(04) Work experience
Financial Modeling Intern
Production-grade quantitative research stack for measuring and forecasting post-earnings announcement drift across equities and the options surface. Combines a DuckDB → Arrow → C++/CUDA compute pipeline over ~1 TB of OptionMetrics data with a leakage-safe, walk-forward LightGBM quantile-regression predictor served from versioned model artifacts. Python 3.11 · C++17 · CUDA · Arrow/Parquet · DuckDB · LightGBM · WRDS (CRSP / Compustat / IBES / OptionMetrics). A statistical research project that estimates the empirical conditional probability of Bitcoin's forward price move by matching its current price-to-moving-average state against every comparable day across 16 years of history, validated out-of-sample and served through a React + FastAPI dashboard.
Software Engineering Intern
A local Model Context Protocol (MCP) server, written in Python 3.11+, that lets AI skills self-improve from ordinary user feedback. It captures soft preferences and hard constraints, then reapplies them automatically at generation time. A Claude Agent SDK harness that turns CSV data and natural-language prompts into publication-ready data-visualization tables through Posit's great_tables library and headless Chrome rendering.
Data Engineer
A production-grade, multi-tenant Next.js 14 web platform built for Texas A&M MEEN research teams to organize additive manufacturing (AM) experimental campaigns as Builds, capture data across a nine-phase materials pipeline (powder → fracture mechanics), version artifacts in Supabase Storage, and explore results through interactive 2D/3D visualizations — replacing ad-hoc folders of CSVs, STLs, and TIFFs with a role-governed, queryable single source of truth.
Projects Officer
A machine learning system built for American Airlines Flight Operations that predicts OE (Operating Experience) sequence spoilage risk using structured scheduling data and NLP-extracted coordinator comment features, enabling proactive reassignment to reduce pilot training waste across 9,025+ sequences. A heat-index forecasting project that compared 4 TensorFlow model architectures and produced a stacked ensemble for a 15-member team. A Python and Streamlit airport-network analysis platform using BTS data, Leiden community detection, and hub/bridge scoring to simulate changes in network volatility for a 25-member team.
Machine Learning Researcher
Audio Gorilla is a Python research benchmark for training language models to translate natural, multistep, and multiturn speech into executable API sequences. The project combines stateful service simulators, synthetic backend generation, structured ground-truth validation, comprehensive API testing, multi-provider speech synthesis, acoustic augmentation, and voice cloning and humanization experiments.
Quality Assurance Intern
Quality assurance and SQL data-reporting work supporting cross-team operations in Agile/Scrum delivery cycles.