Generic JSON Schema

AI extraction should return a predictable JSON envelope. The exact schema can evolve by document type, but the top-level structure should remain stable.

Example payload

{
  "document": {
    "document_type": "RENTAL_CONTRACT",
    "language": "it",
    "source_pages": 5,
    "overall_confidence": 0.86
  },
  "parties": [
    {
      "role": "LANDLORD",
      "party_type": "PERSON",
      "first_name": {"value": "Mario", "confidence": 0.95},
      "last_name": {"value": "Rossi", "confidence": 0.95},
      "fiscal_code": {"value": "RSSMRA...", "confidence": 0.81}
    }
  ],
  "property": {
    "address": {"value": "Via Roma 10", "confidence": 0.92},
    "city": {"value": "Milano", "confidence": 0.93},
    "cadastral_sheet": {"value": "12", "confidence": 0.74}
  },
  "contract": {
    "contract_type": {"value": "RESIDENTIAL", "confidence": 0.88},
    "start_date": {"value": "2026-01-01", "confidence": 0.91},
    "end_date": {"value": "2030-12-31", "confidence": 0.89},
    "rent_amount": {"value": 1200, "confidence": 0.96},
    "payment_due_day": {"value": 5, "confidence": 0.84}
  },
  "payments": [
    {
      "due_date": {"value": "2026-02-05", "confidence": 0.86},
      "amount": {"value": 1200, "confidence": 0.96},
      "currency": {"value": "EUR", "confidence": 0.99}
    }
  ],
  "bank_account": {
    "iban": {"value": "IT60X...", "confidence": 0.79},
    "account_name": {"value": "Mario Rossi", "confidence": 0.73}
  }
}

Top-level sections

Section	Purpose
document	Document classification, language, page count, and overall confidence.
parties	Extracted people and companies with business roles.
property	Property address, cadastral, and ownership information.
contract	Contract-level business facts.
payments	Expected rent, installments, bill payments, or tax payments.
bank_account	Logical bank account details when present.
expenses	Invoice or expected cost details when relevant.
utility	Utility bill or service information when relevant.

Confidence per field

Each extracted field should contain:

• value: the extracted value
• confidence: number from 0 to 1
• source: optional page or text location
• raw_text: optional source text snippet for review

UI behavior

Confidence range	UI behavior
>= 0.90	Show as high-confidence pre-filled value.
0.70 - 0.89	Show as editable value with review highlight.
< 0.70	Require manual confirmation or leave blank with suggestion.

Backend behavior

The backend should reject invalid formats even when confidence is high. Examples:

• invalid fiscal code format
• invalid IBAN checksum
• end date before start date
• unsupported currency
• rent amount less than or equal to zero
• missing required party role for a contract