Background & experience
Curriculum vitae.
I study mathematics and computer science at HKU, working on scientific computing and machine-learning projects. This page brings together my education, research and selected work.
The English PDF is my Quant CV; the Chinese PDF is my algorithm-focused CV. You can also find the project details and research demo on this site.
Education
The University of Hong Kong
Bachelor of Science · Mathematics & Computer Science · Minor in Finance
Dean’s List · HKU Entrance Scholarship · Lee Shau Kee Scholarship
University of California, Los Angeles
Exchange study in Mathematics & Computer Science
Stanford University
Visiting study in Mathematics & Computer Science · International Honors Program
Research Experience
Sharpness-aware optimization
Current capstone · In progress
Exploring sharpness-aware optimization for machine-learning training. The work is ongoing; results are not yet published.
HKU Summer Research Fellowship
Parabolic Inverse Source Reconstruction: From Raw Error to Visible Geometry · 2026
Supervisor: Prof. Zhiwen Zhang
Selected as one of the best poster presenters and invited to present at the HKU Science Undergraduate Research Colloquium (Summer 2026).
- Study how to recover a hidden heat source from temperature measurements taken after diffusion.
- Compare recovered source values and shapes, and test reduced models that speed up repeated reconstruction.
Urban Rail Network Optimization
Independent research · 2023
Modeled rail alignment under geometric and engineering constraints using analytical optimization and Particle Swarm Optimization; recognized in the S.-T. Yau High School Science Award.
Selected Projects
H&M Two-Stage Personalized Recommendation System
Built a system that finds candidate products and ranks them for each customer, using 31.8 million transaction records. It reached 0.034479 MAP@12 on a later, untouched week and generated predictions for 1.37 million customers. Repository
DocuQuest Agent
Built a C++17 document search and question-answering tool, including the index, retrieval logic and tests that run without API credentials. Repository
Behavioral Personality Analytics
Compared 381 model and feature configurations while keeping the final test separate from model selection. A three-feature model reached 0.9651 holdout ROC-AUC on generated data. Repository
Flight Price Prediction & Pricing Analysis
Collaborative XGBoost study over 300,153 rows; contributed Experiments 1 and 2 code and Experiment 2 interpretation. Repository
Technical Skills
Languages: Python, C++, SQL, MATLAB
ML & Data: PyTorch, LightGBM, XGBoost, scikit-learn, pandas, NumPy, SciPy, DuckDB
Recommendation & Search: candidate retrieval, collaborative filtering, two-tower models, LightGCN, LambdaRank, temporal validation
Research & Engineering: Git, CMake, Jupyter, LaTeX, numerical optimization, PDEs, experiment design
Updated October 2026.