Meridian's data engineer is setting up a Vector Search index over the port-circular corpus and must decide whether Databricks computes the embeddings automatically from the Delta table's text column, or whether the team pre-computes embeddings themselves and stores them as a vector column. What does this decision primarily determine?

Whether the team manages embedding generation itself versus letting Databricks manage it is correct because this is precisely the distinction between Databricks-computed and self-managed embeddings: in the Databricks-computed path, the platform generates and refreshes embeddings from a specified text column using a chosen embedding endpoint, while in the self-managed path the team runs their own embedding pipeline and stores the resulting vectors directly in a column that Vector Search then indexes as-is. Metadata filtering is available in either configuration - it is a property of the index and its metadata columns, not something gated by who computed the embedding vectors, so this option misattributes an unrelated capability to this decision. Whether the index can be queried with natural-language questions is also unaffected - both paths ultimately produce a queryable Vector Search index, and query-time semantics do not depend on which side computed the vectors. Whether the source table can be registered in Unity Catalog is unrelated to this decision as well - Unity Catalog registration is a property of the Delta table itself and happens regardless of the embedding-computation approach chosen downstream. The exam-relevant fact is that Databricks-computed versus self-managed embeddings is a division of operational responsibility for the embedding step, not a gate on other index capabilities.