Chunk a Document by Size

The first thing every RAG pipeline does, and the thing most of them never revisit. Start here, then look at the structure-aware chunker in this track to see what this version costs you.

Write chunk_fixed_size(text, size, overlap) returning a list of strings.

Split text on whitespace into tokens, then emit windows of at most size tokens, each starting size - overlap tokens after the last. Join each window back with single spaces.

The overlap is why this is not just text.split(): a sentence that straddles a boundary is unretrievable if neither chunk contains all of it, and repeating the last few tokens is the cheap fix.

Careful: if overlap >= size the step becomes zero or negative and the loop never advances. Guard it.