Every business runs on documents that are almost data: invoices with totals, resumes with skills, statements with transactions. Structured data extraction is the step that turns them into fields your systems can use — and the teams that get it right all do the same thing first: normalize the document before extracting from it.
Why Markdown-first extraction wins
- A PDF is a layout, not a text stream — naive text dumps scramble tables and reading order.
- Converting document to Markdown first restores structure: headings, tables, and lists survive.
- Extraction over clean Markdown is dramatically more reliable than extraction over raw bytes — and auditable, because you can read the intermediate.
The pipeline in Kit for AI
Upload a file or point at a URL; scans and images go through OCR automatically; the result is clean Markdown you can keep, search, or chat over. Need fields? Use the recipe flows — invoice, receipt, resume, paper, bank statement — or take the Markdown into your own extraction step. The PDF guide covers the hardest format in depth.
Extraction or retrieval?
Extract when downstream code needs fields (totals, dates, line items). Retrieve when a human or agent needs answers — ingest into a knowledge base and ask, with citations. Same converted document either way, so you don't choose upfront.
Convert one real document and see the Markdown — free to start.