TF-IDF retrieval with Gemini reranking, reducing generic/off-topic tutor responses by grounding answers in course-specific material.
• Shipped exam mode, topic-based practice retrieval, and Socratic prompting constrained to course-specific APIs (babypandas), logging around 88,000 tutoring sessions to date.

• Estimated how UC faculty summer teaching supply responded to the $10,000-per-course pay cap introduced in 2019, applying regression discontinuity and difference-in-differences to 34,585 administrative course records.
• Built the full cleaning-to-estimation pipeline in Stata and ran robustness checks across bandwidths and specifications; results feed into a working paper.

• Owned an automated grade-reporting pipeline (Python, bash, Gradescope API) that computes final scores across labs, homework, quizzes, exams, and participation (handling slip days, drops, and curving), then pushes personalized reports to more than 200 students weekly, replacing an error-prone manual process and ensuring consistent, reproducible grading.
• Maintained the course website and GitHub repository, releasing weekly labs, midterm and final projects on schedule across 3 quarters.
• Led discussion sections, pod meetings, office hours, and one-on-one exam reviews for more than 200 students per quarter on regression, permutation testing, and Python in Jupyter.

• Engineered more than 50 features from 125,542 real-estate transaction records (CoreLogic Trestle API), using exploratory analysis to isolate the price drivers that shaped the final model architecture.
• Trained and tuned linear regression, random forest, and XGBoost valuation models, reducing MAPE from 20.9% to 0.85% through iterative tuning (log-transforming price, pruning low-value features), reaching a final R² of 0.86; presented results weekly to senior data scientists.
• Documented the end-to-end modeling pipeline and set reproducibility standards (environment pinning, seeded splits, and model versioning) adopted by the team.

• Built a Streamlit application that recommends optimized six-Pokémon teams from user stat preferences through a custom weighted scoring algorithm over 1025 Kaggle records; presented to judges at DataHacks 2025.

• Presented campus resources to more than 300 incoming students and synthesized post-event survey data from respondents to reshape the following year's programming.

• Planned and executed campus events for more than 500 attendees, managing a $27,581 budget across 20 events.