Mosaic produces bespoke curated title sets for the specific viewer at the specific moment. It fuses a supply-side content vector with a demand-side cohort vector to compose mosaics — small, intentional sets of titles, each expressing one coherent curatorial intent.
Under active development. Engineered and demonstrated against synthetic reference catalogs — not yet proven in production. The most forward-looking of the eleven.
Curated sets that express a coherent intent — not just "more like this."
The same curation substrate across home rail, search re-rank and post-completion.
Composition that reacts to real viewer behavior and a target the business wants to grow.
Where conventional recommendation asks "what is similar to what they watched," Mosaic asks "what curatorial intent serves this viewer-moment, and which titles best express it." It is a distributor-side substrate — an integration layer between catalog, engagement signals and frontend surfaces, not a standalone app — that composes bespoke, cohort-shaped, moment-aware mosaics: bounded, intentional sets of titles, not endless rails.
A supply-side vector characterizes every title across 50–100+ dimensions; a demand-side vector resolves viewers into cohort assignments with confidence, cold-start handled structurally by cohort defaults. A desired-audience lever lets a distributor bias composition toward a cohort it wants to grow. The substrate is canonical across surfaces — home rail, search re-rank, post-completion and a live-event companion.
Six substrate capabilities turn catalog and signal into purposeful, moment-aware sets.
Every title characterized across 50–100+ dimensions — thematic, narrative, visual, audio, performer, era, audience-affinity — continuously, at ingest.
Viewers resolve into cohort assignments with confidence; individualization only at the description layer; cold-start handled by cohort defaults.
Different cohorts on the same library produce materially different mosaics; the same cohort at different moments produces different mosaics.
A distributor specifies a target cohort to grow; mosaics bias directionally in vector space, measured as cohort-aggregate growth.
Five phases — pre-game, live, halftime, post-game, next-day — compose five distinct mosaics for the same viewer and library.
A viewer’s signal in one tenant cannot leak to another — enforced at storage and validated against an adversarial test harness (in progress).
Each serve returns mosaic items plus a structured rationale, a freshness-signaling decision and an audit pointer — auditable curation, with the default a single mosaic of eight titles and a supported range of 1–3 mosaics of 6–14 titles.
One canonical substrate composes every mosaic — engineered and demonstrated against synthetic reference tenants.
Composed into the distributor’s existing experiences — no standalone app.
Editorial control over intent and composition.
Mosaics surfaced into existing end-user rails.
Phase-aware mosaics across a live event.
The same substrate, deployed to fit the distributor’s sovereignty and scale.
A delivery-agnostic boundary that consumes catalog and signal — it does not author them.
Mosaic fuses what the catalog contains with who is watching, into a set with a point of view.
Every title across 50–100+ dimensions.
Viewers resolved into cohorts with confidence.
A bounded set expressing one intent.
Demonstrated against reference catalogs — capability-grade, ahead of a first production integration.
Under active development; demonstrated against synthetic reference catalogs, with no paying distributor integrated yet.
Different audience targets produce materially different sets, and adjacent live-event phases stay distinct — measured on simulated tenants.
Mosaic is in active development. If viewer-moment curation is on your roadmap, bring us your catalog and the audiences you want to grow — and help shape where it goes.
info@vela.com · (727) 507-5300