C · Chunking · context · contextual retrieval

Chunking

Splitting documents into retrievable segments. A critical RAG strategy.

In one sentence

Chunking is the process of splitting source documents into smaller passages so a retriever can return only the most relevant segment to a language model, rather than the whole document.

When it matters

For technical manuals with mixed text, tables and code blocks. Bad chunking destroys recall on tables; good chunking preserves table-row context.

A real-world example

A 1,200-page HVAC manual is chunked by section + sliding window; tables are kept intact as one chunk. Recall on error-code queries jumps from 64% to 91%.

Curated by helpcode research team · Last reviewed 2026-05-22