Dual-Brain PDFs: Replacing Brittle OCR & Expensive Vision LLMs with 0ms Embedded JSON-LD

Think about the absurdity of modern B2B document workflows: Company A has clean, structured JSON in their database (invoice line items, SKU codes, tax rates, IBANs). They render that JSON into a visual PDF and email it to Company B. Company B then pays for an OCR pipeline or Vision LLM to look at the pixels of the PDF and guess what the original JSON was.

How Dual-Brain PDFs Work

Whenever you render a document using a hosted or custom template_id, or whenever you pass ai_ready: true and embed_data: { ... } on a raw HTML render, Tailprint automatically attaches tailprint-manifest.json into the PDF container:

// 1. Render a Dual-Brain PDF with embedded structured data
const renderRes = await fetch('https://tailprint.com/v1/render/pdf', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer sec_live_...',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    html: '<div class="p-10"><h1>Purchase Order #PO-9941</h1>...</div>',
    ai_ready: true,
    embed_data: {
      po_number: 'PO-9941',
      currency: 'USD',
      total_net: 48200.00,
      cost_center: 'CC-ENGINEERING-04',
      line_items: [
        { sku: 'SRV-H100-8X', qty: 2, unit_price: 24100.00 }
      ]
    }
  }),
});

Extracting the JSON in 0ms via POST /v1/pdf/extract

Any receiving system, AI agent, or ERP webhook can pass the PDF binary (base64 or URL) to POST /v1/pdf/extract and recover the exact JSON object in under a millisecond — with zero OCR errors and zero LLM token cost:

curl -X POST https://tailprint.com/v1/pdf/extract \
  -H "Content-Type: application/json" \
  -d '{"pdf": "https://cdn.yourdomain.com/po-9941.pdf"}'