Katch · 2018 to present
Audience Genome
A recommendation and comparable-title engine over 100,000+ titles, built to work on the long tail where almost no interaction history exists.
Role · Co-founder and CTO. Architecture, ML, and product.
VisitThe problem
Content owners and distributors make eight-figure acquisition and marketing decisions on instinct and a handful of comparable titles. The underlying signal, who actually watches what and why, is scattered across incompatible datasets and rarely reaches the person making the call. Collaborative filtering does not rescue you here: the titles that most need a comp are the ones with the least behavioral data.
Approach
- 01
Built ingestion and entity-resolution pipelines that reconcile catalog, audience, and performance data across providers into a single title-level identity.
- 02
Developed the genome, a taxonomy that decomposes a title into machine-comparable attributes rather than genre labels, so a title with no audience history still has a position in the space.
- 03
Fine-tuned LLMs for content understanding and for translating natural-language questions into the analytical queries underneath them.
- 04
Shipped clustering and recommendation services that return comparable titles and audience segments with the attributes that produced them, not a bare similarity score.
Outcome
The analytical core of a profitable business, in production with Warner Bros. Discovery, AMC, Meta, Paramount, and Omnicom.