Question

You are building an Azure AI Search vector index to ground a RAG agent. Documents are chunked and embedded using the text-embedding-3-large model at 3072 dimensions. During testing, indexing succeeds but every vector query returns an error and no results. You confirm the query vectors are generated from text-embedding-3-small. What is the most likely cause and correct fix?

Accepted Answer

The vector field's dimensions must match the embedding model that produced both indexed and query vectors; re-embed queries with text-embedding-3-large (3072 dimensions) so query and index vectors align.

Answer

The vector search algorithm is set to exhaustive KNN; switch to HNSW to enable dimension conversion between the two models automatically.

Answer

The index is missing a semantic configuration; add a semantic ranker so the differing vector dimensions are reconciled during scoring.

Answer

The chunk size is too large; reduce chunk size so the embeddings fit within the field's dimension limit and queries succeed.