Projects

An SQL practice platform with a 58-question practice bank plus a 42-question exam bank, covering basic syntax, joins, subqueries, aggregation, and window functions, with progress visualized through a knowledge graph.

Designed a normalized database (BCNF) for users, products, orders, and reviews. Built a responsive web frontend with HTML/CSS/JS supporting registration, login, product publishing, image upload, multi-dimensional filtering, order tracking, and admin functions. Integrated Qianwen API for LLM-powered natural language search with semantic recall and dynamic query relaxation.

Built XGBoost regression model for day-ahead and real-time electricity price prediction (MAE 51.75 and 107.76). Designed Distribution Robust Optimization strategy with Wasserstein-1 fuzzy set. Achieved three-month cumulative return of RMB 11.3184M with annualized Sharpe ratios up to 4.56.

Built a SimPy discrete-event simulation of LLM inference systems. Found optimal batch size 16-32, SJF cuts latency 37% but hurts fairness (Jain 0.87→0.09), and adaptive batching reduces spike latency by 54%.

Applied XGBoost, Transformer, VAE, and GAN for financial time series prediction (Transformer improved RMSE by 19.11% over exponential smoothing). Designed Trustworthy Online Decision Making framework with ε-tolerant OMS algorithm ensuring robustness and consistency.

Analyzed EU data space legislative framework and constructed network models for data sovereignty and compliance. Developed data management interface functionalities using Vue-Element-Admin with identity authentication and API validation.