The audience decision system for talent-led video shows, co-founded and designed end to end

The Venture

SlateRoom is the audience decision system for companies that manage a portfolio of talent-led video shows. You give it an idea, a script, a rough cut, or a published episode; it gives back one accountable read: what will land, where attention falls away, what to change, and how confident it is. Then it holds that prediction against what the audience actually did, so the next read is better than the last. I co-founded SlateRoom in 2026 as a joint venture between SOAR Productions and Buildcraft, and I am the architect of the engine.

The Design Problem

A prediction is worthless if the person reading it trusts it too much or too little. Every SlateRoom report has to show not just a recommendation but its evidence: what the read draws on, where attention falls away, and how confident the engine is. I designed the brand (an editorial cream-and-ink system built for evidence-led credibility), the site, and the decision reports themselves: one engine serving three role views (Editor, Showrunner, Publisher), each translating the same model output into recommendations, retention curves, and scorecards a non-expert can calibrate their own judgment against.

Calibration

On the model side, I designed the four-axis calculation, built the pipeline that produces every report, and calibrated the predictions against real behavioral data: a +0.043 lift measured across a corpus of 2,400+ videos. Every claim carries a validation tier (Validated, Calibrating, or Directional), and the engine never ships a number it can't stand behind: once a show airs, the forecast is held against what viewers actually did, so every read sharpens the next.

The Research Question

SlateRoom is where my applied work became a research question: how should an AI system represent its evidence and uncertainty so a person calibrates trust correctly and stays accountable for the decision? I have lived that question here as a builder, on both sides of the model at once, and it is the question I now want to study as research. It is the clearest expression of the thing I care about most: making complex, quantitative systems feel obvious to the people who rely on them.

Co-founder Product Design Brand Data Visualization ML Calibration

Next project

Verikai Platform Redesign Product Design, UX Strategy

Forecast vs measured: predicted attention curve held against actual audience behavior

Data fuels the learning loop: video, show, slate, and application layers

Segment breakdown: four-axis scores and verdicts for every segment of the cut