Chunking Strategies

Choose a strategy that represents the documents inherent structure.

Strategy Type	Definition	Use case
Fixed size/Sentence aware	Standard splits based on character count and sentence boundaries	General use cases like blogs, memos
Semantic Chunking	Identify boundaries based on embedding variance	Messy text, where logical shifts dont align with paragraph markers
Document aware/adaptive	Segmenting based on DOM structure, headings, tables, code blocks etc.	Technical manuals, financial reports, source code
Parent/child retriever	Decoupling retrieval units from the generation unit	Balancing precision (small child chunks) with context (large parent blocks)