From pioneer-era modeling to multimodal AI.
The last mile is the only one that matters.
I started building predictive models when the industry was still figuring out whether data could drive decisions at all. The tools were primitive. The frameworks had to be invented. The work was slow and the validation was manual. That era taught me something most people practicing analytics today have never had to learn: the model is not the answer. The decision the model enables is the answer.
Thirty years later, the compute has changed beyond recognition. The principles have not. Prediction, validation, framework design, and the discipline of testing your work against reality — those fundamentals are as true in a multimodal AI pipeline as they were in a regression on punch card data.
What I bring to any engagement is the full arc: the rigor of an era when analytical frameworks had to be built from first principles, the fluency of someone who has shipped a production AI platform from concept to deployment, and the judgment that comes from decades of sitting across from C-suite executives who needed to act on what the data was telling them.
I take on a small number of engagements — short-scoped projects and selective advisory work — where the analytical challenge is genuinely interesting and the client is serious about using the output to make better decisions.
Co-founder and serves as Chief Strategy Officer for a multimodal AI platform for brand safety classification and optimal ad placement based on a core taxonomy pairing for content TV. Conceived the product thesis, designed the Core Taxonomy (92 emotions × 189 contexts × 28 media genres), architected the IE Score methodology, and participated in full-stack production deployment — V1 through V8 on Railway, Celery, PostgreSQL, and LLM multimodal AI.
The IE Score Engine applies behavioral science principles to a weighted scoring model (scene-by-scene Emotion, Context, and Genre) that enables premium inventory identification at scale. GARM/IAB brand safety standards and proprietary brand suitability are integrated across 80+ categories. The system is live and production-deployed.
This is the most complete proof point of what it means to translate a behavioral science framework into a commercially deployable AI product — from taxonomy whiteboard to Railway container to revenue generation.
A validated decision framework for predicting NFL game outcomes — not a prediction tool, but a systematic methodology. Includes the Fragility Index (defensive depth, OL sustainability, coaching conservatism, explosive play vulnerability), the Reversibility Index, and a Red Flag Analysis with quantitative score adjustments.
Validated retrospectively with an 86% improvement metric. The model illustrates how analytical frameworks built for one domain apply universally: the same framework thinking that predicts fourth-quarter lead collapses applies to organizational risk and market positioning.
Request methodology overview →Designed and built an AI-powered analytics reporting application integrating large language model APIs with a custom data pipeline. Demonstrates the implementation side of AI strategy: not just the recommendation, but the working system.
Built independently, from architecture to deployment. The application automates analytical narrative generation from structured data outputs — turning model results into executive-ready reports without manual interpretation.
Learn more →The current AI conversation treats 2024 as year zero. It is not. The principles that make an AI system useful — rigorous problem framing, validation against reality, the discipline of knowing what your model cannot tell you — have been true since the first regression ran on a mainframe. The delivery mechanism changed. The hard part did not.
This is a practitioner's perspective on what has actually changed, what hasn't, and what most organizations are still getting wrong.
Read ArticleMost mornings I use AI the way I would use a sharp colleague — not for quick answers, but for the question I haven't asked yet. This is an honest practitioner's account of what serious AI collaboration actually looks like, including the parts that don't work.
The tool does not change the discipline. The discipline changes what the tool can do. That distinction is the difference between someone who has used AI for a year and someone who has used it seriously.
Read ArticleThe Four-Pass Model — built to predict NFL game outcomes — is not really about football. It is about what happens when you encode thirty years of pattern recognition into a disciplined framework and then test it against reality. The 86% improvement metric is the part people notice. The methodology is the part that matters.
A practitioner's case study in how analytical frameworks built for one domain apply universally — and what the process of building them reveals about how decisions actually get made under pressure.
Publishing July 2026Advertising has always known that emotion drives purchasing behavior. The industry has spent decades trying to engineer around the fact that emotion was too expensive and too variable to measure at scale. Contextual targeting was a proxy. Brand safety scoring is a proxy.
Multimodal AI makes direct measurement possible for the first time. This is what that actually means commercially — and why most of the current "AI in advertising" conversation is focused on the wrong problem.
Publishing August 2026I played football in high school and college — that is the college gave me a prediction more urgent than anything my model has produced since. The inside experience of the game shaped the Four-Pass Model in ways no algorithm could replicate. The same principle built IntellEmotion™.
Two frameworks. Thirty years apart. One lesson about what domain expertise means when encoding it into an AI system — and the single question every brand should ask before selecting an AI marketing platform.
Publishing September 2026AI is powerful. Extraordinarily so. But the narrative spreading across LinkedIn — that experience matters less than ever, that the playing field is leveled — confuses the instrument with the musician. Thirty years of building models teaches you things no training dataset captures.
A practitioner's case for why domain expertise isn't diminished by AI. It's amplified by it — and why the gap between those with only tools and those with tools plus experience is wider than it has ever been.
Publishing October 2026"Agentic AI" has become the phrase of the moment — used largely by people who have never built one. This is a practitioner's guide to the architecture decisions, failure modes, and discipline required to make multi-agent systems work in production, not just in a demo.
Covers the Orchestrator-Specialist model, state management, incremental build sequencing, and the five failure modes every production agentic system will encounter — and how to plan for all of them before launch.
Publishing November 2026For most of my career, work came to me by reputation. When I decided to wind down my active practice, companies still called — explained their challenge — and then told me they hoped I found it interesting. That dynamic has shaped how I work.
I am not available for every engagement. I am available for problems where the analytical challenge is real, the organization is serious about acting on the output, and the work has a defined beginning and end.
The right engagement is one where, six weeks in, you have a decision framework you didn't have before — one you can explain, defend, and build on. Not a dashboard. Not a report. A framework.
The Calendly link below takes fifteen seconds. It asks for your name, your company, and the challenge you're trying to solve. That's the vetting step — for both of us.
If it looks like a fit, we schedule thirty minutes. If thirty minutes confirms the fit, we discuss an engagement. If not, you'll leave with a clearer picture of the problem than you came in with.
Note: I do not respond to vendor outreach, software sales, or SEO inquiries through this form.