Session 5 - Specialized AI Associate Series: Build a Document Understanding Automation in Studio

🚀 Welcome to Session 5/ AI Associate Developer Series 2025!

In this session, we will learn how to use the UiPath Document Understanding framework and the specialized AI capabilities for intelligent document processing in real-life scenarios.

📕 Agenda:

Introductions
Recap of DU Framework and Components
Setting up UiPath Studio
Project Requirements and Solution Design
Develop using the Document Understanding framework
Zoom in: the Studio activity packages for document understanding
Q&A
For links to the self-study modules, go here: https://academy.uipath.com/learning-plans/specialized-ai-associate-training, and click on the second tab.

1. Build a Document Understanding Automation in Studio AI Associate Developer Series 2024 Session 5

2. Introduction Recap – DU Framework and Components Automations in Document Understanding DU in Studio Web – Simple Processes DU in Studio Desktop – Medium to Complex Processes Project Requirements and Solution Design Develop using the DU Framework Zoom in: The Studio activity packages for document understanding Confidence Explained Additional Resources Questions Upcoming Sessions Agenda

3. Technology Leader, AI & Automation qBotica Director, Intelligent Automation Anika Systems Intelligent Automation Lead Anika Systems Russel Alfeche Chris Bolin Mason Turvey Team Slide

4. Recap – Document Understanding Framework and Components

5. Automations in Document Understanding IntelligentOCR Activities Package DocumentUnderstanding Activities package Deployment Automation Cloud Automation Suite Standalone Automation Cloud Automation Suite Best Suited For RPA Developers RPA Developers Citizen Developers Integrated Development Environment Studio Desktop Studio Desktop Studio X Studio Web Benefits Flexibility You can mix extraction and classification models, and can also use extractors and classifiers as fallback You can modify the taxonomy and extraction results using RPA code during run-time. Extensible and open framework You can bring your own classifier, extractor, or OCR engine using the respective interfaces. You have full control over the configuration as an RPA developer. Document Understanding Process Template based on REFramework. Ease of adoption: Easy to use, available on cloud, no setup required for consuming out-of-the-box models. Can be consumed using the  Create Automation  option in Document Understanding and Marketplace. Suggested by UiPath®  Autopilot TM  in workflows. Seamlessly integrated with Document Understanding modern projects, isolated configuration in a Document Understanding project, enabling reusability. Relying on Document Understanding cloud APIs, leading to quicker bug fixes. Single input/output object,  Document Data .

6. Automations in Document Understanding …cont IntelligentOCR Activities Package DocumentUnderstanding Activities package Drawbacks High learning curve Complex configuration, reducing reusability Passing explicit arguments from one activity to the other repeatedly: Taxonomy Document Object Model Text Classification reults Extraction results Compared to IntelligentOCR , there are some missing features, which are planned to be added: Splitting Business Rules Training (fine-tuning models) Support for multiple extraction methods per document type Use Case Use if you have an existing automation relying on it for which adopting DocumentUnderstanding.Activities is either impossible (functionality currently not available), or costly. Use this package if you are just getting started with Document Understanding. Use for new automations relying on: Modern projects Generative capabilities Out-of-the-box specialized models

7. DU in Studio Web – Simple Extraction Process

8. Studio Web – Create a new automation starting from a file

9. DU in Studio Desktop – Medium to Complex Extraction Process via DU Framework

10. 1. Create a blank process in UiPath Studio 2. Install the required activities packages 3. Test your setup Setup UiPath Studio

11. Source Where does the document come from? Single or multiple sources? File structure Is it fully structured, semi-structured or unstructured file? Classification Whether classification is required Extraction What information is needed Validation Decide on what validation methods are used Post Processing Where do we upload this information Project Requirements

12. Solution Design

13. Create and modify a taxonomy and load it into the automation project. Digitize documents by choosing the right OCR engine. Use and configure classifiers for document classification. Use and configure extractors for retrieving data from the classified documents, in the extraction stage. Configure the human validation stages and the retraining features. Develop using the Document Understanding framework DEMO

14. The Load Taxonomy wizard. The Digitize Document activity using OCR. The Classify Document Scope activity. The Data Extraction Scope activity. The Classification Station and Validation Station. The Train Classifiers Scope and Train Extractors Scope activities. The Export Extraction Result activity. Zoom in: The Studio activity packages for Document Understanding

15. Confidence is a numerical value that is meant to output an algorithm's estimation of how well it has performed a certain task. Confidence Explained What is Confidence? Types of Confidence OCR confidence An estimation of its ability to recognize the characters Classification confidence An estimation of its ability to identify the type of document correctly Extraction confidence Field confidence Table Confidence Table Cell Confidence

16. Upcoming Session Session 6: The GenAI Experience in UiPath DU Session 7: UiPath Communications Mining Overview Session 8: Fundamentals of Model Training

17. Q&A

18. Thank you!