Learn

A five-part course on getting structured data out of PDFs. Each page builds on the previous one, mostly using a single running document — a fake inspection report — first as clean digital text, then as a scan.

Every page doubles as a notebook: use the “Open in Colab” badge to run it yourself.

1. Text and tables — open a PDF, find things with selectors, walk the page spatially, pull a table, exclude headers.
2. OCR — when there’s no text layer: apply OCR, judge its quality, compare engines, re-OCR the stubborn parts.
3. AI extraction — ask questions, extract structured records with LLMs, classify documents — and see exactly where AI extraction fails.
4. Page structure — multi-column reflow, content that spans pages, and layout models.
5. Grids — build table grids from headers, whitespace, content anchors, or pixel lines, and read drawn checkboxes.
6. Putting it together — the capstone: turn a five-page library weeding report into one clean DataFrame, then batch the whole thing.

If you’re working on a specific problem instead, jump to Solve and pick the document that looks like yours.