commerce-vector-search

Hybrid search combines semantic embeddings with BM25 keyword indexing to retrieve commerce records. This balances natural language understanding with exact-match precision across products, customers, orders, and inventory using OpenAI embeddings and SQLite FTS5.

Hybrid ranking merges semantic relevance from OpenAI embeddings with exact-match scores from SQLite FTS5 BM25 indexing. This unified approach ensures natural language queries retrieve precise text matches across various commerce entity types.

You can filter e-commerce search results by applying entity-type filters to your query. This allows you to restrict hybrid search retrieval to specific commerce records such as products, customers, orders, or inventory.

Yes, generating OpenAI embeddings is required for semantic search across customer and order records. These embeddings provide the natural language understanding necessary to locate specific customer notes and order details.

You can set minimum relevance thresholds to filter vector search results. This ensures only commerce records that meet a specific combined semantic and keyword matching score are returned in the unified search experience.

BM25 indexing handles exact-match keyword precision for product search, while semantic embeddings manage natural language understanding. You should use both simultaneously via hybrid ranking to ensure accurate retrieval of products by description.