Blog Post

XtractFlow: Using Generative AI to Make a Quantum Leap in Intelligent Document Processing

Jonathan D. Rhyne
Illustration: XtractFlow: Using Generative AI to Make a Quantum Leap in Intelligent Document Processing

The ORPALIS team has been part of PSPDFKit since 2022, but we’d already begun working on document processing and machine vision years before, and as such, we’d experienced its limitations, such as the necessity for predefined extraction rules and rigid templates. Even with those limitations, the first intelligent document processing (IDP) technology offered business benefits when compared to manual processing; however, we’ve always believed we could achieve more with it — more accuracy, more speed, more agility, and more intelligence — and require less setup and tuning for each workflow.

Today, we’re pleased to introduce the XtractFlow SDK and API from PSPDFKit — a generative AI-powered intelligent document processing engine.

XtractFlow simplifies IDP deployment by reducing it from days to hours, significantly reduces setup time for IDP workflows by eliminating the need for predefined rules or key-value pairs (KVPs), and enhances data extraction accuracy. Additionally, XtractFlow saves time and reduces complexity by enabling you to set document classification conditions and data extraction requirements using natural language.

Core Functionalities and Use Cases

XtractFlow is an engine that operates as a headless processor that easily integrates with large language models (LLMs) like ChatGPT from OpenAI or Azure OpenAI. It’s designed to be compatible with most available LLMs, enabling you to select the generative AI provider that best meets your business needs.

XtractFlow brings human-level accuracy and intelligence to document classification and data extraction in a way that many people have always expected.

  • AI-Powered Document Processing — XtractFlow employs generative AI, using OpenAI and Azure OpenAI for document classification and data extraction in its current version.

  • Supports Hundreds of Formats — XtractFlow efficiently extracts data from hundreds of document formats, including PDF, JPEG, Office, and CAD files, regardless of document complexity.

  • Customizable, with Robust Data Integrity — It features tailormade components and extensive validation rules for diverse industries, ensuring accuracy in data elements like postal addresses, bank account numbers, emails, and more.

  • Integration and Compliance — XtractFlow is available as a .NET SDK that can be used to create a REST microservice or API suitable for global hosting, without storing documents or extracted content.

The engine uses natural language as input for document classification and data extraction instructions. This means you can describe the types of data you want to extract and the documents to classify in the same way you’d express this information to a colleague. For example, you can say, “Extract the customer’s name, address, ID number, and signature date,” and XtractFlow will identify and extract the data, regardless of its location in a document.

XtractFlow can categorize documents into predefined or custom semantic-based templates based on their content and structure. It easily classifies a variety of documents — including invoices, contracts, legal filings, medical records, academic papers, and more — from any unstructured document storage of your choice. This saves time by eliminating the need to set classification rules and identify the correct documents manually. Additionally, you can edit the predefined templates to adjust them for specific use cases.

Extract Virtually Any Data Found in a Document

XtractFlow is built to extract a multitude of data. Examples include but aren’t limited to:

Textual Data

Numerical Data

Identification Numbers

Workflow-Specific Data

Names, addresses, emails, descriptive paragraphs, and other freeform fields.

Dates, monetary values, quantities, statistical data, and other numerical information.

Social Security numbers, account numbers, invoice numbers, employee IDs, and other alphanumeric identifiers.

Various semantic text structures, such as insurance policy details, medical codes in healthcare, legal clauses in contracts, and educational qualifications.

By utilizing generative AI, XtractFlow identifies and extracts data without the need for rigid extraction patterns, reducing reliance on traditional IDP technology. Generative AI significantly enhances efficiency and accuracy in any data extraction workflow, saving valuable time and effort.

XtractFlow vs. Traditional IDP

XtractFlow delivers a quantum advance in simplifying deployment, implementation, and accuracy when compared to traditional IDP technologies. XtractFlow dramatically reduces the go-to-market time from days to hours, along with cutting down on the investment needed for IDP applications. It also simplifies setting up new classification and data extraction workflows, avoids the requirement for specialized skills or training, and boosts accuracy to prevent expensive errors.

Traditional IDP XtractFlow

Capability Scope

Extracts prequalified/categorized entities and all key-value pairs.

Extracts strictly and semantically predefined entities; suitable for defined key association tasks; faster development with NLP.

Processing Time and Architecture

High-performance processing speed when using tailored architectures and local inference.

Flexibility to scale performance as needed based on the hardware selected.

Accuracy and Development Time

Lower initial accuracy (~70 percent on a predefined dataset), longer development time (~7 days), plus ongoing maintenance.

High accuracy (>90 percent for invoices), with solution deployment within a day.

Cloud-Based vs. Local Inference

Adaptable to both cloud-based solutions and local inference.

Adaptable to both cloud-based solutions and local inference.

Use Case Suitability

Better for extracting categorized entities or key-value pairs; suitable for more supervised approaches.

Ideal for quick, accurate extraction of predefined/semantically defined information; versatile across a wide range of document templates.

With its semantically driven instructions and no need to predefine extraction rules for specific document templates, XtractFlow overcomes the most challenging hurdles associated with IDP.

  1. Diverse Format Processing

Simplifies the classification and extraction of data from various formats, including PDFs, Office files, and images.

  1. Unstructured Data Storage Issues

Streamlines the extraction of information from inconsistently stored documents (e.g. invoices mixed with contracts), including documents with varying structures and years.

  1. Challenges Recognizing Document Type

Improves accuracy in identifying and extracting data from a wide range of document types, such as applications, invoices, contracts, and patient intake forms.

  1. Complexity of Targeting the Correct Data to Extract

While key-value pair extraction excels at retrieving defined data, XtractFlow addresses the complexity of extracting non-explicitly defined information.

The core XtractFlow technology is built upon an access point design that has been specifically engineered to ensure long-term compatibility, regardless of the architectural changes in your solutions or the XtractFlow engine. This approach streamlines both the design and integration processes and greatly reduces latency for client-server applications.

Advantages of XtractFlow Over the ChatGPT API

Many developers may wonder why they should use XtractFlow rather than the widely accessible ChatGPT API.

While OpenAI’s ChatGPT adeptly processes documents, extracting pivotal data like invoice numbers and supplier details in JSON, XtractFlow also excels at processing speed, supplemented by advanced features such as document classification, ready-to-use models, and data validation. XtractFlow boasts several advantages, including enhanced privacy and security features, particularly with the upcoming local inference capability.

Feature Available Now (January 2024) XtractFlow: OpenAI ChatGPT 3.5 OpenAI GPT-4 Turbo (Vision API) [Coming Soon] XtractFlow: Local Inference

Supported Formats

Images, PDF, Office


Images, PDF, Office

Document Classification Support




Multipage Document Support




Average Processing Time Per Page

5 seconds (1 second on 8-core systems)

5–10 seconds

2–4 seconds

Internet Access Requirement




Price Per Page

$0.0015 (on average)

$0.01105 for 200 DPI A4


Data Privacy

OpenAI Enterprise Privacy

OpenAI Enterprise Privacy

No data is shared

Multiple-Language Support




Maximum Input File Size

No limit

20 MB

No limit

Predefined Models Available




Data Validation Support




Key-Value Pair Extraction




API-Forward Compatibility




Effort Required to Build Application




XtractFlow represents a new era in designing IDP workflows using no-code components. This innovation makes integrating and utilizing the most powerful IDP technology easier and more cost-efficient than ever before.

After three months in private preview, XtractFlow is now available for you to try for free. Our Solutions Engineer and Sales teams are always eager to address any questions you might have and provide a complimentary trial license, so please don’t hesitate to reach out.

We’re leading a revolution to evolve the human experience with documents. Join us and stay ahead of the curve!


To showcase the full potential of XtractFlow, we invite you to join our exclusive webinar on 8 February 2024 (17:00 CET/11:00 AM ET). This is a fantastic opportunity to see XtractFlow in action and understand how it can revolutionize your document processing workflows.

Jonathan D. Rhyne Co-Founder and CEO

Jonathan joined PSPDFKit in 2014. As CEO, Jonathan defines the company’s vision and strategic goals, bolsters the team culture, and steers product direction. When he’s not working, he enjoys being a dad, photography, and soccer.

Share Post
Free 60-Day Trial Try PSPDFKit in your app today.
Free Trial