• Built a financial vector database leveraging text embedding techniques on extracted financial datasets, enabling efficient retrieval for investment modeling, credit risk analysis, and advanced machine learning applications.• Implemented RAG with LLMs to automate financial insights extraction, generating high-quality earnings call highlights and analytical reports, boosting investment strategy optimization.• Deployed LLMs on AWS SageMaker, enhancing scalability for credit scoring, sentiment analysis on market news, and real-time quantitative computations.• Designed a hybrid coroutine/thread state machine framework to process trading inputs and market queries concurrently, achieving a 3x acceleration in computation speed.• Engineered a parallel extraction algorithm optimized for multi-core clusters, reducing financial data processing time by 6x for large-scale datasets.• Debugged and deployed production-quality Python code, leveraging CI/CD pipelines for high reliability and scalability.

• Deployed and fine-tuned Qwen and GLM-4 models in a local environment, processing millions of text entries from social media, financial news, and forums. Conducted high-precision sentiment analysis and mapped the correlation between sentiment indices and stock returns, providing robust support for financial analysis and investment decisions.• Engineered a new Python genetic programming package to replace traditional packages like gplearn, expanding the structure of time series data and operators. The custom-designed algorithm framework addresses the balance between high performance and expressiveness, making the package excel in handling complex time series data.• Implemented the NSGA-II algorithm for multi-objective genetic programming, overcoming traditional single-objective and linear weighted multi-objective limitations. This algorithm facilitates bi-objective optimization, providing optimal performance at each cost level and yielding comprehensive alpha mining results.• Accelerated the runtime of genetic programming simulations by leveraging PyTorch with NVIDIA CUDA for GPU acceleration and Numba with JIT compilation, outperforming the naive implementation by 85 - 110 times.• Achieved 19.6% average annualized excess return in backtesting (2022-2024), validating the robustness of multi-objective genetic programming models in financial strategies.

MOOC (Massive Open Online Course) Data Analysis• Compiled the anonymized MOOC data from multiple sources in ND system, including homework submission record, video watching log, Q&A participation history, and prior course survey• Performed the feature engineering to prepare the modeling dataset• Analyzed the data via multiple data analytics techniques, e.g., ANOVA to evaluate difference between classes; K-mean to cluster unobserved learning styles; and random forest to classify final grade categories. Time Series Forecasting for Class Grade• Constructed data frames for ARIMA time series modeling• Evaluated the modeling assumptions by multiple criteria, including stationarity, autocorrelation function (P/ACF), BIC (Bayesian Information Criterion), and MAE (Absolute Error)