A text chunk is a coherent segment of a document treated as an individual unit for indexing and retrieval. In a retrieval-augmented generation pipeline, chunks are converted into embeddings, stored, and matched against a user’s query to retrieve relevant passages. section 6.3.2
Paragraph-level chunking
A document can be divided at paragraph boundaries, producing smaller units for indexing. Preserving headings, tables, and lists during ingestion can also support search accuracy and relevance. section 6.3.2 section 6.4.3
When preparing documents for retrieval:
- Treat chunking strategy as a factor that can affect retrieval results. section 6.3.2
- Partition large documents into representative subdocuments for more granular indexing and retrieval. section 6.4.3
