Verification and Filtering

Standard RAG suffers from "Lost in the Middle" or irrelevant chunks. Corrective RAG pattern:

1. Retrieve
2. Filter: An LLM Node reads each chunk and labels it Relevant or Irrelevant.
3. Branch:
   • If 0 chunks are relevant -> Route to WebSearchNode.
   • If >0 chunks are relevant -> Route to GenerateNode.

This self-correction loop drastically improves quality.