ML//RAG//chunking strategy

A chunking strategy is a rule for splitting documents into pieces before embedding, and it decides what units a retrieval system can find and hand back.

A chunking strategy is a rule for splitting documents into pieces before embedding, and it decides what units a retrieval system can find and hand back.

Fixed-size chunks miss semantic boundaries. Recursive splitting (into paragraphs, then into sentences) does better.

Overlap between chunks prevents losing context at their edges.

Chunk size trades precision against recall: small chunks match better but lose the surrounding context.