How NoteIA reads a 200-page PDF

Our approach to chunking and page-level citation.

A 200-page PDF isn’t text. It’s a hybrid object: table of contents, appendices, tables, footnotes, shifting layout. Feeding it as-is into an engine gives mediocre results. Here’s how we do it.

Chunk, don’t truncate

The first mistake would be to cut the document into fixed-size blocks, straight off. You get fragments that split a sentence in two and lose the context.

Our approach: chunk according to the document’s structure — headings, sections, paragraphs — while keeping the original page numbering. Every fragment knows where it came from.

Keep the page as the anchor

Page-level citation isn’t a detail. It’s what separates a verifiable summary from a hallucination.

When NoteIA writes a summary, every claim is tied to a fragment, and every fragment to a page. You can go back to the source, in the original document, without guessing.

Read in full, cite at the right moment

We read the whole document — not just the beginning, not just the executive summary. But we don’t dump everything into the answer: we cite what’s relevant, at the moment it’s relevant.

That’s the difference between a tool that has “read” the doc and a tool that has done something with it.

What we learn from every document

Every PDF is a case study. Tables, footnotes, appendices each pose a different problem. We adjust the chunking document after document, without ever sacrificing citation.