GCP Generative AI Leader Practice Question 67: Embeddings Vector Search RAG
Google Cloud Generative AI Leader · Test 7 · question 7 of 10
An engineering lead is designing a custom retrieval-augmented generation pipeline instead of using a prebuilt search offering. The plan is to convert a large internal document library into numerical representations that capture semantic meaning, store them so that queries can find the most conceptually similar passages even when the wording differs from the query, and pass the retrieved passages to a model as grounding context.
Which pair of Google Cloud building blocks does this design rely on?
Show answer and explanation
Answer:
A custom RAG pipeline needs two building blocks working together: an embeddings model that turns text into dense numerical vectors capturing meaning, and a vector database that indexes those vectors and finds the nearest ones to a query vector at retrieval time. The Embeddings API produces those vector representations, and Vector Search is the managed, scalable service that indexes them and performs the similarity search, together supporting semantic matches even when the wording differs from the query.
- A is wrong because full-text keyword search matches literal terms, not semantic similarity, so it would miss conceptually relevant passages that use different wording than the query. - B is wrong because translation solves a language-normalization problem, and BigQuery ML is a general machine learning toolkit in the warehouse, neither of which performs vector-based semantic retrieval. - C is wrong because Document AI only extracts text and structure from source files, and Cloud Logging just records events; neither generates embeddings or performs similarity search.
Reference: Vector Search overview