Drew Boynton
Software/ML Engineer
I build production-focused ML and data products end to end, from feature pipelines and model inference to APIs, frontend delivery, and automated monitoring.
About Me
I am a software and machine learning engineer focused on building systems that hold up in production, not just in notebooks. My strongest work combines data engineering, model development, and product delivery.
Most of my recent projects are end-to-end pipelines: data ingestion, feature generation, model scoring, API layers, and frontend experiences that expose model outputs in a usable way. I regularly work with Python, TypeScript, BigQuery, Supabase/Postgres, and cloud-based AI tooling.
This portfolio highlights projects where I can demonstrate measurable outcomes, architectural decisions, and operational tradeoffs. If you are evaluating for DS/ML/SWE roles, the flagship systems are the best place to start.
Selected Work
Production-oriented machine learning and data systems are my core focus. These flagship projects represent the strongest examples of modeling, automation, and full-stack delivery.
Flagship Systems
End-to-end systems with measurable outcomes, live interfaces, and automated data pipelines.
NBA Hall of Fame Predictor 🏀
Interactive machine learning model that predicts NBA players' Hall of Fame chances with 99% accuracy. Features real-time player lookup and detailed prediction analysis using XGBoost trained on 5,250+ players since 1976. Try entering any NBA player name!
- -5,250+ historical player careers
- -Interactive probability search for any player
- -Feature-level model reasoning on each prediction
NBA Hall of Fame Predictor
XGBoost machine learning model trained on 5,250+ players from 1976-2025
Search for any NBA player below to see their Hall of Fame prediction
Search from 5,250+ NBA players • Click the screen above to see behind-the-scenes
ICTML Advanced Trading System 📈
Real-time machine learning trading system achieving 84.4% accuracy in daily market bias prediction for QQQ, SPY, and IWM. Features ensemble models, premium session filtering (9:30-12:00 EST), and daily bias probability vectors.
- -84.4% daily market-bias classification accuracy
- -Live QQQ/SPY/IWM probability vectors
- -Automated prediction refresh before market open
Loading daily bias predictions...
Live daily bias predictions for QQQ, SPY, and IWM • 84.4% accuracy • Updated daily at 9:30 AM EST • Click the screen above to see model analysis
Sports Edge: NFL/NBA Betting Analysis 🏈
Machine learning pipeline that computes model spreads and home win probabilities for NFL/NBA games, compares against sportsbook lines, and identifies betting edges. Features real-time odds integration, feature engineering (rest days, form metrics, opponent strength), and automated daily predictions.
- -BigQuery as source of truth + Supabase serving layer
- -Automated daily/weekly prediction pipeline
- -Live portfolio card pulling sportsbook deltas
🏈⚽ Sports Edge Analysis
Model predictions vs sportsbook lines
Live NFL/NBA game predictions vs sportsbook lines • Examine the spreads in the screen above to see model analysis
LLM Advisor: Agentic Trading System 🤖
Autonomous trading agent that uses Google Gemini 1.5 Flash to analyze market sentiment and adjust statistical mean-reversion thresholds in real-time. Features automated risk management, backtesting engine, and Alpaca trade execution.
- -Hybrid ML + LLM trading decision stack
- -Risk controls with drawdown guardrails
- -Backtesting workflow tied to execution rules
Additional Projects
Smaller builds and experiments that highlight breadth across data science, frontend, and algorithmic thinking.
Mancala AI with Game Theory (Try to beat the AI!)
Intelligent Mancala game implementing minimax algorithm with alpha-beta pruning optimization. The AI evaluates game states 5 moves ahead, achieving 70-80% win rate against random opponents with 10x performance improvement through pruning. Features Monte Carlo simulation analysis for strategic validation.
Advanced Data Cluster Sorting
Project for my Advanced Data Science class. This project was a individual effort to sort data into clusters based on their similarity. We used a variety of data structures and algorithms to achieve this.
CU Boulder Police Department Heatmap
A simple heatmap of the CU Boulder Police Department data and its most common location occurrences.
Simple Fitness (Tracking App!)
A native iOS app for tracking strength training and cardio workouts. Built with Swift and CoreData, this was a fun introduction to iOS development and its ecosystem compatibility. This was more a fun project just to learn more about iOS development and its language capabilities.
