Definition
An advanced RAG pattern that uses a self-grading mechanism to evaluate the relevance of retrieved documents, triggering alternative actions like web search or knowledge refinement if retrieval quality is low. It improves accuracy and grounding but introduces architectural trade-offs such as increased latency and higher token costs due to iterative evaluation steps.
Focuses on validating the relevance of retrieved context before generation, whereas standard RAG proceeds regardless of retrieval quality.
"A quality control inspector on a factory line who reroutes defective raw materials for extra processing before they reach the assembly robot."
Conceptual Overview
An advanced RAG pattern that uses a self-grading mechanism to evaluate the relevance of retrieved documents, triggering alternative actions like web search or knowledge refinement if retrieval quality is low. It improves accuracy and grounding but introduces architectural trade-offs such as increased latency and higher token costs due to iterative evaluation steps.
Disambiguation
Focuses on validating the relevance of retrieved context before generation, whereas standard RAG proceeds regardless of retrieval quality.
Visual Analog
A quality control inspector on a factory line who reroutes defective raw materials for extra processing before they reach the assembly robot.