Automated Intelligent Document Processing Architecture

Introduction

Organizations often face challenges in managing and extracting insights from vast amounts of unstructured data contained in emails, PDFs, images, and scanned documents. The variety of formats, document layouts, and text types complicates the extraction process for standard Optical Character Recognition (OCR) technologies.
To address these challenges, AWS offers connected, pre-trained artificial intelligence (AI) service APIs that enable organizations to derive meaningful insights from document-based data sources. This blog post presents a cost-effective, scalable automated intelligent document processing solution using Amazon Text

Document Management Challenges

Across various industries, customers encounter the following document management challenges:

• Inconsistent Extraction Accuracy: The accuracy of extraction processes varies significantly, particularly with handwritten text, images, and scanned documents.
• Inadequate Solutions: Existing scripting and rule-based solutions lack the flexibility to create domain-specific classifiers.
• Limited Feedback Integration: Traditional document management systems do not incorporate feedback from domain experts to improve the learning process.
• Data Privacy Concerns: Handling of Personally Identifiable Information (PII) is not robust or customizable, raising concerns about data privacy leakage.
• Manual Interventions: Many steps in the process require manual intervention, reducing overall efficiency.

Automated Intelligent Document Processing Solution

To address these challenges, we developed an automated intelligent document processing solution centered on a Natural Language Processing (NLP) engine, which includes:

• Amazon Textract
• Amazon Comprehend
• Amazon SageMaker
• Custom regular expression-based Python parser

The solution leverages other AWS services to create a cost-effective, scalable architecture for document processing.

Solution Overview

The automated intelligent document processing solution operates as follows:

Document Upload

Business users upload documents through a custom web application to a designated Amazon Simple Storage Service (Amazon S3) bucket.

Event-based Processing

An Amazon S3 event triggers an AWS Lambda function to start document pre-processing.

Pre-processing

The Lambda function evaluates the document payload, uses Amazon Simple Queue Service (Amazon SQS) for asynchronous processing, prepares document metadata, stores it in Amazon DynamoDB, and invokes the NLP engine for information extraction.

Text Extraction

The NLP engine uses Amazon Textract to extract text from various document types and optimizes API calls based on document metadata (e.g., form, tabular, or PDF).

Entity Parsing and Analysis

Amazon Comprehend processes the extracted text, performing entity parsing, sentiment analysis, and document classification. Custom classifiers within Amazon Comprehend enhance accuracy. PII data is masked using configurable rules.

Custom Parsing

A custom Python parser running in a Lambda function handles data from Microsoft Excel workbooks, invoked based on document metadata.

Machine Learning Integration

Output from Amazon Comprehend is fed into ML models deployed with Amazon SageMaker for additional use cases like recommendations, predictions, and personalization.

Post-processing

Upon job completion, another Lambda function updates the status in the Amazon SQS queue. The function parses the NLP engine’s output, augments data, validates key entities, assigns default values, and stores the results in Amazon DynamoDB and Amazon S3.

User Interface and Feedback

Users can review and compare extracted information with original documents via a custom UI, providing feedback to improve extraction and parsing accuracy. Amazon Cognito manages user authentication and authorization.

Customer Benefits

The automated intelligent document processing solution offers several benefits:

• Increased Efficiency: Automation reduces manual interventions, increasing overall document management efficiency by 50-60%.
• Reduced Administrative Work: Integrated workflows decrease the need for in-house team involvement in administrative tasks by up to 70%.
• Enhanced Visibility: Features like Document Classification and Obligation Extraction improve visibility into key contractual obligations.
• Feedback Mechanism: A UI-based feedback mechanism allows domain experts to validate extracted information and contribute to model training.
• Cost Optimization: The solution only calls the necessary Amazon Textract APIs based on document type and required information.

Industry Use Cases

This solution is applicable across various industries:

• Insurance: Accelerates claim processing and KYC-related processes by extracting key entities, mapping them to a defined taxonomy, and detecting anomalies with SageMaker models.
• Healthcare: Processes medical records and reports, extracting key medical entities and masking customer data.
• Banking: Automated check processing, extracting key entities like payer, payee, date, and amount.

Integrating Cloudairy Cloudchart for Effective Design and Collaboration

Cloudairy Cloudchart empowers architects to visually design and collaboratively refine 3-tier AWS architectures. Drag-and-drop pre-built shapes for AWS services (ELB, EC2, RDS) simplify the visual representation. Real-time collaboration ensures everyone is on the same page, while annotations capture design choices and security considerations. This centralized documentation streamlines the design process for a well-defined and secure 3-tier AWS architecture.

Conclusion

Manual document processing is resource-intensive, time-consuming, and costly. It requires significant resources, reducing business agility and employee morale. Intelligent document processing automates the classification, extraction, and analysis of data, expediting decision cycles, reallocating resources to high-value tasks, and reducing costs.

AWS AI services' pre-trained APIs facilitate quick document classification, extraction, and analysis. This blog discussed the foundational architecture to accelerate the implementation of specific document processing use cases.