Blog

Retail: semantic catalog search and shelf computer vision

Semantic Search for Gourmet Food and Drink: Pairing and Dietary Queries

Semantic search for gourmet food handles pairing and flavour queries, with allergen and dietary constraints kept as exact, grounded filters - not guesses.

Semantic Search for Toys: Age, Interest, and Developmental Queries

Semantic search for toys reads age, interest, and developmental intent, with age suitability and safety labelling kept as grounded filters, never guessed.

Semantic Search for Office and MRO Supplies: Reorder and Bulk Intent

Semantic search for office and MRO supplies is precision-first - reorder intent, exact matches, equivalents for discontinued items, and pack-size awareness.

Semantic Search for Automotive Retail: Vehicle, Trim, and Feature Queries

Semantic search for automotive retail treats vehicle, year, and trim as hard fitment filters, then ranks the descriptive rest - a wrong-fit part is a failure.

Semantic Search for Gifting: Recipient, Occasion, and Budget Queries

Semantic search for a gift finder extracts recipient, occasion, and budget from a query and ranks across categories by fit to the person, not keyword match.

Semantic Search at Marketplace Scale: Millions of SKUs, Thousands of Sellers

At millions of SKUs, semantic search outgrows a single Postgres instance - a dedicated vector store, ANN tuning, and a CDC re-embedding pipeline take over.

Semantic Search for a Small Catalog: When It Is Overkill

For a small catalog, semantic search is often overkill - good synonyms, clean filters, and a tidy attribute set usually beat an embedding pipeline.

Semantic Search for Books and Media: Theme, Mood, and Read-Alike Queries

Semantic search for books and media handles theme, mood, and read-alike queries, where the text is rich but the structured metadata is thin.

Semantic Search for Electronics Retail: Spec Queries and Comparison Intent

Semantic search for electronics retail turns spec queries into structured filters and handles comparison intent, so "32GB RAM under 1000" returns real matches.

Semantic Search for a Multi-Seller Marketplace: Ranking Across Sellers

Marketplace search ranking has to group the same product across sellers, weigh seller quality, and balance relevance against fairness and take rate.