AI Architecture
Parent Document Retriever: Small Chunks for Retrieval, Large Chunks for Context
Improve RAG recall and coherence: index 200-character granular chunks for vector matching, but return the full parent section to the LLM.
3 min
The Small-to-Big Retrieval Principle
Small snippets yield higher cosine similarity against specific queries, while surrounding parent paragraphs give the LLM full context to construct accurate answers.