Forward Deployed AI Engineer · Machine Learning
I build machine-learning systems that live inside your product —
and I can explain them to the board.
Selected Work — 2025 → 2026
Each project has a public repo, reproducible code, and — where it counts — a live demo you can click. No slides.
01 — Flagship
A physics-based ideal racing line computed on real Suzuka circuit geometry, wrapped in a 4,096-sample Monte Carlo uncertainty band and rendered as a live 3D CesiumJS experience — delta heatmap ribbon above the track, timeline scrubbing, camera tracking, WebSocket LIVE mode.
02 — Geospatial · Probability
Bayesian grid search plus ocean-drift simulation to estimate where the plane most likely is. Radar constraints, satellite handshake times, and confirmed debris finds feed a posterior over the southern Indian Ocean — drift calibrated on Réunion, validated against the Mozambique find.
03 — End-to-end pipeline
Full pipeline over 221,000 plays (2015–2025): play-by-play data cleaned to parquet, situation features into a win-probability model, then ranking the plays that actually moved games. Baseline logistic regression is fully reproducible; the tuned ensemble stays in a gitignored config.
04 — Explainable ML
Predict which B2B SaaS customers will churn — and why. XGBoost with proper cross-validation, SHAP explanations a non-technical VP can read, and per-account risk scores in a Streamlit app. It's the mirror of my career: customer success plus predictive analytics.
05 — Local LLM agents
A local LLM agent iteratively rewrites a Haas VF-2 program while a PyTorch physics evaluator checks every block against the machine's real spindle torque curve and axis acceleration limits. The loop runs until the toolpath is feasible — then keeps the fastest version it ever found, not just the last one. No cloud: Ollama serves the model locally.
About
"Finally, a people person who knows tech."
Most ML projects die in the gap between the model and the business. I work in that gap. As a forward-deployed engineer I embed where the problem actually lives — building local LLM pipelines, custom CRMs, and offline-first AI for clients who can't send their data to the cloud.
Before the models: fifteen years of operations, sales, and consulting. Running a Glass Doctor location, closing warranties at CarMax, leading B2B teams at Leadium. That's where I learned what "done" costs, who signs off on it, and why a model nobody can explain is just an expensive spreadsheet.
MBA from ASU W.P. Carey. Postgraduate data-science certificate from Caltech. Based in Las Vegas — the city that runs on probability.
Models that survive contact with production: proper cross-validation, honest baselines, explainability built in.
Local LLMs, custom CRMs, offline-first AI. I ship where the client's constraints are — not where my comfort zone is.
SHAP waterfalls a VP can read in ten seconds. Heatmaps that make a LinkedIn feed stop scrolling.
MBA, fifteen years in sales and ops. I know what "done" costs — and who has to sign off on it.
Experience
MBA · Finance, Marketing & Supply Chain Management (2017–2019)
Certificate in Data Science — postgraduate program
Deep Learning with TensorFlow · Python for Data Science · R Programming for Data Science · HubSpot Sales Software · NAASE
