TasteBrain Foundations
The global geometry of taste
A transformer-based, multimodal model for understanding human taste and aesthetic preference. Foundation-derived from Qwen-Omni, trained on ~20M curated product images across ~7M products.
128-dimensional multi-vector embeddings. Lightweight per-user taste models. Weekly-refreshed crawling across retail, restaurants, music, and culture.
The Model
Qwen-Omni backbone with fine-tuned attention layers, custom projector, and 128-dim multi-vector embedding space.
SVK Primitives
Stateless API primitives — Prism and Shopkeep — for cross-modal search, sentiment reranking, and brand matching.
Persona Services
Per-user taste models, Windows, Sentiments, Tasteboards, context ingest pipelines, and retailer intelligence.
Example Applications
InstaShop
Instagram aesthetic analysis mapped to curated product sets.
insta.bestomer.comStreams
Infinite scroll product discovery with image upload and personalization.
streams.bestomer.comBookTaste
Literary prose mapped to aesthetic product matches.
ishmael.bestomer.comDress Code
Wedding event parsing into complete looks and shoppable categories.
examples/dresscodeThe SVK is optimized for LLM-assisted development. Install the Claude Code plugin and start searching in minutes.
Start building