Invoices are structured by convention but encoded as PDF. Converting them to Markdown surfaces the structure: a header section with metadata, a line-items table, a totals block. The result feeds straight into AI summarisers, expense trackers, or accounting workflows.
What we extract from a typical invoice
Invoice number and date go to the metadata block at the top (or YAML front matter if requested). The line items become a GFM table with columns for description, quantity, unit price, and line total. Subtotals, taxes, and grand total are emitted as a separate section after the table. Vendor and customer addresses come through as labelled blocks. Anything else (purchase order references, payment terms, notes) is preserved in its original position.
Why this is useful for AI workflows
An LLM asked to extract invoice data from a raw PDF will struggle — it has to find the line items in a wall of layout noise. Same LLM asked the same question on a Markdown invoice answers correctly in one pass: the table is already a table, the totals are labelled, the dates are obvious. Useful for expense automation, fraud detection, or accounts-payable pipelines.
Frequently asked questions
What invoice fields are extracted?
Invoice number, issue date, due date, vendor name and address, customer name and address, line items (description, quantity, unit price, line total), subtotal, taxes, grand total, payment terms, and any free-text notes. All emit as Markdown in roughly the order the invoice presented them.
Are line items converted to a Markdown table?
Yes — line items become a GFM table with columns for description, quantity, unit price, and line total. The table renders identically in any Markdown viewer and pastes cleanly into spreadsheets.
Does this work on European VAT invoices?
Yes — VAT lines are recognised separately from net subtotals, and tax rates are preserved per line where the invoice itemises them. Multi-currency invoices keep their currency markers.
Can it handle multi-page invoices?
Yes — line items that wrap across pages are stitched into a single GFM table in the output. Page numbers and repeating headers/footers are stripped automatically.
How does this compare to dedicated invoice OCR tools?
Dedicated tools (Rossum, Hypatos, Veryfi) extract structured JSON for accounting integration. Our converter produces human-readable Markdown — better for AI workflows and ad-hoc review, less ideal for automated ERP ingestion. Use the right tool for the job.