SERVER SDK · SMART DATA EXTRACTION
Document data extraction from complex PDFs, scans, forms, tables, and engineering drawings.
Most tools guess at layout and meaning in one pass. Apryse Smart Data Extraction recovers document structure deterministically, then applies purpose-built AI to interpret it, so extraction holds up on the complex, layout-heavy documents where generic tools break down. Embed it in your application as an add-on to the Apryse Server SDK.
Works with your stack
Apryse Smart Data Extraction runs natively in the languages and environments your team already uses.
Native bindings for .NET, Java, Python, Node.js, C++, Go, PHP, Ruby, and Objective-C, plus a C interface for other environments — on Windows and Linux (x64). Apple Silicon runs through the tested Linux/AMD64 Docker configuration.
Trusted for document workflows where accuracy, control, and deployment flexibility matter.
Smart Data Extraction at a glance
Apryse Smart Data Extraction is an embeddable, self-hosted SDK for offline document data extraction. It recognizes document structure, classifies pages, extracts key-value pairs and tables, and detects form fields, all returned as structured JSON.
When to consider Smart Data Extraction
Use it for complex documents with varied formats, files contain sensitive information, extraction must be embedded in your application, or per-page cloud pricing makes costs difficult to predict at scale. It is a self-hosted document extraction alternative to cloud document AI services and enterprise IDP platforms.
How the extraction pipeline works
Here's what happens between upload and structured output. Smart Data Extraction is a set of utilities, not a fixed pipeline: call the capabilities your documents require, skip the ones you don't.
1. Prepare the input
Normalize file formats, split or reorder bundled documents, redact sensitive information, and apply OCR or ICR when pages do not contain usable text. These preparation capabilities are available as an add-on to the same Server SDK license and can run in the same controlled environment.
2. Analyze document structure
The SDK identifies reading order and structural elements such as headings, paragraphs, tables, and form fields. This structural analysis is deterministic: the same input produces the same structural representation.
3. Classify documents and extract data
After the structure has been identified, document-focused models classify pages and extract the information required by the application. Depending on the workflow, this can include document types, dates, totals, account or invoice numbers, labeled values, table contents, form fields, or engineering metadata. This step handles PDF data extraction and automated data extraction from pdf without requiring predefined templates.
4. Return structured output
Extraction results are returned as structured JSON with confidence scores. Table data can be returned as JSON or exported to Excel, ready for routing into a database, business application, ERP system, search index, workflow engine, or LLM pipeline.
Why separate structure from interpretation?
Complex documents often contain multiple columns, nested tables, inconsistent spacing, repeated sections, scans, or several document types in one file. An extraction pipeline must preserve those relationships before it can reliably identify the information they contain.
Smart Data Extraction establishes a consistent structural representation first and then applies extraction to that representation. This is particularly useful for documents where flattening the page into plain text would lose reading order, table relationships, or label-to-value context. See how this works in the extraction workflow guide.
Why Apryse over alternatives
Teams evaluating document data extraction software typically compare four approaches: cloud document AI APIs, enterprise IDP platforms, open-source pipelines, and embedded SDKs. Smart Data Extraction is the embedded SDK approach, and it is built as an add-on to the Apryse Server SDK , and designed for a specific set of requirements.
Over cloud document AI APIs
Cloud APIs run on infrastructure you do not control. Every page leaves your environment, pricing scales per operation, and offline or air-gapped deployment is not available. Smart Data Extraction runs inside your application on your own infrastructure. Documents stay in your environment, processing runs on CPU rather than GPU, and licensing is a package rather than per-page metering. For teams in regulated industries or with data sovereignty requirements, that is often the deciding factor.
Over enterprise IDP platforms
Enterprise IDP platforms are designed for business-user workflows with low-code configuration, document queues, and review interfaces. Smart Data Extraction is an SDK add-on, not a platform. It gives developers direct control over the extraction pipeline, embedded inside the application they are building. Teams that want to own the pipeline architecture rather than adopt a broader platform choose the SDK approach. Learn more: Why document preparation goes beyond OCR →
Over open-source pipelines
Open-source components offer flexibility, but the team assembles, maintains, and troubleshoots the full pipeline. Smart Data Extraction provides a commercially supported, tested pipeline with deterministic structure recognition and purpose-built extraction models out of the box. The tradeoff is a license cost against engineering time and maintenance.
Over general-purpose AI models used alone
A general-purpose model guesses at document layout and meaning in a single pass, which is why results vary on complex, layout-heavy documents. Smart Data Extraction recovers the structure deterministically first, then applies purpose-built models to interpret it. That separation is what keeps extraction reliable on the documents where a single-model approach degrades: multi-column reports, nested tables, scanned pages, and mixed-document packets. See how Smart Data Extraction works with PDF to JSON with AI.
Use Smart Data Extraction with or without an LLM
Smart Data Extraction can run as the document-processing layer in an application, or as a preprocessing step ahead of an existing LLM-based workflow.
In an LLM-based pipeline, it can handle OCR, reading order, document structure, page classification, table recognition, and field extraction before structured data is passed downstream. This lets the application control which information reaches the model and can reduce the amount of source content that must be processed. This approach can cut the amount of source content the model has to process, and the token cost that comes with it. Teams using this as self-hosted document extraction alongside an LLM reduce both token cost and the number of pages requiring manual review.
Deployment and integration
Smart Data Extraction is embedded in your application and runs on infrastructure controlled by your organization. Deploy on-premises, in a private cloud, in a hybrid environment, or in a fully air-gapped system without sending document content to a public document AI service.
Predictable licensing for high-volume processing
Smart Data Extraction is licensed as an add-on package to the Apryse Server SDK rather than metered per page or extraction operation. This gives teams a more predictable commercial model as document volume changes. Infrastructure requirements will still depend on document types, scan quality, enabled capabilities, throughput targets, and the architecture of the application.
Prepare documents before extraction
Extraction quality depends in part on the source document. Apryse offers many capabilities that can prepare documents in the same environment before extraction begins.
Common use cases
Use Smart Data Extraction to add document processing to financial systems, claims workflows, contract applications, records platforms, engineering systems, and other products that need reliable, structured data from varied or layout-heavy documents. Extract data from PDF, DOCX, and scanned files with a modular SDK you embed directly in your application.
Financial services
Classify and extract data from statements, annual reports, prospectuses, filings, and other dense, multi-column financial documents. Extract data from PDF filings with complex layouts where structure must survive extraction.
Insurance
Process claims packets, loss forms, policies, and supporting records received as scans or mixed-document files.
Legal
Classify contracts, discovery materials, correspondence, and long-form case files while keeping privileged content inside controlled infrastructure.
Healthcare and life sciences
Extract information from clinical documents, lab reports, forms, and regulatory submissions in private deployment environments.
Government and public sector
Digitize records, classify forms, and process scanned archives on-premises or in disconnected environments.
Logistics and trade
Extract data from bills of lading, customs documents, invoices, and shipping records received from varied sources.
Engineering and manufacturing
Extract title block metadata from engineering drawings and process supporting business documents in the same application.
How Smart Data Extraction compares with other approaches
Cloud API services, enterprise IDP platforms, open-source components, and embedded SDKs can all support document extraction. The most important differences are how they integrate, where processing runs, who owns the workflow, and how costs are structured.
How to evaluate document extraction accuracy
Accuracy should be tested against the documents the application will process in production. A useful evaluation set includes common and uncommon layouts, different document sources, high- and low-quality scans, mixed packets, multi-page tables, similar document categories, and files with missing or incomplete fields.
Review more than the final extracted value. Evaluate classification confidence, field-level confidence, table structure, reading order, failure handling, throughput, infrastructure requirements, and the amount of manual review required.
Confidence scores can support automation thresholds: high-confidence results can continue through the workflow, while lower-confidence results can be routed for review. Confidence scores are not a substitute for testing. Results can vary with scan quality, document type, category similarity, and layout.
Frequently asked questions
Apryse Smart Data Extraction is a self-hosted SDK package for document classification, key-value pair extraction, table extraction, form field detection, OCR and ICR, and CAD title block extraction. It is embedded in an application and runs inside infrastructure controlled by the customer.
Yes. Document Classification assigns a category and confidence score at the page level, allowing mixed packets to be separated and routed. A broad predefined category set is available without training a model first.
Yes. Key-Value Pair Extraction identifies relationships between labels and values, such as “Invoice Number” and “INV-001,” without requiring fixed coordinates or predefined zones.
Yes. Table Extraction recovers rows, columns, and table structure from layouts that can include nested tables, multi-column pages, and tables that continue across pages. Results can be returned as JSON or exported to Excel.
Yes. Form Field Detection identifies and labels the fields themselves, including text fields, checkboxes, and signature fields. Key-Value Pair Extraction identifies relationships between labels and business values in free-form or semi-structured layouts.
Yes. OCR converts scanned and image-only documents into machine-readable text, and ICR supports handwritten content. Results depend on factors such as image quality, resolution, handwriting clarity, and layout.
Yes. Smart Data Extraction can run on-premises, in a private cloud, in a hybrid environment, or in a fully air-gapped system. Teams planning a disconnected deployment should confirm activation requirements during evaluation.
Smart Data Extraction supports Windows and Linux on x64, with bindings for .NET, Java, Python, Node.js, C++, Go, PHP, Ruby, Objective-C, and a C interface for other environments. Apple Silicon can use the tested Linux/AMD64 Docker reference configuration. See all supported platforms →
Extraction results are returned as structured JSON with confidence scores. Table data can be returned as JSON or exported to Excel.
Yes. It can provide OCR, document structure, classification, tables, and extracted fields before structured data is passed to an LLM or downstream application, saving token costs when LLM analysis is required. It can also be used without an LLM when its native extraction capabilities provide the required output. Review the guide for Augmenting LLMs with Smart Data Extraction →
It is an alternative for teams that need document extraction inside customer-controlled infrastructure rather than through a public cloud API. The right approach depends on deployment, integration, document types, pricing, and operational requirements.
Smart Data Extraction is an embeddable SDK package rather than a complete IDP solution. Teams that need document processing inside their own application may prefer an SDK; teams that need business-user workflow design may evaluate a full IDP platform. Learn more →
Smart Data Extraction is licensed as an add-on package to the Apryse SDK rather than metered per page or extraction operation. Contact Apryse for pricing based on the required capabilities and deployment.
Start a free trial to test Smart Data Extraction on representative documents, or request a guided evaluation for help assessing document categories, extraction requirements, deployment architecture, and difficult source files.
Documentation and resources
Documentation
- Smart Data Extraction — overview and setup
- Document Classification
- Key-Value Pair Extraction
- Form Field Detection
- Tabular Data Extraction
- Document Structure Recognition
- CAD Title Block Extraction — see sample code
- SDK Reference and APIs
- Download / Supported Platforms
- Data Extraction Module add-on
Guides and further reading
- Smart Data Extraction: The Complete Guide
- Why Document Preparation Goes Beyond OCR
- PDF to JSON with AI: Smart Data Extraction for Structured Data
- Introducing Document Classification in Smart Data Extraction
- Intelligent Document Processing for AI Readiness
- Augmenting LLMs with Smart Data Extraction (RAG guide)
- Offline PDF Document Classification: Cloud AI vs. Open Source vs. Commercial SDKs
- PDF Key-Value Pair Extraction: Cloud AI vs. Open Source vs. Commercial SDKs
- Does Apryse Handle Flattened and Scanned PDF Forms
- Apryse vs. Nutrient
