TotalAgility Third-Party Extraction Framework

This framework demonstrates how TotalAgility can be connected to third-party extraction (TPE) providers such as Microsoft Azure Document Intelligence and Google Cloud Document AI for intelligent document processing. It routes documents to specialized AI models, extracts fields and line items with bounding box coordinates, and maps results into TotalAgility extraction groups for validation. The same framework can be extended to support additional document types or new extraction providers.

Multi-provider extraction framework Connects Tungsten TotalAgility with third-party AI extraction providers — Microsoft Azure Document Intelligence and Google Cloud Document AI — to extract fields, line items, and bounding box coordinates from documents.

Unified DLL entry point A single Extract method handles provider routing, document retrieval via the TotalAgility SDK, model resolution, and API communication — no need to call provider-specific classes directly from TotalAgility processes.

Central JSON-based routing configuration Document types are mapped to provider-specific model or processor IDs via a single TPE-MODEL-MAP server variable. Adding a new document type requires only a configuration change — no code modifications needed.

Field-level bounding box highlighting Extracted field values are mapped with precise bounding box coordinates, enabling field highlighting on the document image in TotalAgility's validation screen for Driving License and Receipt document types.

Line item and table extraction Supports multi-row table extraction for Receipts and Invoices, mapping individual item fields (Description, Quantity, Price, Amount, ProductCode) to extraction group table columns.

Cloud-environment compatible The Google Document AI integration includes a pure managed RSA-SHA256 signing implementation using BigInteger.ModPow, bypassing Windows Crypto API and CAS restrictions in Tungsten Cloud environments.

Reusable subprocess architecture Common operations — field extraction, position calculation, entity matching — are built as reusable, document-type-agnostic subprocesses shared across all document types.

Fully extensible framework The same framework can be extended to support additional document types (by updating configuration and cloning processes) or entirely new extraction providers (by adding a new DLL class and TotalAgility processes).

Enables TotalAgility to leverage specialized third-party AI extraction services from Microsoft Azure Document Intelligence and Google Cloud Document AI, reducing the need for custom integration development.

Provides a multi-provider framework that allows organizations to route different document types to different AI providers based on extraction quality, cost, or preference.

Delivers extracted field values with precise bounding box coordinates, enabling field-level highlighting in TotalAgility's validation screen for faster and more accurate document review.

Supports line item and table extraction for complex document types like Invoices and Receipts, mapping structured data directly into TotalAgility extraction groups.

Offers a fully extensible architecture — new document types can be added with configuration changes only, and new extraction providers can be added by following documented patterns in the Blueprint guide.

Works in both on-premise and Tungsten Cloud environments, with a pure managed RSA implementation that bypasses CAS restrictions for Google OAuth2 authentication.

The Third-Party Extraction Framework is designed to simplify and accelerate the integration of external AI-powered document extraction providers with Tungsten TotalAgility. It provides a ready-to-use, extensible solution that enables TotalAgility workflows to route documents to Microsoft Azure Document Intelligence or Google Cloud Document AI, extract field values with bounding box coordinates, and map results directly into TotalAgility extraction groups for validation — without requiring custom development.

All framework configuration, including provider selection, model/processor routing, authentication credentials, and API settings, is managed centrally using TotalAgility Server Variables and a single JSON routing map (TPE-MODEL-MAP). Document types are mapped to provider-specific AI models at the configuration level, meaning new document types can be added without any code changes.

The solution includes a TotalAgility package, complete C# source code, and a comprehensive Blueprint & Implementation Guide covering architecture, setup, process reference, troubleshooting, and step-by-step How-To guides for adding new document types or entirely new extraction providers.

This framework is ideal for organizations looking to complement TotalAgility's native extraction capabilities with specialized third-party AI services, support multi-provider strategies, or build a reusable extraction architecture that can be extended across document types and providers. It works in both on-premise and Tungsten Cloud environments.