Taste is difficult to capture because it's expressed indirectly. A shopper rarely states "I prefer structured silhouettes in muted tones" — but their likes, skips, and purchases reveal exactly that pattern over enough interactions. An AI taste model's job is to extract this pattern from behavior rather than requiring shoppers to articulate it. Taste models improve with more data and more diverse interaction types. A model built only from purchases is thinner than one that also incorporates likes, skips, saved items, and outfit-level feedback (not just individual products). Elara's Style Graph is a taste model that specifically incorporates negative signal (skips) alongside positive signal, which produces a more precise profile than purchase history alone.