Retrieval & RAGbeginner11 min · 8 steps
Chunking Strategies
Where you cut a document decides what retrieval can ever find — split on meaning and structure, not on a character count.
indexed unit300tok
sent to LLM300tok
dataerrorrequest
metrics are illustrativeProblemStep 1 of 8
A fixed-size splitter cuts wherever the counter lands
The simplest chunker slices every N tokens. It is content-blind: it splits sentences mid-thought, separates headers from their text and breaks equations and tables.
Adding overlap (say 10% of text repeated across neighbouring chunks) softens the edges, but the model still has to reconstruct meaning from fragments.
Watch outBad chunks can't be rescued downstream. If the fact is split across two vectors, no embedding or reranker will put it back together.
Lesson 2 of 37 in Retrieval & RAG