179 questions
No questions match those filters.
← All questions
RAGMedium· RAG & Retrieval
You're chunking policy documents. What's your strategy?
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansFixed-size chunking ignores the document’s own structure and routinely splits a clause from the heading that makes it findable — you end up with a chunk that contains the answer but not the context a retriever needs to surface it.
My approach:
- Chunk on natural boundaries where the source has them — headings, sections, numbered clauses — rather than an arbitrary character count.
- Keep enough overlap between chunks to avoid severing a sentence or a cross-reference mid-thought.
- Attach metadata to every chunk — section, document type, effective date — so a query can filter before it searches, not just rank after the fact.
The failure mode to actively watch for in testing: a chunk that reads correctly in isolation but has lost the heading or clause number that made it retrievable in the first place.