Unlocking dependable responses with Gemini Enterprise Agent Platform’s Agentic RAG
Research paper
Google Research introduces a new agentic RAG framework that breaks down complex enterprise queries, iteratively searches for sufficient context, and includes a Sufficient Context Agent to ensure completeness before generating a response.
Compared to standard RAG, it increases accuracy on factuality datasets by up to 34% and improves grounding and reasoning accuracy on domain-specific tasks.
The framework is hosted on the Gemini Enterprise Agent Platform as Cross-Corpus Retrieval powered by Agentic RAG.