The correct answer is Vertex AI Search with managed RAG over private data. It provides a managed connection from a generative model to private repositories so answers are grounded in the latest indexed documents.

This service ingests and indexes enterprise content using built in connectors and then performs retrieval augmented generation that returns citations to source passages. It handles chunking, embeddings, relevance ranking, and access control so teams do not need to build or operate a custom retrieval stack. Because it continuously indexes the source library, responses stay aligned with the most current policies and the assistant can justify answers with links back to the original documents.

Cloud Search focuses on enterprise search across Workspace and other data sources but it is not a managed RAG solution for grounding large language model responses. It does not directly provide generative question answering with citations over your private corpus.

Build a custom retrieval pipeline with the Gemini API and a vector store can work but it is not the best way to streamline this requirement. It requires you to design and maintain ingestion, chunking, embedding generation, vector indexing, retrieval, security filtering, and citation logic, which the managed service already provides.

Fine tune a foundation model only on the policy corpus bakes content into model weights and does not guarantee use of the most up to date documents. It offers no built in citations and would require frequent retraining to keep pace with policy updates, which is not aligned with the need for grounded answers.

Full AWS Practitioner Certification Question