Data Scientist focused on Sports Analytics, Local Tourism, AI Safety, and Supply Chain Optimization.
Featured Projects
Mapping humanity’s hardest issues: what’s known, the open questions, the main positions, and
where they split, so readers can find where they stand and why.
• Structured Topic Maps:
50 published topics on one six-section spine (Overview, Basics, Questions, Positions,
Crux, Actions), each readable at a glance and expanding in place, with sourced data
visualizations.
• AI Authoring Pipeline:
Five fresh-context Claude Code agents research, edit, illustrate, render and audit each
topic before a human publishes it.
• Web App: Django, HTMX
and Alpine platform with anonymous voting, leans and follow-up prompts, self-hosted in
Docker behind a Cloudflare Tunnel.
Predicting NBA game spreads and winners with deep learning models that learn directly from
play-by-play, box scores and rosters, evaluated against Vegas lines.
• Automated Data Pipeline:
Twice-daily collection of games, play-by-play, box scores, injury reports and betting
lines from the NBA API and ESPN into SQLite, followed by fresh pre-game predictions.
• Prediction Engines:
A hierarchical network (Kalman-filter player abilities, GATv2 player synergy, team and
matchup layers), a roster-aware temporal transformer, Ridge, XGBoost and a PyTorch MLP,
combined in an ensemble.
• Web App & Dashboard:
Flask app showing predictions next to Vegas lines and results, with a dashboard tracking
each model's record against the spread.
Other Work
A Warning About AI - The Control Problem, Explained for Everyone
A one-page, plain-language briefing on the AI control problem, written for readers with no technical background. Distills months of AI-safety research into a single visual poster: what AI actually is, why alignment is hard, why the race dynamics prevent anyone from slowing down, and what the failure modes and the reward look like.
(See NBA AI for newer version) Using data analytics and machine learning to create a comprehensive and profitable system
for predicting the outcomes of NBA games.
• Data acquisition architecture that
leverages Scrapy and SQLite to analyze and store data on NBA teams,
players, and games from a diverse range of data sources.
• Data modeling setup employing AutoML for
quick iteration and optimized performance.
• Public-facing web application and
dashboard to showcase predictions and results.
Predicting Yelp Review Quality
Utilizing the Yelp Open Dataset, this project predicts review quality to enhance user engagement and satisfaction. It incorporates advanced feature engineering techniques and machine learning models. Through text analysis and sentiment analysis, it offers improved insights into user behavior and drives data-driven decision-making on Yelp.
In this project, I implemented machine learning and natural language processing techniques to predict fraudulent events from transaction data. The results were visualized through an intuitive Flask web application. This project highlights my ability to transform complex data into actionable insights.