The Doc-AI document extraction API

If you have built document processing before, you know the shape of it. PDF parsing that works until it meets a scanned page. An OCR integration that returns words but not table structure. Splitting logic for multi-document packets that never quite gets the boundaries right. Extraction prompts that work on the ten documents you tested and fail on the eleventh. Then an evaluation harness, because you cannot tell whether a change helped. Then a review interface, because some documents will always need a human.

That is two to four engineers, permanently, on something that is not your product. The pitch here is not that document AI is hard — it is that it is a solved, undifferentiated problem you are paying full engineering salary to re-solve.

1. Submit a document

POST the file with the document type you expect. Doc-AI returns a transaction ID immediately and processes asynchronously.

curl -X POST https://api.doc-ai.wisetrend.com/v1/transactions \
  -H "Authorization: Bearer $DOC_AI_API_KEY" \
  -F "file=@invoice-8842.pdf" \
  -F "project=accounts-payable" \
  -F "callback_url=https://your-app.example.com/hooks/doc-ai"

{
  "transaction_id": "txn_01HQ8...",
  "status": "processing",
  "received_at": "2026-09-04T14:22:07Z"
}

2. Define what you want, in plain language

A document type is a description plus a field list. No template, no labeled samples, no training run. This is the whole configuration.

{
  "document_type": "supplier_invoice",
  "description": "A supplier invoice from any vendor. May be multi-page with a
                  line-item table that continues across pages.",
  "fields": [
    { "name": "invoice_number", "type": "string",  "description": "The vendor's own invoice number" },
    { "name": "invoice_date",   "type": "date"   },
    { "name": "vendor_name",    "type": "string" },
    { "name": "vendor_tax_id",  "type": "string",  "description": "VAT, GST or EIN, whichever appears" },
    { "name": "currency",       "type": "enum",    "values": ["USD","EUR","GBP","CAD"] },
    { "name": "subtotal",       "type": "decimal" },
    { "name": "tax_total",      "type": "decimal" },
    { "name": "grand_total",    "type": "decimal" },
    { "name": "line_items",     "type": "array",
      "items": [
        { "name": "description", "type": "string"  },
        { "name": "quantity",    "type": "decimal" },
        { "name": "unit_price",  "type": "decimal" },
        { "name": "line_total",  "type": "decimal" }
      ]
    }
  ],
  "rules": [
    { "name": "totals_foot",   "expr": "abs(subtotal + tax_total - grand_total) < 0.01" },
    { "name": "lines_reconcile", "expr": "abs(sum(line_items.line_total) - subtotal) < 0.01" },
    { "name": "vendor_known",  "lookup": "vendor_master", "on": "vendor_tax_id" }
  ]
}

3. Retrieve the result

Every value carries a confidence score and the page region it came from, so you can drive your own review interface or overlay boxes on the original page.

GET /v1/transactions/txn_01HQ8.../result

{
  "transaction_id": "txn_01HQ8...",
  "status": "completed",
  "document_type": "supplier_invoice",
  "classification_confidence": 0.99,
  "validation": { "passed": true, "rules_failed": [] },
  "review_required": false,
  "fields": {
    "invoice_number": { "value": "INV-8842",   "confidence": 0.99,
                        "page": 1, "bbox": [412, 96, 528, 118] },
    "invoice_date":   { "value": "2026-08-19", "confidence": 0.98,
                        "page": 1, "bbox": [412, 124, 520, 144] },
    "vendor_name":    { "value": "Northwind Components Ltd", "confidence": 0.97 },
    "grand_total":    { "value": 14820.55, "confidence": 0.99 },
    "line_items": [
      { "description": { "value": "Bearing assembly, 40mm", "confidence": 0.96 },
        "quantity":    { "value": 120,     "confidence": 0.99 },
        "unit_price":  { "value": 98.40,   "confidence": 0.99 },
        "line_total":  { "value": 11808.00,"confidence": 0.99 } }
    ]
  }
}

4. Receive a webhook instead of polling

For volume workloads, register a callback and Doc-AI posts when a transaction reaches a terminal state — completed, needs review, or failed.

POST https://your-app.example.com/hooks/doc-ai
X-Doc-AI-Signature: sha256=...

{
  "event": "transaction.completed",
  "transaction_id": "txn_01HQ8...",
  "document_type": "supplier_invoice",
  "review_required": false,
  "result_url": "https://api.doc-ai.wisetrend.com/v1/transactions/txn_01HQ8.../result"
}

5. Submit reviewer corrections back

Corrections made in your own interface feed the improvement loop, so accuracy on your document types improves without a retraining cycle.

POST /v1/transactions/txn_01HQ8.../feedback

{
  "corrections": [
    { "field": "vendor_tax_id", "was": "GB4471820", "should_be": "GB447182055" }
  ],
  "reviewer": "ap-clerk-14"
}