Agentic corrective RAG

This architecture adds self-correction to retrieval. It is built around the evaluator-optimizer pattern: it judges its own retrieved documents and fixes them when they fall short.

The core idea is to evaluate retrieved documents dynamically, take corrective action, and refine queries to improve the answer. It is built on five specialized agents:

1. A context retrieval agent fetches initial documents from a vector database.
2. A relevance evaluation agent assesses those documents and flags irrelevant or ambiguous ones.
3. A query refinement agent rewrites the query to improve retrieval, using semantic understanding.
4. An external knowledge retrieval agent performs web searches or hits other sources when the context is insufficient.
5. A response synthesis agent integrates the validated information into the final answer.

Strengths: iterative correction (dynamically identifying and fixing poor retrieval), dynamic adaptability (real-time web search and query rewriting), modularity, and factuality assurance (validating content minimizes hallucination). The cost, as the lessons chapter notes, is added latency from the correction loop.

The survey's example is an academic research assistant: it retrieves papers, evaluates their relevance, rewrites the query and searches the web if needed, and only then synthesizes a summary.