Vertex AI Search and Agent Builder power enterprise RAG and grounded AI applications on Google Cloud. These exercises cover grounding, data stores, extractive content, and the fully managed RAG pipeline.
0 / 5 completed
1 / 5
At standup, a colleague asks what grounding with Google Search does in Vertex AI. What is the correct answer?
Grounding with Google Search augments a Vertex AI model response with live web search results. The model retrieves relevant search snippets and uses them as context, then cites sources in the response via a search_entry_point object. This reduces hallucinations on recent or niche topics where the model's training data may be stale or incomplete.
2 / 5
During a PR review, a teammate asks what a search_entry_point object contains in a grounded Vertex AI response. What is accurate?
The search_entry_point object in a grounded Vertex AI response contains a rendered search widget — an HTML snippet and inline CSS — that you must display to the user when showing grounded responses, as required by Google's grounding usage terms. It visually attributes the search queries made during grounding. Displaying this widget is a contractual requirement, not optional.
3 / 5
In a design review, the team discusses data stores in Vertex AI Agent Builder. Which statement best describes them?
A data store in Vertex AI Agent Builder is a managed document index. You ingest content — websites (via sitemap or URL crawl), Cloud Storage documents (PDFs, HTML, JSON), or BigQuery tables — and Vertex AI indexes them. Search apps query one or more data stores to retrieve relevant chunks for grounding or RAG. Each data store has its own ingestion pipeline, schema, and refresh schedule.
4 / 5
An incident report shows poor answer quality from a Vertex AI Search RAG setup. A senior engineer asks what extractive answers vs extractive segments are. What is correct?
Vertex AI Search returns two types of extractive content alongside search results. Extractive answers are short verbatim spans (typically one or two sentences) most likely to directly answer the query. Extractive segments are longer verbatim passages (up to a paragraph) providing more surrounding context. Both are extracted directly from the source document without paraphrasing, making them useful as RAG context chunks for an LLM.
5 / 5
During a code review, a senior engineer asks what RAG on Vertex AI Agent Builder requires compared to building RAG from scratch. What is the key difference?
RAG on Vertex AI Agent Builder is a fully managed pipeline. You ingest documents into a data store and the platform handles chunking, embedding, and indexing automatically. At query time, you call the answer API (or integrate via grounding), and Vertex AI handles retrieval and context injection into the model. This contrasts with DIY RAG where you manage embedding models, vector databases, chunking strategies, and retrieval logic yourself.