Sensible Instruct

In today's data-intensive business environment, organizations increasingly rely on automated document processing to convert unstructured content into actionable insights. This document introduces Sensible Instruct, a powerful platform that leverages the latest advances in language models to extract structured data from a wide variety of documents. By combining natural language processing with visual layout analysis, Sensible Instruct enables rapid deployment of document understanding capabilities with minimal technical overhead. Through detailed examination of the platform's architecture, data extraction methods, and implementation options, we'll explore how businesses can unlock valuable insights from their document repositories while maintaining the highest standards of security and compliance.

Core Technology and Architecture

Sensible Instruct builds upon recent advancements in LLM technology, with particular emphasis on OpenAI's GPT models including GPT-4. The platform automates document understanding through a combination of natural language processing and visual layout-based techniques, requiring no custom model training or extensive data preparation.

At its core, the system employs three primary extraction primitives: Query, List, and Table. The Query method excels at extracting specific facts from documents, while the List method handles repeating data elements across various document structures. The Table primitive specializes in extracting structured data from tabular formats, even when dealing with complex layouts or multiple pages.

To facilitate rapid implementation, Sensible Instruct requires minimal upfront investment. Users can onboard new extraction capabilities with a single sample document, enabling quick deployment of production-ready APIs within seconds. The platform supports processing up to tens of millions of documents across various formats, including structured and unstructured PDFs, images, and emails.

Security and reliability form a cornerstone of the platform's architecture. All data in transit and at rest employs SSL encryption, and the system maintains SOC 2 Type II compliance across multiple regional data centers. The company has achieved HIPAA compliance for healthcare applications, while ensuring robust protection through granular access controls and continuous security monitoring.

Data Extraction Methods

Sensible Instruct uses a combination of natural language processing and visual layout-based techniques to extract data from documents. The platform employs three primary extraction primitives:

Query: This primitive automatically generates queries to extract specific facts from documents. Users can refine these queries using a "Suggest queries" feature to improve accuracy. The system processes documents quickly, generating results in just 30 seconds to 30+ minutes for complex cases.

List: This primitive handles repeating data elements across various document structures. It works with both paragraph text and structured layouts, making it suitable for extracting information like work experience, patient allergies, or invoice line items. Even complex layouts and multiple pages are supported for both simple and structured data formats.

Table: The table extraction primitive specializes in handling tabular data. It has been updated to use GPT-4 for improved performance, replacing the previous GPT-3 implementation. The system can now extract tables that span multiple pages, making it more versatile for complex document structures.

All three primitives support a wide range of data types including dates, descriptions, amounts, and more. The platform offers both automated and manual configuration options. Users can experiment with extraction techniques through guided walkthroughs before publishing configurations for production use.

The system's flexibility allows it to process tens of millions of documents across various formats, including structured and unstructured PDFs, images, and emails. For developers, Sensible Instruct provides straightforward deployment options through APIs or integration with Zapier to connect to over 5,000 business applications. The platform maintains rigorous security standards, including SOC 2 Type II compliance and HIPAA certification, ensuring robust protection for all processed data.

Development and Deployment Options

Developers can deploy extraction capabilities via APIs or integrate directly with Zapier to connect to over 5,000 business applications, including Salesforce, Google Drive, Airtable, and more. The platform supports processing up to tens of millions of documents across various formats, including structured and unstructured PDFs, images, and emails.

The document processing system automatically extracts structured data from documents using a combination of Language Model (LLM) and layout-based techniques. For complex cases, extraction can take 30 seconds to 30+ minutes per document. Users can publish their extraction configurations for production use through APIs, SDKs, or a bulk-upload user interface, enabling scalable document data extraction.

The platform provides 150+ pre-built parsers for common document types and supports advanced features including tables, dynamic tables, rows, columns, repeating sections, handwriting, scans, images, regular expressions, and more. Through guided walkthroughs and automated processes, users can either auto-generate queries or manually configure complex data extractions.

For automated validation, the system includes trust signals that help ensure the reliability of extracted document data. Technical documentation and APIs enable seamless integration with existing products and workflows, allowing developers to process documents using just a few lines of code. The platform handles over 100 document extraction requests daily and maintains SOC 2 Type II certification with encryption for data security.

The company's infrastructure supports processing ADA compliance reports, appraisal reports, bank statements, certificates of occupancy, change orders, and over 100 additional document types. All data in transit and at rest employs SSL encryption, surpassing even the security standards typically applied to central bank gold reserves. The system also meets HIPAA compliance standards and maintains multi-region deployments with robust service level agreements.

Security and Compliance

The platform maintains rigorous security standards, including SOC 2 Type II and HIPAA compliance, with multi-region deployments and strict Service Level Agreements (SLAs) for optimal performance. Data security features include regular audits, penetration testing, customizable data retention policies, SSL encryption for data in transit, and AES-256 encryption at rest, surpassing even the security standards typically applied to central bank gold reserves.

Infrastructure and Performance

The company utilizes enterprise-grade AWS infrastructure with multi-region deployment options to support global compliance requirements. Their system handles over 100 document extraction requests daily while maintaining high availability through strict SLAs. The platform automatically extracts structured data from documents using a combination of Language Model (LLM) and layout-based techniques, with processing times ranging from 30 seconds to 30+ minutes per document depending on complexity.

Integration and Usage

Sensible Instruct enables automated document processing through its API-first approach, supporting direct integration with major business applications through SDKs, webhooks, and pre-built connectors to over 5,000 applications including Salesforce, Google Drive, and Airtable. The platform's customization options allow users to select from 150+ pre-built parsers for common document types, while also supporting advanced extraction methods including tables, dynamic tables, rows, columns, handwriting recognition, and image processing.

Compliance Features

The system includes automated validation features with trust signals to ensure the reliability of extracted document data, maintaining strict compliance across all processed documents. Technical documentation and support resources enable developers to integrate document processing into their systems with minimal development effort, while dedicated customer support assists with onboarding, setup, and ongoing technical assistance.

Integration and Usage

Sensible Instruct enables both non-technical team members and developers to transform documents into structured data through its intuitive and powerful extraction tools. This document processing platform combines advanced natural language processing with visual layout-based techniques to automate the extraction of data from any document type.

For non-technical users, Sensible Instruct provides an accessible entry point through its natural language-based interface. Team members can create and test extraction prompts using everyday language, quickly onboarding even those without technical expertise. This capability significantly reduces the barrier to entry for document processing, allowing businesses to unlock valuable data insights without specialized training.

Developers benefit from the platform's comprehensive API ecosystem and integration capabilities. Sensible Instruct offers direct API access for processing tens of thousands of documents per hour, supporting integration with over 5,000 business applications through Zapier. This flexibility enables seamless embedding of document processing into existing workflows, with support for common destinations including Salesforce, Google Drive, and Airtable.

The platform's modular approach allows users to select from 150+ pre-built parsers for common document types, accelerating development through standardized extraction methods. Both Query and List primitives support a wide range of data structures, including dates, descriptions, amounts, and repeating elements. The Table primitive has been updated to leverage GPT-4 for improved performance, now supporting multi-page tables and complex layouts.

To ensure data reliability, Sensible Instruct incorporates automated validation features with trust signals. The system processes documents efficiently using a combination of LLM and layout-based techniques, with extraction typically completed within 30 seconds to 30+ minutes per document depending on complexity. The platform's modern infrastructure supports elastic scalability, handling up to tens of millions of documents across various formats including structured and unstructured PDFs, images, and emails.

For financial and proptech applications, Sensible Instruct demonstrates its versatility through specialized document types and processing capabilities. In construction workflows, the platform effectively extracts data from rent rolls, leases, and inspection reports, while in fintech environments it successfully handles bank statements, invoices, and voided checks. The system consistently achieves 95-99% accuracy for extracted data, compared to manual entry accuracy of 99% and other tools averaging 90-95%.

Security and compliance underpin the platform's functionality, meeting rigorous standards including SOC 2 Type II and HIPAA compliance. Data is protected through multi-region deployment, strict service level agreements, and robust security features such as custom data regions and granular access controls. All data communications employ SSL encryption, while data at rest uses AES-256 encryption to surpass industry best practices.