What DocExtract Mega adds after extraction
OCR-style correction, field-specific date and numeric formats, literal text substitution, and a change-by-change audit. What the Mega plan adds, with verified postprocessing examples and its limits.
Working notes on what we build and how — document extraction, API design, and autonomous software delivery.
OCR-style correction, field-specific date and numeric formats, literal text substitution, and a change-by-change audit. What the Mega plan adds, with verified postprocessing examples and its limits.
Use the "general" document_type to get a best-effort structured summary (title, entities, topics, free-form fields, summary) for any text that doesn't match a known schema like invoices or resumes.
Most text people need as JSON is not an invoice or a resume — it is a listing, an incident writeup, a spec sheet. The general type (the default) returns a best-effort structured summary of any text: title, entities, topics, free-form fields, and a summary. Two verified responses, and an honest list of where it stops.
Choose flat monthly quota pricing for predictable per-call billing (a 199-char invoice and 6,600-char contract each cost one call), or per-page pricing if you prefer variable costs tied to document length.
Per-page pricing bills pages × rate and is hard to forecast; a flat monthly call quota bills calls × rate, which you can predict. The trade-off with real numbers, plus a verified response showing a 199-character invoice and a 6,600-character contract each cost exactly one unit.
Feed raw OCR text (Textract, Document AI, Tesseract) into DocExtract to get typed JSON in one call — it handles OCR noise but cannot fix missing characters or hallucinated words.
You have already run Textract, Document AI, or Tesseract — now you have a wall of recognised text, not JSON. Here is how to turn that OCR output, noise and all, into typed JSON with one API call, and exactly which errors it will and will not fix.
Use pdftotext or pdfminer.six to pull the existing text layer from generated PDFs, then send that text to DocExtract for typed JSON — no OCR needed.
Most generated PDFs — ERP exports, print-to-PDF, reporting tools — already carry a real text layer. Pull it with pdftotext or pdfminer.six, then turn that text into typed JSON with one API call. No OCR, and a flat monthly quota instead of per-page pricing.
Send contract text (CLM export, pasted clause, text-layer PDF) to DocExtract to extract parties, roles, effective/expiration dates, value, and governing law in one API call.
The contract is usually already text by the time someone needs data out of it — a CLM export, a pasted clause, a text-layer PDF. Here is how to turn it into typed JSON with one API call: parties and roles, effective and expiration dates, value, governing law — and where the output still needs checking.
Send meeting transcripts, bullets, or email recaps to DocExtract to get action items with owners and due dates, decisions, and attendees as typed JSON in one call.
The useful part of a meeting is usually already text — a transcript, pasted bullets, an email recap. Here is how to turn it into typed JSON with one API call: action items with owners and due dates, decisions, attendees — and where the output still needs checking.
Paste plain-text CV content into DocExtract to get structured JSON with sections, experience, education, and skills — no file upload or OCR required.
A lot of resume text never arrives as a file — it is pasted into a form, sent as an email body, or a text-layer PDF. Here is how to turn that CV text straight into typed JSON with one API call, and where the output is still rough.
Use DocExtract when you need schema enforcement, deterministic pricing, and zero prompt engineering at scale; call the model yourself only for open-ended or one-off extractions.
You can already get JSON out of GPT or Gemini. Here is an honest look at when a narrow extraction API is worth it instead — and when it is not.
Send invoice text (email body, portal export, text-layer PDF) to DocExtract to get line items, totals, taxes, and vendor data as typed JSON — skip OCR entirely when you already have the text.
A lot of invoices never needed OCR — they arrive as email bodies, portal exports, or text-layer PDFs. Here is how to turn that invoice text straight into typed JSON, and how to tell when you can skip OCR entirely.
Send any document text (invoices, resumes, contracts) to DocExtract to get typed JSON in one API call with a fixed schema — flat monthly quota, no OCR, no prompt engineering.
Every project that touches invoices, resumes, or contracts ends up rewriting the same prompt-and-parse glue. Here is how DocExtract turns document text into typed JSON in a single request — and where it deliberately stops.