Glossary

Chunking — Glossary

Splitting documents into smaller passages before embedding and retrieval in RAG systems.

Reviewed:

Full definition

Chunking determines the unit of retrieval. Chunks that are too large mix topics; chunks that are too small lose context. Structure-aware splitting on headings and paragraphs usually beats naive character cuts.

Store metadata—titles, URLs, ACLs—with each chunk to enable citations and secure filters. When answers miss known facts, inspect retrieved chunks before blaming the LLM.

Related encyclopedia topic

Chunking — Chunking splits documents into smaller passages before embedding and retrieval so models receive focused, relevant context.

← Back to glossary · Learning hub