Meridian Line's data science group produces its own embedding vectors for crew-report text using a specialized multilingual model that runs outside Databricks, and wants to deploy a Vector Search index that simply stores and serves those pre-computed vectors without Databricks re-embedding anything. Which index type and deployment approach fits?
Select an answer to reveal the explanation.
Short Explanation
When you already have the vectors in hand, you don't want an index that insists on computing its own — you want the empty container you can push into directly. That's Direct Vector Access, populated by explicit upserts, with no Databricks embedding step in between.
Full Explanation
Direct Vector Access is the index type designed for exactly this case: vectors are computed outside Databricks (here, by a specialized multilingual embedding model run externally) and pushed into the index via explicit API upsert calls, with no Delta table sync or Databricks-side embedding computation involved. This gives the team full control over the embedding process while still getting Vector Search's serving and query capabilities. A Delta Sync index with Databricks-computed embeddings enabled does the opposite of what is requested — it has Databricks generate embeddings from source text in a Delta table, which would discard the team's already-computed, specialized multilingual vectors and force a different (likely lower-quality for this use case) embedding model. A Delta Sync index with a Change Data Feed trigger to 'recompute embeddings on every sync' misunderstands both concepts: Change Data Feed is a Delta table change-tracking prerequisite for sync, not an embedding recomputation trigger, and this option still ties the index to Databricks-side embedding rather than accepting pre-computed vectors. Serving vectors through a Model Serving endpoint as if they were model outputs is a category error — Model Serving hosts models that produce inferences on request, it is not a vector store or similarity-search index. For the exam, the signal for Direct Vector Access is always 'vectors are produced elsewhere and pushed in,' as opposed to Delta Sync's 'Databricks reads a Delta table and keeps the index current automatically.'