Chunking

What it is

Chunking is breaking long documents into smaller pieces for processing. Since models have context limits, long documents must be split. Chunking strategies consider where to split (paragraphs, sentences, semantic units) and how much overlap between chunks to maintain context.

Why it matters

How you chunk affects quality. Poor chunking—splitting mid-sentence or separating related content—produces worse results. Good chunking preserves semantic units and includes overlap so context isn't lost at boundaries. This is crucial for RAG systems and long document processing.