AI-103 Practice Question 20: Query Time Vectorizer
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 2 · question 10 of 10
An ingestion pipeline is working: an Azure AI Search indexer runs a skillset with the Text Split skill and the AzureOpenAIEmbedding skill, and the vector field in the index is populated correctly. The RAG application sends the user's question as plain text in a vector query and lets the search service embed it, so that the embedding model is configured in one place rather than in the app. Those queries fail; the index has no way to turn the query text into a vector.
What is missing?
Show answer and explanation
Answer:
Integrated vectorization has two halves, and this pipeline only has one. Indexing-side vectorization comes from the skillset. Query-side vectorization comes from a vectorizer, which is "defined in the index schema, assigned to a vector field, and used automatically at query time to convert a text query to a vector". The pairing is fixed: an AzureOpenAIEmbedding skill at index time pairs with an Azure OpenAI vectorizer at query time, and Microsoft warns that the vectorizer must match the embedding model used to encode the content; mixing models produces vectors that do not compare meaningfully.
- A is wrong because skills run inside the indexer pipeline during ingestion. A skillset is never invoked by a query, so no skill can embed query text. - C is wrong because semantic ranking is an L2 reranker over results that a query has already returned. It improves ordering; it does not produce the query vector that has to exist first. - D is wrong because in a hybrid query the vector half still requires a vector. BM25 would return keyword matches while the vector clause keeps failing, silently degrading the recall the team built the vector field for.
Reference: Integrated vectorization in Azure AI Search