AI product strategy for streaming and content intelligence. A decade of platform product work across Warner Bros. Discovery, Koch Industries, IBM and Acquia, and the founder of AirDate.TV, an AI-native TV premiere platform that has been live since June 2026.
Where I've been, what's on air now, and what's up next.
Streaming companies have more content and more data than they can make sense of. These are the places where AI turns that into decisions, and where I have shipped something you can check.
Clean, disambiguated catalog data is the foundation every recommendation and report sits on.
In production: AirDate's tagging engine pairs curated allowlists with zero-shot tagging on Bedrock, and an ID-locking layer keeps "same title, different show" collisions from reaching users.
Helping viewers find the next show, without letting the algorithm quietly narrow what they see.
In production: My Persona, an AI viewer profile behind a paid tier, plus a representation floor so Black-created and Black-led series aren't buried in discovery.
Giving development teams a faster, more consistent first read on scripts and what a project will need.
In development: The Verdict turns scripts and treatments into coverage, a success estimate and a forecast of stage and facility needs.
Predictions are only useful if someone can hold them to account.
In production: AirDate's renewal model publishes its calls before networks announce. It had Your Friends & Neighbors Season 3 at 95.1% ahead of the renewal.
I take products from idea to production myself: the strategy, the architecture and the code. It keeps me honest about what AI can actually do and what it costs to run.
An AI-native platform that tracks TV premieres before they trend. Viewers ask in plain language, like "HBO premieres in 2026" or "Taylor Sheridan", and get dates, briefings and predictions instead of a list of links.
Bedrock (Claude Haiku 4.5), Lambda, DynamoDB, API Gateway, Cognito, SageMaker, CloudFront, React
Decision intelligence for film and TV development. Upload a script or treatment and get a structured first read before money and stage time are committed.
Built for studios, production facilities and financiers. Bedrock, Step Functions, DynamoDB, React
An NBA question-answering engine. A DistilBERT model fine-tuned on a custom basketball QA corpus handles factual questions, alongside Monte Carlo playoff simulations, cross-era player comparisons and a conversational layer on Claude via Bedrock. The next phase adds vector retrieval and multi-turn dialogue.
Try HoopsIQI started as a Linux systems engineer, which is why I can sit with an engineering team and talk tradeoffs, then walk into an executive review and talk ROI. Most of my career has been platform products: the infrastructure users never see but always feel.