A living map of who your users are.
The Taste Graph is Galya’s global graph, trained on human web data and queried at inference.
Archetype clusters
It computes where they sit, stylistically and semantically, relative to everything in your catalog.
Every query measures the distance between a user’s behavioral signals and the meaning encoded in your content.
The closest matches aren’t the most popular, they’re the most fitting to your user.
Galya runs two processes in parallel.
On the content side, vision-language models extract meaning from everything in your catalog: images, text, video.
Not categories or tags. Aesthetic quality, mood, emotional tone. The attributes that actually drive preference but rarely make it into a database.
On the behavior side, Galya tracks how users engage with that content (dwell time, scroll depth, saves, skips) and encodes those patterns as embeddings.
A precise representation of taste, mapped into the same geometric space as your catalog.
When a user’s taste moves in one direction, the right content or inventory moves with it.
At inference, Galya returns that as structured context in natural language, something any model or agent can act on directly.
Ready to give your AI taste?
Book a demo and see your first Taste Graph composed live, from cold-start to a callable preference layer.