UiPath Document Understanding

From PRIMO.ai
Jump to navigation Jump to search

YouTube ... Quora ...Google search ...Google News ...Bing News


UiPath Document Understanding (often shortened to DU) is the intelligent document processing (IDP) capability of the UiPath Platform, built to read invoices, receipts, forms, identity documents, contracts, and other business paperwork and turn them into structured data that software robots and AI agents can act on. It combines robotic process automation with optical character recognition (OCR), machine learning classifiers and extractors, UiPath's own fine-tuned large language model called the Helix Extractor, and prompt-driven generative extraction, and it can handle scanned images, PDFs, handwriting, signatures, checkboxes, and tables. Every extraction comes with a confidence score, so low-confidence fields can be routed to a person for review in Validation Station while high-confidence documents flow straight through. Since 2025, Document Understanding has sat inside UiPath IXP (Intelligent Xtraction and Processing), an umbrella that also includes Communications Mining for emails and messages and a generative extraction capability for long, unstructured, and complex documents. For background on the wider field, see Document Understanding AI and Documents.

Document processing matters because most business processes still begin with a document: an invoice that must be paid, a claim that must be adjudicated, a loan application that must be checked, or a referral that must be routed. Think of Document Understanding as the mailroom and data-entry department of an automated company: it opens each envelope, decides what kind of document is inside, copies the important details into the right fields, and sets aside anything smudged or unusual for a human clerk to check. UiPath, a New York-based automation company founded in Bucharest in 2005, has been named a Leader in Gartner's Magic Quadrant for Intelligent Document Processing in both 2025 and 2026, and it reported annual recurring revenue (ARR) of $1.938 billion for the quarter ended July 31, 2026. Customer results cited by UiPath include Omega Healthcare automating more than 100 million healthcare transactions at 99.5% accuracy and Thermo Fisher Scientific processing 53% of its invoices with no human intervention.


UiPath Document Understanding turns the paper trail of the enterprise into trusted, structured data, pairing specialized and generative AI models with human validation so that robots and agents can act on documents with measurable confidence.


History and Company

UiPath was founded in 2005 in Bucharest, Romania, by Daniel Dines and Marius Tîrcă, grew into one of the leading vendors of robotic process automation, and listed on the New York Stock Exchange in April 2021 under the ticker PATH. Document Understanding grew out of a practical problem: software robots were good at clicking through applications but struggled whenever the input was a scanned or variable document. UiPath packaged its answer as a framework inside the IntelligentOCR activities package, organized around a taxonomy of document types and fields plus a standard sequence of digitization, classification, extraction, validation, and training steps. The product reached first availability in February 2020, and by mid-2020 UiPath was selling it alongside AI Fabric (later renamed AI Center) for hosting machine learning models and Action Center for human review tasks.

The framework design deliberately allowed mixing techniques in one workflow, so a team could use regular expressions for one document type, a template-based form extractor for another, and a machine learning model for a third. In June 2023 Document Understanding became available through a cloud-hosted REST API as an alternative to calling it from RPA workflows, and UiPath's 2023.10 platform release introduced an enhanced experience for building and monitoring projects, which developed into what UiPath calls Document Understanding modern projects, with Build, Publish, Measure, and Monitor stages. The older framework, now called the classic experience, remains supported for existing automations. At its AI Summit in March 2024, UiPath announced two specialized LLMs, DocPATH for documents and CommPATH for communications, fine-tuned on industry-specific data and built into Document Understanding and Communications Mining.

DocPath reached Document Understanding modern projects for US-based tenants on October 31, 2024, and was renamed the UiPath Helix Extractor in January 2025 as part of a portfolio alignment. In March 2025, at its Agentic AI Summit, UiPath launched IXP to bring Document Understanding, Communications Mining, and a new generative extraction capability for unstructured documents under one roof, closely tied to its new agent-building and orchestration tools. In April 2025 UiPath also retired support for its oldest out-of-the-box OCR, extraction, and classification packages, pushing customers toward modern projects and the newer models. Founder Daniel Dines returned to the chief executive role in mid-2024 and has since repositioned the company around agentic automation and business orchestration, with document processing presented as the step that feeds reliable data to agents.

How Document Understanding Works

Every Document Understanding solution starts with a taxonomy that lists the document types to be processed and the fields to capture from each, such as invoice number, vendor name, line items, and total amount. The pipeline then runs in a predictable order: digitization converts a file into machine-readable text and layout using an OCR engine, classification decides which document types are present and where one document ends and the next begins in a multi-document file, and extraction fills in the fields for each identified document. Each extracted value carries a confidence score and a reference to where it was found on the page, which makes results auditable and lets automations decide what to trust. Validation then sends uncertain results to a person, and the corrections can be fed back through training so the models improve over time.

Digitization can use UiPath Document OCR, which UiPath updates frequently for handwriting, signatures, logos, and Arabic numerals, as well as an extended-language OCR covering more than 200 languages and a dedicated Chinese, Japanese, and Korean engine, and third-party engines such as OmniPage can be plugged in. Classification options range from simple keyword-based and intelligent keyword classifiers to machine learning classifiers, modern project classifiers, and a Generative Classifier that uses a large language model and plain-language descriptions of each document type. Extraction options include the RegEx Based Extractor for predictable text, the Form Extractor and Intelligent Form Extractor for fixed layouts with handwriting or signatures, machine learning extractors and Forms AI for semi-structured documents, the Helix Extractor in modern projects, and a Generative Extractor that answers field-by-field instructions written in plain English. This menu of techniques is the core design idea: cheap, deterministic methods handle the easy documents, and more expensive AI models are reserved for the variable ones.

Modern projects simplify this work considerably compared with the classic framework. Teams upload sample documents to a project in Automation Cloud, choose predefined document types or create their own, annotate a few examples, and let the project train automatically, with no separate pipeline runs and no AI Units charged for training or serving models. The Measure stage reports project and field-level scores so teams can see which fields need more examples, and published versions can be tagged (for example, Staging or Production) and called from UiPath Studio, Studio Web, coded workflows, or the REST API. Developers can still drop down to the classic IntelligentOCR scope activities when they need custom classifiers or extractors, and the framework's contracts packages let partners build their own components.

AI Models and Extractors

The Helix Extractor is the foundation model of modern Document Understanding projects. According to UiPath Research, earlier Document Understanding models used encoder-only Transformer networks that classified individual tokens as belonging to a field, followed by post-processing rules to stitch the spans together, whereas Helix Extractor is a fine-tuned generative LLM that reads document text plus special positional tokens and outputs structured JSON. Because a generative model does not naturally point to where an answer came from, UiPath built a separate method to attribute each predicted field back to the page, and it derives confidence scores from the model's token probabilities. UiPath says the model improves accuracy on semi-structured documents such as invoices, receipts, and purchase orders, handles tables better, needs fewer annotated examples for new layouts, and produces confidence scores that track accuracy more closely, which lets more documents skip human review at the same error rate.

Helix Extractor powers a large library of predefined document types in modern projects, including invoices, receipts, purchase orders, US tax forms such as the 4506-T, 941-X, and 9465, and ACORD insurance forms such as the ACORD 25 and 125, while a few types such as financial statements and Chinese, Hebrew, and Japanese invoices still run on the previous generation of models. UiPath's documentation in early 2026 still listed the Helix Extractor as available only to US-based tenants, excluding GxP and Government Cloud environments, so customers elsewhere rely on the earlier models or on generative options. Existing project versions keep their original models until a team deliberately retrains them, which protects production automations from unexpected changes. Separately, a Predefined Non-Latin Languages project type covers Japanese and Chinese invoices and Japanese receipts.

For documents that do not fit a predefined type, the Generative Predefined project type exposes pre-trained generative extractors (Long Document Simple Layout, Long Document Complex Layout, and Short Document Complex Layout) that take a dictionary of field names and plain-language instructions, an approach closer to prompt engineering and in-context learning than to model training. Field names are limited to 30 characters, instructions to 1,000 characters, and the Short Document Complex Layout extractor to 20 pages, while the Extract Document Data activity accepts files of up to 500 pages. Generative Validation, introduced in 2024, adds a second LLM pass that checks extracted values so that fewer high-confidence fields need human review, at an extra charge per page. Administrators can govern these features through UiPath's AI Trust Layer policy, which controls who may access the AI-powered features in Document Understanding.

UiPath IXP and Agentic Automation

UiPath IXP (Intelligent Xtraction and Processing) is UiPath's umbrella for converting unstructured documents and communications into structured data for automation and analytics. It offers three capabilities matched to different content: Communications Data (Communications Mining) for emails, tickets, and messages, Structured and semi-structured documents (Document Understanding) for forms, purchase orders, applications, and invoices, and Unstructured and complex documents (Generative Extraction) for contracts, leases, annual reports, brokerage statements, and other long or highly variable files. Document Understanding remains a standalone service in Automation Cloud with the same access and governance, and it can also be opened from within IXP, while Communications Mining is now reached only through IXP. UiPath says its long-term goal is a single unified experience that can extract data from mixed content such as an email together with its attachments.

The unstructured and complex documents capability is designed for files with paragraphs of free text, complex tables, charts, checkboxes, signatures, handwriting, inferred values that must be derived from context, and stacks of mixed documents that would otherwise need to be split first. Instead of annotating hundreds of examples, builders define an extraction schema and write prompt instructions, review predictions against a small set of annotated documents, and iterate while IXP reports precision, recall, and F1 scores. Builders can choose among several LLMs and settings such as chunk size and zero-shot or few-shot prompting, and IXP combines techniques including retrieval-augmented generation to handle long documents. In April 2026 UiPath made IXP available on Google Cloud Marketplace and made Google Gemini the default third-party model for new IXP projects, reporting about 40% faster predictions, up to 15% higher F1 scores than other third-party LLMs in its evaluations, and average processing time per document falling from 40 to 25 seconds.

IXP is also the document layer of UiPath's agentic strategy. Analysts at Deep Analysis observed at the launch that IXP's generative extraction was not new for the IDP market, but that its tight integration with UiPath's Agent Builder and orchestration tools gave AI agents direct access to document reading and understanding. In 2026 that integration deepened through an Extract node in UiPath Maestro Flow that runs IXP models as a deterministic step on a process canvas with every LLM request audited, IXP models packaged as resources inside UiPath Solutions, and a preview uipath-ixp skill in UiPath for Coding Agents that lets coding assistants create IXP projects, review predictions, refine prompts, and publish. UiPath's industry solutions, such as the agentic Purchase-to-Pay solution announced in March 2026 for procurement and accounts payable, combine AI agents, IXP, governed exception handling, and Maestro orchestration in one package.

Human Validation and Governance

Human review is built into Document Understanding rather than bolted on, reflecting a human-in-the-loop design. Validation Station shows a reviewer the original document beside the extracted fields, highlights where each value was found, and lets the reviewer correct values, fix classifications, or reject a document, and it is available in classic and compact display modes with right-to-left language support. In unattended automations, the Create Validation Task and Wait activity creates a validation task in Action Center and suspends the workflow until a person completes it, and a confidence threshold can limit the fields a reviewer must check. A July 2026 preview added validation artifacts that applications can use to create UiPath App Tasks, giving teams control over when, where, and with which interface the review happens.

Governance features have grown as document AI has moved into regulated industries such as banking, insurance, and healthcare. Runtime permission checks, generally available since March 2026, restrict who can discover and run models in a project to users, groups, and robots holding a role such as DU Automation User, DU Developer, or DU Project Administrator. Modern projects can be versioned, tagged, and referenced across organizations and tenants, and IXP adds field-level source attribution, audit trails, controlled releases, and rollback. Deployment options include Automation Cloud, Automation Cloud Public Sector, Automation Cloud Dedicated, and the on-premises Automation Suite, although modern projects on Automation Suite require GPU infrastructure and some generative extractors are not available in every environment, which matters for data governance, privacy, and compliance planning.

UiPath Document Processing Offerings

UiPath's document AI is spread across several products, models, and activity packages, which can be confusing for newcomers. The table below summarizes each offering as of late 2026 and how it relates to Document Understanding. In practice, most new projects start in IXP or Document Understanding modern projects and are called from Studio workflows, agents, or Maestro processes.

Offering What it is Notes
Document Understanding modern projects Cloud-based projects to build, measure, publish, and monitor classification and extraction models Recommended starting point; no AI Units charged for training or serving models
Document Understanding classic experience The original framework of taxonomy, digitize, classify, extract, validate, and train activities Still supported; built on the IntelligentOCR activities package
UiPath Helix Extractor UiPath's fine-tuned extraction LLM, formerly called DocPath Used by predefined and custom document types in modern projects; documented as US-only in early 2026
Generative Classifier, Generative Extractor, and Generative Predefined projects LLM-based classification and instruction-driven extraction with no training Field instructions written in plain language; subject to page and length limits
Generative Validation A second LLM pass that checks extracted values Reduces manual review at an extra per-page charge
UiPath Document OCR and other OCR engines Digitization for printed text, handwriting, signatures, and more than 200 languages Third-party OCR engines can be plugged into the framework
UiPath IXP Umbrella for Communications Mining, Document Understanding, and generative extraction for unstructured and complex documents Gemini is the default third-party model for new IXP projects since April 2026
Communications Mining Intent, entity, and tone extraction from emails, tickets, and messages Accessed through IXP as the Communications Data capability
Validation Station, Action Center, and App Tasks Human review interfaces and task routing Validation artifacts for App Tasks in preview since July 2026
Maestro Flow Extract node Runs IXP extraction models as a step in an orchestrated flow Every LLM request passes through the AI Trust Layer and is audited
UiPath for Coding Agents (uipath-ixp skill) Lets coding agents build, test, and publish IXP projects IXP support in preview; UiPath for Coding Agents itself became generally available in September 2026
Activity packages and APIs UiPath.DocumentUnderstanding.Activities, UiPath.IntelligentOCR.Activities, UiPath.PDF.Activities, and the Document Understanding REST API Coded workflow support added in the 26.7 preview packages of July 2026

Pricing

Document Understanding is consumed through UiPath's platform licensing rather than sold as a separate per-seat product, and charges depend on whether a customer is on the older Flex Plan (AI Units) or Unified Pricing (Platform Units). On July 8, 2026, UiPath switched modern projects from a flat per-page rate to per-operation charging: digitization is now free, classification and extraction are charged individually, and the total remains capped at 1 AI Unit or 0.2 Platform Units per page. The same update allowed Document Understanding and IXP licenses to be used together with no additional cost per page. On Unified Pricing, Document Understanding is included in the Basic, Plus, and Pro platform tiers but not in Express, and the cost in currency of each AI Unit or Platform Unit depends on the customer's plan and contract.

Capability Flex Plan (AI Units) Unified Pricing (Platform Units) Notes
Modern projects (digitize, classify, extract) 0 for digitization, 0.2 per page for classification, 0.8 per page for extraction after classification or 1 without 0.2 per page Never more than 1 AI Unit per page; per-operation charging since July 2026
Generative Validation 1 additional per page 0.2 additional per page Optional LLM check of extracted fields
Classic projects Charged per component used 0.04 per page for digitization, 0.2 for ML extraction or Forms AI, 0.04 for the Form Extractor, 0 for the RegEx extractor Classification is charged by page ranges and capped at 1 Platform Unit per document
IXP unstructured and complex documents 1 per page for extraction, plus 1 per page for generative or modern classification 0.2 per page at runtime Prompt-based generative extraction for long and variable documents
Model training and serving in modern projects 0 0 Teams can retrain and experiment without consumption charges
Validation Station and Taxonomy Manager 0 0 Human review itself does not consume units

Getting Started: Extract Invoice Data in Studio

The quickest way to see Document Understanding work is to extract fields from a few invoices with a generative extractor, because it needs no taxonomy design, annotation, or model training. You need a UiPath Automation Cloud account (a trial or Community account is fine for experimenting), UiPath Studio connected to that account, and a folder with two or three sample PDF invoices of a few pages each. Make sure your tenant has AI Units or Platform Units available for Document Understanding, which an administrator can confirm under Admin and Licenses in Automation Cloud. Community accounts have historically carried tight limits on document size and request rates, so keep the samples small and expect occasional throttling.

With those prerequisites in place, build the workflow in five steps. The expression in the second block below is written in Visual Basic, the default language for most Studio projects; if your project uses C#, replace each ampersand (&) with a plus sign (+). If you prefer working in a browser, the same activities are available in Studio Web.

  1. Create a project. In Studio, create a new Process named InvoiceReader and confirm under Manage Packages that the UiPath.DocumentUnderstanding.Activities package is installed.
  2. Add the extraction. Add an Extract Document Data activity, set its Input to one of your sample invoices (use the property's plus menu to browse to a local file), set Project to Generative Predefined, choose the Short Document Complex Layout extractor, and name the output variable DocumentData; if a Generate Data Type option is shown, leave it set to False.
  3. Describe the fields. In the extractor's Dictionary, add the field names and plain-language instructions shown in the first block below.
  4. Show the results. Add a Log Message activity after the extraction and paste the expression from the second block below, which reads each value and its confidence with the GetFieldValue method.
  5. Run and review. Run the workflow and check the Output panel, which should show one line with the invoice number, vendor, total, and a confidence percentage; then point the Input at your other samples to compare results.
Field name      Instruction
InvoiceNumber   What is the invoice number?
InvoiceDate     What is the invoice date? Return it as YYYY-MM-DD.
VendorName      What is the name of the company that issued the invoice?
TotalAmount     What is the total amount due, including tax? Return only the number.

"Invoice " & DocumentData.Data.GetFieldValue("InvoiceNumber").Value &
" from " & DocumentData.Data.GetFieldValue("VendorName").Value &
" totals " & DocumentData.Data.GetFieldValue("TotalAmount").Value &
" (confidence " & DocumentData.Data.GetFieldValue("TotalAmount").Confidence.ToString("P0") & ")"

If a value comes back empty or wrong, the fix is usually a clearer instruction rather than more code, for example naming the label the document actually uses ("Amount Due" or "Balance") or telling the model which value to prefer when several totals appear. Once results look right, add a human review step by placing Create Validation Task and Wait after the extraction, pointing it to an Orchestrator storage bucket, and setting Extracted fields validation confidence % to around 80 so reviewers focus on uncertain fields. This activity creates a validation task in Action Center and suspends the workflow until a person completes it, so the project must support long-running workflows and be published to Orchestrator. From there, natural next steps are looping over a folder of files, writing the validated values to a spreadsheet or business system, switching to a predefined Invoices type in a modern project for higher-volume work, or moving complex contracts into IXP's unstructured and complex documents capability.

Why This Matters

Document Understanding addresses one of the oldest and least glamorous bottlenecks in business: people retyping information from documents into systems. For finance and accounting teams it automates invoice and receipt capture, for insurers it reads claims and ACORD forms, for healthcare providers it handles referrals and prior authorizations, and for supply chain, human resources, legal, and government teams it processes purchase orders, applications, contracts, and case files. The combination of confidence scores, human validation, and retraining gives managers a way to raise automation rates gradually without losing control of accuracy, which is often the deciding factor in regulated industries. Because UiPath also sells the robots, agents, and orchestration that act on the extracted data, its document tools appeal to organizations that want one platform from document intake to completed transaction.

The trade-offs deserve equal attention. Generative extraction makes it easy to start, but LLM outputs can still be wrong or inconsistent, so teams need ground-truth test sets, evaluation metrics, and sensible confidence thresholds rather than trusting a demo. Regional availability varies, with the Helix Extractor documented as US-only in early 2026 and some generative features unavailable on-premises or in public-sector clouds, and usage-based pricing in AI Units or Platform Units can be hard to forecast for large, multi-page volumes. Many organizations also weigh UiPath against general-purpose cloud services such as Amazon Textract, Google Document AI, and Microsoft Azure AI Document Intelligence, or against building their own extraction pipelines on frontier models, which UiPath counters by emphasizing governance, validation, and integration with automation.

For readers exploring where document AI fits among broader automation and agent tools, the Visual and Low-Code Agent Builder Comparison and the Ready-to-Use AI Agent Comparison place UiPath alongside other platforms. Document processing increasingly serves as the input layer for agents: an agent can only reason about a contract or claim once the document has been reliably converted into data. That is why UiPath now describes intelligent document processing less as a standalone product and more as the foundation that makes agentic automation trustworthy.

Recent Developments

Between 2025 and late 2026, UiPath turned Document Understanding from a standalone IDP product into one capability within IXP and its agentic automation platform, while steadily upgrading models, governance, and developer tooling. Most of these changes are documented in UiPath's release notes, which list Document Understanding and IXP updates almost every month. Readers following older tutorials should expect different menus, project types, and pricing rules. The most significant developments were:

  • IXP Launch (March 2025): At its Agentic AI Summit on March 25, 2025, UiPath launched IXP to combine Document Understanding, Communications Mining, and generative extraction for unstructured and complex documents, initially in private preview, with tight integration into its new Agent Builder and orchestration tools.
  • Retirement of Legacy Models (April 2025): UiPath removed support for its oldest out-of-the-box OCR, data extraction, and document classification packages, encouraging customers to move to modern projects and newer models.
  • Unstructured Documents and Gartner Recognition (August to September 2025): The unstructured and complex documents capability appeared in IXP release notes in August 2025, Gemini was added as an extraction model option in September, and UiPath was named a Leader in Gartner's inaugural Magic Quadrant for Intelligent Document Processing Solutions.
  • Regional Expansion and Governance (January to March 2026): IXP's unstructured documents and Communications Mining capabilities became generally available in the United Kingdom region in January 2026, runtime permission checks for Document Understanding projects became generally available on March 3, and new activity releases added project tags, versions, and cross-tenant connections.
  • Agentic Purchase-to-Pay (March 2026): UiPath announced an agentic Purchase-to-Pay solution for procurement and accounts payable that combines AI agents, IXP, governed exception workflows, and Maestro orchestration, with controlled general availability to follow.
  • IXP on Google Cloud Marketplace (April 22, 2026): IXP became available on Google Cloud Marketplace, Gemini became the default third-party model for new IXP projects, and UiPath reported faster predictions, higher F1 scores, and a drop in average processing time from 40 to 25 seconds per document.
  • Forrester Recognition (May 2026): UiPath was named a Leader in The Forrester Wave for Document Mining and Analytics Platforms, Q2 2026, a result UiPath attributed to IXP's combination of document processing, agents, orchestration, and human validation.
  • Pricing and Validation Changes (July 2026): On July 8, Document Understanding moved to per-operation charging with free digitization and allowed Document Understanding and IXP licenses to be combined, and on July 21 a preview introduced validation artifacts for App Tasks.
  • Developer Tooling (July 2026): The Document Understanding 26.7 preview added coded workflow support, new PDF conversion activities for email, HTML, and text, an activity for extracting embedded e-invoice XML files, per-field redaction options, and IXP models as Solution resources, while UiPath for Coding Agents entered preview with an IXP skill.
  • Gartner Leader Again and OCR Updates (September to October 2026): UiPath was named a Leader in the 2026 Gartner Magic Quadrant for Intelligent Document Processing for the second consecutive year, and a series of UiPath Document OCR releases improved handwriting recognition, cut processing latency by 30%, and recalibrated handwriting confidence scores, which UiPath warned may send more handwritten documents to validation.
  • FUSION 2026 (September 22 to 25, 2026): At its annual conference in Las Vegas, UiPath made UiPath for Coding Agents and the Cartographer process-knowledge tool generally available and announced Maestro Orchestrate and Maestro Automate, new governance controls such as Model Hub and LLM-as-Judge guardrails, and Automation Suite on Linux for on-premises deployments, all of which shape how document workflows are built and governed.
  • Debates Over Build Versus Buy, Accuracy, and Growth: Supporters argue that IXP's governance, validation, and integration with robots and agents make it safer than building custom LLM extraction pipelines. Critics note that generative extraction features are now common across IDP vendors and cloud providers, that LLM-based extraction still needs careful evaluation, and that regional limits and usage-based pricing complicate global rollouts. Investors have also watched whether AI products can reaccelerate UiPath's growth, with ARR rising 12% year over year to $1.938 billion in the quarter ended July 31, 2026.

Additional Viewing

Load video
YouTube

In this complete guide, UiPath and Microsoft MVP Anders Jensen builds an end-to-end Document Understanding automation that watches an inbox, downloads PDF invoices, and extracts key invoice data using the digitize, classify, and extract steps. He explains how the taxonomy defines document types and fields and how the same structure scales to purchase orders, receipts, and other documents. It is a practical companion to the How Document Understanding Works and Getting Started sections.

Load video
YouTube

This beginner-level introduction from Tutorials by Mukesh Kala opens a step-by-step series on UiPath Document Understanding and intelligent document processing. It introduces Document Understanding and the intelligent document processing concepts behind it, and later videos in the series build each step of a working process. It is a good first video for readers with no prior exposure to the product.

Load video
YouTube

In this full tutorial, Marcelo Cruz shows from scratch how to process and extract data from several document types using the classic Document Understanding framework in UiPath Studio. It walks through the framework's main stages in a single workflow rather than treating each activity in isolation. Because it uses the classic experience, it is most useful for understanding the underlying framework and for maintaining older automations.

Load video
YouTube

In this video from the UiPath with Jai channel, the presenter explains the Machine Learning Extractor in detail and shows how to configure it to pull data correctly from a document. It is a useful deep dive into one of the trained extractors described in the How Document Understanding Works section. It pairs well with the AI Models and Extractors section of this page.

References

Retrieved from "https://primo.ai/w/index.php?title=UiPath_Document_Understanding&oldid=52584"