Document extraction platform

The technological core of the solution is based on an "agentic AI" architecture. In this ecosystem, various specialized models (agents) are trained to perform specific tasks under the supervision of a central orchestrator. This modular approach ensures high flexibility and scalability, allowing the system to dynamically adapt to multiple document types and integrate advanced techniques such as Retrieval-Augmented Generation (RAG) or specific fine-tuning on customer data.

• Multichannel Acquisition: The system automatically intercepts incoming documents through various methods, including direct acquisition from dedicated email inboxes (e.g., automatic receipt of delivery notes via email) or manual upload via the web interface.

• Intelligent Classification: Each acquired document is analyzed by AI, which determines its type. The system provides a confidence level and a logical explanation (reasoning) for its choice, routing the file to the appropriate workflow.

• Semantic Extraction: Using advanced linguistic models, the system extracts relevant information based on context understanding rather than fixed coordinates, ensuring robustness against format variations.

• Data Structuring: The extracted information is organized according to a predefined data schema (JSON Schema), which is user-configurable for each specific document type.

• Management System Integration: Once validated, the structured data is automatically exported and loaded into the company's business management software (ERP, CRM, etc.) via dedicated integration procedures.

‍

Platform Capabilities and Features

The BlueIT solution offers an intuitive and comprehensive web interface, designed to maximize operational efficiency while maintaining full human control over the process. The system's main sections and functionalities are outlined below.

‍

Dashboard and Document Upload

The platform's homepage serves as the main access point for operators. From here, users can quickly initiate new processing workflows and monitor recent activities.

The 'New Upload' area supports drag & drop and accepts a wide range of file formats (PDF, DOCX, XLSX, CSV, images, etc.) up to 50MB per file. On the right, the operator can manually select the desired workflow (e.g., DDT, Invoice, F24-730, Injunction) if known in advance. The interface is designed to support international contexts, offering multi-zone and multi-language management.

‍

Process Monitoring and Management

The 'Processes' section offers a comprehensive and detailed view of all documents processed or currently being processed, ensuring full traceability of operations.

The summary table shows each document's name, associated workflow, current status (e.g., REVIEWED), processing date, and available actions. Operators can use advanced filters to search for specific documents by name, status, workflow type, or time period. An export function is also available for reporting purposes.

‍

Data Validation and Human-in-the-Loop Approach

Despite the high degree of automation, the platform adopts a 'Human-in-the-Loop' paradigm. This means that the human operator always maintains final control over data quality before it is sent to business systems.

The review interface features a convenient side-by-side (split-screen) view. The original document (e.g., the DDT PDF) is displayed on the left, while the extracted and structured data in JSON format is presented on the right. The operator can visually verify the correctness of the extraction (e.g., item codes, quantities, descriptions), make any manual corrections by clicking 'Edit', and once the content is validated, click 'Mark as Reviewed'. Only after this approval are the data unlocked for integration with the management system.

‍

Batch Processing and Automatic Classification

To efficiently manage high document volumes, the platform features a powerful Batch Processing function enhanced with AI-based automatic classification capabilities.

When multiple files are uploaded simultaneously, the AI engine analyzes the content of each document and automatically suggests the most appropriate workflow (e.g., recognizing one file as a Modello 730 and another as a DDT). To ensure maximum transparency, the system provides a confidence percentage and, crucially, a textual 'Reasoning': the AI explicitly explains which elements of the document led to that classification (e.g., presence of specific regulatory references or typical layout). This allows the operator to quickly approve correct classifications.

‍

Flexible Workflow Management

The true strength of the solution lies in its extreme flexibility. The 'Workflow Management' section allows administrators to configure and manage custom workflows for any type of business document.

From this panel, you can view all active workflows (e.g., DDT, F24-730, Invoice, Injunction, Mediation), organize them into logical categories (e.g., RDA, SIAV), and activate or deactivate them with a simple toggle. This architecture allows the solution to be rapidly scaled to new business use cases without the need for additional software development.

‍

Extraction Mask Configuration (JSON Schema)

Creating a new workflow is a guided and highly customizable process that does not require advanced programming skills but relies on defining an 'extraction mask'.

In this phase, the user defines the 'JSON Schema for AI', which is the exact structure of the data that the Artificial Intelligence needs to extract from the document. For example, for a DDT, the schema will instruct the AI to look for an array of 'orders', each containing 'lines' with specific fields such as 'quantity', 'description', 'unitOfMeasure', and 'itemCode'. It is also possible to provide a textual description that helps the AI contextualize the document, and to define supported file formats and processing priority.

‍

Conclusion

The BlueIT solution goes beyond simply digitizing a process; it introduces true operational intelligence, freeing human resources from repetitive tasks to focus on higher-value activities.