Applied AI Lead · Principal ML Scientist · Head of AI Engineering
Twenty years building production ML systems and, for the last three, GenAI systems end to end. Former Senior Applied Scientist at Amazon and Research Scientist at Google; twice the technical founder of an AI-native product taken 0→1. Deep NLP and LLM engineering — retrieval, agentic pipelines that rank and explain their own output, and the evaluation harnesses that keep them honest — paired with retail, CPG and marketing-data fluency. Alongside it I run a registered advisory firm as principal, where the applied-mathematics side stays load-bearing against live capital. M.S. Applied Mathematics.
Six engagements, in detail
Every figure below is tagged, because they are not the same kind of claim: realized it happened and the money or the time moved · identified found and quantified, capture was someone else’s to execute · validated measured against a held-out or human-labeled set. A page that blurs those is a page you cannot trust twice.
Twenty years, condensed
A.Team
Principal-level engagements for Fortune 100 and enterprise clients across CPG, retail and finance; fractional and concurrent by design, directing teams of up to three. AB InBev/Ambev (2025), Mercer Advisors (2024), and Mudflap — first hire on the data science function, Snowflake stood up from zero with adoption driven across non-technical teams, Amplitude experimentation integrated, and reinforcement-learning dynamic pricing across 100,000+ drivers.
Yesora.ai
First technical hire. Built the AI-native operating layer reconciling fragmented ERP, CRM, EDI, portal and spreadsheet data into ranked, explained, owner-assigned actions. Launched to paying customers on recurring contracts.
Neo Health
End-to-end agentic consumer-health platform on AWS and GCP unifying wearables, labs and records. Hired and led six engineers. Established the evaluation discipline that benchmarked the system 10% above OpenAI's baseline.
AEOM Investment Advisors LLC — registered investment adviser, Series 65
A quantitative practice run alongside consulting: a systematic trading system operating unattended against live capital, walk-forward and Monte Carlo backtesting, volatility-regime classification, and portfolio construction blending mean-variance optimization with scenario forecasts. Plus a raw land fund as lead partner in a four-investor partnership. No performance figures are published here — the SEC Marketing Rule governs performance claims for a registered adviser.
Amazon — Fashion Tech & Consumer Electronics Tech
Personalization and recommender R&D at nine- and ten-figure revenue scale. Founded a cross-organizational tiger team serving all of Amazon Retail. CETech Tech Star award — top engineer among roughly 100 peers.
The Trade Desk
Led the three-person Boulder data science team on real-time bidding optimization at auction scale: targeting, bid optimization, performance forecasting and fraud detection.
Royalty Analytics
S&P 500 and startup clients across energy, finance, telecom and cybersecurity. CVR Energy geospatial fleet optimization; reinforcement-learning models for high-frequency futures trading at Northfield Trading; user engagement optimization for Time Warner Cable; Bayesian cybersecurity risk estimation for Dexa Systems.
Developed the patented algorithm reducing display-ad over-delivery, improving inventory management by over 30%.
What I actually reach for
GPT/OpenAI, Claude, agentic workflows and tool use, RAG, embeddings and vector search, prompt and context engineering, structured output and function calling, evaluation harnesses and benchmark design, guardrails and safety constraints, generative image pipelines, latency and unit-cost optimization.
Topic modeling, semantic clustering, signal/noise filtering, ranking and re-ranking, recommender systems, RNN-LSTM and transformer architectures, causal inference, reinforcement learning, active learning, time series and forecasting, large-scale and geospatial optimization, Bayesian methods.
Python (expert), SQL, R, PyTorch. AWS full stack (RDS, DynamoDB, S3), GCP (BigQuery, event tracking and analytics), Snowflake, PostgreSQL. Batch and streaming pipelines, messy external API integration, schema-drift handling, data cataloging.
Systems architecture, production code review, hiring and team building, technical roadmapping, executive communication, translating business goals into technical requirements. Speaker: “Algos, AI, and Humans: Where Each Belongs in a Quant Stack.”
and the rest
University of Colorado Boulder, 2006. Focus: time series, statistics, machine learning. Teaching assistant for Calculus III, Differential Equations and Linear Algebra.
University of St. Thomas, Houston, 2004. Magna Cum Laude; Outstanding Student of the 2004 graduating class.
Tecarro, Van Rooy & Obeyesekere, Computational studies of a cell cycle model (2006).
The Firm — an operating model for running several coding agents on one codebase: five roles, a labelled board, and seven rules about authority and evidence. Extracted from a project it still runs.
50+ public repositories at github.com/greentheo, including RTDAmeritrade, an R package for the TD Ameritrade API. The side projects on this site are the current ones.