- Introduction
- Capability types
- Choosing the correct capability
- Access Control
IXP overview guide
Capability types
- Communications data (via Communications Mining™)
- Unstructured and complex documents (via Generative Extraction for Unstructured Documents™)
- Structured and semi-structured documents (via Document Understanding™)
To process communications, use Communications data.
To process documents, use the following capabilities
- The Unstructured and complex documents capability in IXP. You can use this capability to process documents up to 50 pages. This is a temporary limit and will be increased in future releases.
- The Document Understanding Generative Extraction activities available only in Studio Web and Studio Desktop. You can use these activities to process documents that contain more than 50 pages.
- The classic or modern experience in Document Understanding™.
Use all applicable capabilities if your use case contains both communications and documents. For example, process emails through the Communications Mining capability through the Communications data tab, and attachments based on their document type. Some possible approaches include:
- Option 1: Communications data and Structured and semi-structured documents.
- Option 2: Communications data and Unstructured and complex documents.
- Option 3: Communications data, Structured and semi-structured documents and Unstructured and complex documents.
For more details on this topic, check Choosing the correct capability.
All short-form unstructured communications such as emails, messages, tickets, reviews, and so on, should be processed through the Communications data capability, Communications Mining™.
Communications data extraction is based on a combination of specialized AI and generative AI to enhance user experience and time-to-value.
You can disable all generative AI features within this capability, such as Generative Annotation and Generative Extraction, at the dataset level in Communications Mining.
The Unstructured and complex documents capability, as well as the Document Understanding Generative Extraction activities, are best suited for unstructured and complex documents.
Such documents contain paragraphs of free-form text and complex elements such as:
- Complex tables
- Graphics
- Charts
- Checkboxes
- Call-out boxes
- Signatures
- Handwriting, and more.
Unstructured documents often come in different formats or layouts and might require different extraction schemas. Multiple document types can be combined in a single stack of documents, for example, a mortgage application, such as proof of identity, proof of address, bank statements, and so on. Such combined document might require Generative Extraction, if it is not split into separate document types.
Additionally, you might need to extract values that are not explicitly present in the document, that is they need to be inferred. Examples of inferred values can include:
- Values that are not present anywhere in a document, but are implied from its context.
- Values that need to be concatenated across different areas in a document.
- Values that span across multiple paragraphs, lines, or columns.
Both IXP and the Document Understanding Generative Extraction activities rely on generative AI capabilities. As a result, it is not possible to use these products without enabling generative AI. If your organization policies restrict you from using generative AI capabilities in production, use classic or modern projects in Document Understanding™.
The new IXP capability for unstructured and complex documents currently supports documents up to 50 pages. This limit will be significantly increased in future releases. Meanwhile, if you need to process unstructured documents with more than 50 pages, you can use the Document Understanding Generative Extraction activities with built-in RAG for long documents.
Document Understanding Generative Extraction activities are only accessible through Studio Web or Studio Desktop. The following extractors are available within this capability:
- Long Document Simple Layout - optimized for long form documents with mostly text and headings.Uses the GPT4-turbo AI Trust Layer LLM. Only supports text processing, has a buit-in RAG, and a limit of 500 pages per document.
- Long Document Complex Layout (Preview) - oprimized for long-form documents with complex elements such as tables, images, handwritting, form elements, and floating callout boxes. Uses the GPT-4o AI Trust Layer LLM. Supports both text and image processing, has a buit-in RAG, and a limit of 500 pages per document.
- Short Document Complex Layout (Preview) - optimized for short-structured or semi-structured documents with complex elements such as tables, images, handwritting, form elements, or floating callout boxes. Uses the GPT-4o AI Trust Layer LLM. Supports both text and image processing, has no buit-in RAG, and a limit of only 20 pages per document.
Among other benefits, the Unstructured and complex documents capability provides the following:
- User interface for document annotation and validation.
- Performance statistics and confidence scores for all extractions.
- Ability to quickly iterate the extraction schema and prompt instructions.
- Ability to save and compare different model versions.
-
Ability to choose between different LLMs. Currently, IXP supports only GPT-4o for the moment. However, more models will be available in future releases.
- Configurable model settings, such as Temperature, Seed, and so on.
The Structured and semi-structured documents capability in UiPath® IXP leverages the Document Understanding™ classic and modern projects. These projects are best suited for structured or semi-structured documents, and tend to follow the same or very similar layout without any complex elements.
Document Understanding classic and modern experiences use a combination of specialized out-of-the-box models and generative AI features. You can manage all generative AI features within this capability, such as automatic classification and generative annotation, from AI Trust Layer.
For more details on when to choose classic or modern projects, check Choosing the project type.
