Create a Typesense collection with a product schema and implement hybrid vector + keyword search for ecommerce

Steps

1. Create a collection via POST /collections with a schema that includes standard string/int32 fields (title, brand, price, in_stock) and an auto-embedding field configured with an embed.from list of source fields and an embed.model_config pointing to an embedding model (e.g., openai/text-embedding-3-small); the num_dim must match the model's output dimensionality.
2. Index product documents via POST /collections/{collection}/documents (single) or POST /collections/{collection}/documents/import?action=upsert (bulk JSONL); Typesense generates the embedding vector automatically at index time.
3. Issue a hybrid search by setting query_by to a comma-separated list of both regular fields and the embedding field in a single search request; Typesense performs keyword search on text fields and vector search on the embedding field, merging results via Rank Fusion.
4. Tune the hybrid balance using the alpha parameter in vector_query (0.0 = pure vector, 1.0 = pure keyword; default vector weight is 0.3); adjust based on your catalogue's natural-language vs. attribute query mix.
5. In v30+, Curation Sets (formerly per-collection Overrides) and Synonym Sets are top-level resources; attach a Curation Set to a collection via its collection binding. If upgrading from v29 or earlier, synonym and curation data is auto-migrated but API key ACLs may need updating (synonym:* → synonym_sets:*).