DocProcessor Summary

Copy

• Copy for LLM

  Copy page as Markdown for LLMs
• View as Markdown

  Open this page as Markdown
• Open in ChatGPT

  Get insights from ChatGPT
• Open in Claude

  Get insights from Claude

DocProcessor is an AI-powered document understanding engine leveraging Large Language Models (LLM) and Vision-Language Models (VLM) to read, interpret, and extract structured data from diverse text-based documents across multiple languages and formats.

‌

Key Value Propositions

• Flexibility: Choose precisely which data to extract via configuration—no model training required
• Speed: Integrate new document types quickly without model retraining, reducing time from weeks to days
• Robustness: High-quality extraction powered by LLM/VLM technology
• Multilingual: Native support for 5 major European languages
• Easy Integration: RESTful API for seamless system integration

Technical Specifications

Supported Languages (V1)

DocProcessor supports the following languages for document text:

• French
• English
• German
• Italian
• Spanish

Supported Document Types (V1)

Document Format (V1)

• Formats: PDF, JPEG, PNG
• Pages: Multi-page support
• Document Type: Text document
• Maximum Size: 20 MiB

Performance & SLA

• Target Response Time: 10 seconds
• Availability: 100%
• API Standard: REST
• Output Format: Structured JSON

How It Works

DocProcessor follows a simple workflow to process your documents and extract structured data:

Step 1: Submit Your Document

Upload your document in one of the supported formats:

• PDF
• JPEG
• PNG

The system accepts documents up to 20 MB in size.

Step 2: Automatic Processing

Once submitted, DocProcessor automatically:

• Reads and interprets the document content
• Identifies the document type
• Processes the document in its original language (French, Italian, German, English, or Spanish)
• Extracts the specified data fields according to the configuration

Step 3: Receive Structured Data

The system returns extracted data as structured JSON, with typed data structures:

• TEXT: Simple textual values
• ADDRESS: Structured address with street, zip code, and city
• NAMES: First names and last names separated
• DATE: ISO-formatted dates (YYYY-MM-DD)

Key Capabilities

Multilingual Support: DocProcessor automatically detects and processes documents in French, Italian, German, English, or Spanish without requiring language specification.

New Document Types: The system can adapt to new document formats without model retraining, ensuring no downtime when introducing new document types.

Data Types Supported

DocProcessor extracts fields with the following data types:

• TEXT: Simple textual values (e.g., company name, fiscal number)
• ADDRESS: Structured address with street, zip code, and city
• NAMES: First names and last names separated
• DATE: ISO-formatted dates (YYYY-MM-DD)

Use Cases

Example Use Case: Payslip Data Extraction

A customer needs to extract specific data from employee payslips across multiple countries and formats.

Traditional Approach:

• Requires approximately 8 weeks to train models for each document variant
• Limited flexibility for field customization
• Significant development effort for new formats

DocProcessor Approach:

• Extract exactly the fields needed (e.g., last name, first name, gross salary)
• Integration in days instead of weeks
• Easy modification via prompt-based configuration
• Support for documents from different countries without retraining

Example Use Case: KYC Onboarding

Financial institutions can automate customer data extraction from:

• Tax notices for income verification
• Bank account statements for IBAN validation
• KBIS documents for company verification
• Multiple document types in a single workflow

API Integration

Sample JSON Output

{
  "textDocumentInfo": {
    "documentTypeDetail": "PAYSLIP",
    "fields": {
      "EMPLOYER_NAME": {
        "data_type": "TEXT",
        "value": "NETHEOS"
      },
      "NET_SALARY": {
        "data_type": "TEXT",
        "value": "250,76"
      },
      "GROSS_SALARY": {
        "data_type": "TEXT",
        "value": "151,67"
      },
      "EMPLOYER_ADDRESS": {
        "data_type": "ADDRESS",
        "address": "avenue bernard claude Parc Club du Millenaire",
        "zipCode": "34000",
        "city": "MONTPELLIER"
      },
      "EMPLOYEE_NAME": {
        "data_type": "NAMES",
        "firstNames": "JOHN",
        "lastName": "CENA"
      },
      "EMPLOYEE_ADDRESS": {
        "data_type": "ADDRESS",
        "address": "Les impasses de la Mer Appt 34 70 rue de Pivert",
        "zipCode": "34000",
        "city": "MONTPELLIER"
      },
      "PAYSLIP_DATE": {
        "data_type": "DATE",
        "value": "2015-01-01"
      }
    }
  }
}

Constraints & Limitations

Technical Constraints

• Documents must be text-based (not handwritten)
• Maximum file size: 20 MB
• Documents must be in A4 format or a similar standard size
• Requires reasonable image quality for accurate extraction

Benefits & Competitive Advantages

Speed to Market

• New document type integration in days vs traditional 8 weeks
• No model training required for new formats
• Rapid adaptation to customer-specific requirements

Flexibility & Scalability

• Customizable field extraction via configuration
• Prompt-based approach allows easy modifications
• Extensible architecture for future document types

Quality & Accuracy

• Powered by state-of-the-art LLM/VLM technology
• High extraction quality across multiple languages
• Robust handling of varied document structures

Integration with Namirial OnBoarding

Overview

DocProcessor integrates with Namirial OnBoarding (NOB) to enable automated document processing within customer onboarding workflows.

Current Status: Demo integration - simplified workflow for evaluation purposes.

Current Integration (Demo)

Available Features

The current integration offers a single, pre-configured workflow designed for demonstration and evaluation purposes:

• Single document upload per request
• Pre-configured document processing (no customization available)
• No input parameters required
• Processing through DocProcessor backend
• Results available in NOB backoffice

Limitations

• Configuration options (parameters and settings) are not yet available
• Document cannot be passed directly via API (upload link only)
• Single workflow configuration
• Limited to demonstration scenarios

Integration Methods

1. Request Creation from Backoffice

Setup

The integration uses a Request Type based on a specific model.

Note: The name references the legacy Text Engine system and will be updated to reflect the DocProcessor integration.

Process

Step 1: Create Request

1. Access the Namirial OnBoarding back office
2. Select the Request Type that covers
3. Click "Create"

No parameters need to be configured - the system uses a pre-defined configuration.

Step 2: Upload Document

1. The system generates a unique link
2. Share the link with the end user
3. User accesses the link and uploads a single document via the web interface

Step 3: Processing

• Document is sent to DocProcessor
• Automatic processing and data extraction
• Results available in the NOB back office

2. Request Creation via API

Endpoint

POST https://test-eu-ie1-api.namirialonboarding.com/api/v2/requests

Headers

Authorization: Bearer {YOUR_ACCESS_TOKEN}
Accept: application/json
Content-Type: application/json

Request Body

{
  "requestTypeId": "8870fa7a-2e51-4af4-9724-2ac4230163db",
  "parameters": {},
  "settings": {}
}

Note: The parameters and settings fields are currently empty and not configurable. They are reserved for future enhancements.

cURL Example

curl 'https://test-eu-ie1-api.namirialonboarding.com/api/v2/requests?language=en' \
  -H 'Authorization: Bearer {YOUR_ACCESS_TOKEN}' \
  -H 'Accept: application/json' \
  -H 'Content-Type: application/json' \
  --data-raw '{"requestTypeId":"8870fa7a-2e51-4af4-9724-2ac4230163db","parameters":{},"settings":{}}'

Response

{
  "requestId": "unique-request-id",
  "link": "https://...",
  "status": "created"
}

The response contains:

• requestId: Unique identifier for tracking
• link: URL to share with end user for document upload
• status: Current request status

Important Notes

• The requestTypeId is specific to each integration and environment
• Currently, the document cannot be passed in the API call body
• Users must use the generated link to upload documents

Planned Enhancements

1. Internal Operator Review Step

Enable manual review and validation after automatic processing

Workflow:

1. Document is processed automatically by DocProcessor

2. Request enters "Pending Review" status

3. Internal NOB operator reviews:

   • Original document
   • Extracted data
   • Processing results

4. Operator can:

   • Approve the request
   • Reject the request
   • Correct extracted data if needed

5. Request proceeds to next workflow step

Features:

• Automatic or manual assignment of requests to operators
• Workload monitoring dashboard
• Review history and audit trail

2. Document Upload via API

Ability to pass the document directly in the request creation call, eliminating the upload link step.

Benefits:

• Fully automated workflow (no user interaction required)
• Direct integration with external systems
• Faster processing time

3. Configurable Parameters

Enable configuration of document processing parameters per request.

Expected Parameters:

• documentType: Specify expected document type for optimized processing
• language: Override language detection
• extractionFields: Customize which fields to extract
• validationRules: Apply custom validation logic

Expected Settings:

• confidenceThreshold: Minimum confidence score for extracted data
• manualReviewRequired: Force manual review step
• webhookUrl: URL for asynchronous notifications