Meridian Line's data science group has already produced and validated its own embedding vectors for the bill-of-lading corpus with a model they host themselves, and those vectors sit in a column of the Delta table. They now want a Vector Search index over that table. What should they configure?

A Delta Sync index can use Databricks-computed embeddings, where you provide a text column and embedding model endpoint and Databricks embeds documents and incoming queries, or self-managed embeddings, where you provide an existing vector column. Since Meridian Line has already validated its vectors, configuring a self-managed-embeddings index to index that column preserves its work; queries must be embedded with the identical hosted model before search. Configuring Databricks-computed embeddings while naming the vector column confuses text input with precomputed vectors: that mode takes text and invokes a model endpoint, rather than using an existing vector column. Dropping the column and asking Databricks to recompute is unnecessary, and the claim that self-managed vectors are supported only by Direct Vector Access is false: they are supported on Delta Sync indexes as well. Keeping Databricks-computed embeddings alongside the self-managed column for fallback also fails because the index has one vector space and no such fallback mechanism. Exam caveat: document and query vectors must come from the same model; changing models makes old and new vectors incomparable, so rebuild the index. Practical check: verify that the index selects the existing vector column and that the query path uses the same model that generated it.