Why it matters
Vertex AI Search removes RAG infrastructure work. Understanding enables enterprise LLM apps.
Advertisement
The architecture
Data stores: web, docs, structured data.
Retrieval: hybrid (embedding + keyword). Ranker for relevance.
Advertisement
How it works end to end
ADK tool wrapper: expose search as callable tool. LLM decides when to search.
Filters: metadata-based scoping (per-user, per-project).
Grounding: cite sources from retrieved documents.