LEVEL 02 · LESSON 13
Chunking
A good chunk is small enough to find and complete enough to understand.
What it is
Chunking splits larger source material into smaller units for indexing and retrieval. A useful chunk keeps a coherent piece of meaning together.
Why it matters
Whole documents may be too large or unfocused to retrieve efficiently, while tiny fragments may omit the context that makes a rule meaningful.
A business example
An employee handbook is divided by policy sections. A travel-expense chunk includes the reimbursement rule and its exceptions, rather than cutting immediately before the exception paragraph.
How it actually works
Start from the document structure: headings, paragraphs, lists, and tables. Choose chunk sizes that balance search precision with enough context for interpretation. Overlap can preserve continuity but also creates duplication. Retain source identifiers and section relationships. Test with real questions, especially those requiring exceptions or information across pages. There is no universal word count or overlap percentage that works for every document collection.
A common misconception
Smaller chunks are not always better. A highly relevant fragment can still be unsafe to interpret without its heading, date, or exception.
ONE SENTENCE TO REMEMBER
A good chunk is small enough to find and complete enough to understand.
CHECK YOUR UNDERSTANDING
A contract search returns a precise sentence about a discount, but the answer misses the exception stated in the same section. What is the most likely chunking issue?
Show the answer
Answer: A.
A useful chunk keeps a coherent piece of meaning together. A highly relevant fragment can still be unsafe to interpret without its heading, date, or exception. There is no universal chunk size or overlap; test with real questions, especially ones that depend on exceptions.
Take it into your next meeting
What meaning would be lost if your most important document were split mid-paragraph?
One more concept, understood.
Mark this lesson complete to keep your place.