Unlocking dependable responses with Gemini Enterprise Agent Platform’s Agentic RAG

Google Research · 1 month ago

Google introduced Agentic RAG, a multi-agent retrieval-augmented generation system within the Gemini Enterprise Agent Platform that breaks down complex queries across multiple data sources through specialized agents for planning, routing, and context validation. The system achieved 90.1% accuracy on FramesQA's 824 multi-hop questions across 4 different data corpora, compared to lower accuracy from standard RAG approaches, while maintaining comparable latency. The key innovation is a Sufficient Context Agent that detects missing information and triggers iterative searches until enough grounded context exists to answer complex queries completely.