Loom uses hybrid AI to recommend outfits with 3.3x better scoring

Loom is a hybrid outfit recommendation system that bridges the gap between purely learned embeddings and rigid rule-based approaches. Given an anchor clothing item, it first retrieves complementary pieces using slot-constrained approximate nearest neighbor search over FashionCLIP embeddings. Then, it scores candidate outfits through a multi-objective function that weighs six distinct signals: embedding similarity, color harmony, formality consistency, occasion coherence, style direction, and within-outfit diversity. This two-stage retrieval-scoring pipeline allows Loom to generate complete, coherent outfits without exhaustive search.

The system introduces two key innovations. Semantic material weight leverages the geometry of CLIP embeddings to infer garment "heaviness" (e.g., denim vs. silk), enabling layer compatibility without hand-coded material taxonomies. Vibe/anti-vibe occasion priors embed prose descriptions of occasion contexts (e.g., "casual brunch") as anchor vectors in CLIP space, scoring items by differential affinity. Ablation experiments on a 620-item catalog show Loom achieves a mean outfit score of 0.179 with a 9.3% hard violation rate, compared to 0.054 and 16.0% for a category-constrained random baseline — a 3.3x score improvement and 42% reduction in violations. Direction reranking proved indispensable: removing it drops the score to 0.052, essentially random. The system generates three stylistically distinct outfits in under 5 seconds on commodity hardware.