Why it matters
Chunking splits documents for retrieval. Too large and answers dilute. Too small and context loses nuance. Bad boundaries split tables and void refund rules across unrelated chunks.
How it works
Use structure-aware splits: headings, paragraphs, tables. Overlap cautiously. Attach heading breadcrumbs as metadata for citations.
Example
Nintendo chunks Refund Policy by H2 section so each chunk carries eligibility rules intact. Help Center procedural lists stay whole so step order does not break across chunks.
Common mistakes
- 1Fixed token splits through markdown tables
- 2No heading metadata on chunks
- 3Identical chunk strategy for legal PDFs and API JSON
Definitions are useful. Governed knowledge is better.
Wiki helps you turn these concepts into a real system for the knowledge behind your AI.
See Wiki in action