databricks-vector-search

To set up Databricks vector search, you create a STANDARD or STORAGE_OPTIMIZED endpoint using the provided patterns. Endpoint creation configures the compute infrastructure needed to host vector indexes and process similarity queries over Delta Lake data.

Delta Sync indexes automatically synchronize with underlying Delta Lake tables in a managed or self-managed workflow. Direct Access indexes require manual data ingestion but offer more control over the embedding pipeline and vector storage updates.

Yes, you can compose text, vector, and hybrid search queries with optional filters. These query patterns allow you to restrict semantic retrieval results based on specific metadata columns within your Delta Lake tables.

Building a RAG pipeline involves creating a vector search endpoint and a Delta Sync index over your documents. You then query the index with embeddings to retrieve and rank relevant context before passing it to a large language model.

Not necessarily. Delta Sync supports both fully managed and self-managed workflows. Managed sync automatically updates the vector index as your Delta Lake data changes, while self-managed workflows give you control over the synchronization schedule.