Convert PDF to JSON

How it works

1. 1Upload the PDF.
2. 2We extract the table — and, for invoices/receipts, the header fields.
3. 3Download JSON and load it straight into your pipeline.

What you download: A JSON array of row objects keyed by the document's own column titles; invoices and receipts also carry their extracted fields and a totals-reconcile verdict. Pair it with the Markdown export when you're feeding an LLM.

PDF data extraction that behaves like an API response

Scraping text out of PDFs in code means regexes that break on the next document. This gives you the shape an API would: rows as objects, keys from the real column titles, values verbatim. Deterministic extraction means the same report layout produces the same JSON schema next week — your parser stays boring, which is the point.

Invoices in JSON — with the maths checked

For invoices and receipts, the JSON includes the extracted header fields (vendor, invoice number, dates, PO, subtotal, tax, total) and a reconciliation verdict: we independently check that line items sum to the subtotal and that subtotal plus tax equals the total. Your pipeline can branch on reconciled: true instead of trusting every extraction blindly.

JSON for machines, Markdown for models

If the consumer is code, JSON is right. If the consumer is an LLM — a summarisation prompt, a RAG index — Markdown usually beats JSON: the model reads headings and tables natively without you flattening structure into strings. Both come from the same upload; take whichever fits the job, or both.

What about scanned PDFs?

Digital PDFs with a real text layer give exact results — we read the document's own characters and coordinates, not a picture of them. If you upload a pure scan with no selectable text, we detect it and tell you honestly instead of returning garbage rows.

Questions