This project provides a scalable, serverless solution for automated document processing and information extraction on AWS, leveraging generative AI and OCR. It targets users needing to convert unstructured documents into structured data efficiently, offering benefits like automated workflows, enhanced data extraction accuracy, and integrated human validation.

How It Works

The solution employs a modular, serverless architecture built on AWS services like Lambda, Step Functions, SQS, and DynamoDB, deployed via nested CloudFormation stacks. It supports two primary processing modes: a default "Pipeline mode" orchestrating OCR, Bedrock classification and extraction, assessment, rule validation, and summarization; and a "BDA mode" utilizing Bedrock Data Automation for end-to-end processing. This serverless and modular approach ensures scalability, cost-efficiency, and flexibility for diverse document processing needs.

Quick Start & Requirements

Deployment is initiated by launching a CloudFormation stack via provided buttons in the AWS console for specific regions. Key prerequisites include an AWS account and prior access approval for specific Amazon Bedrock models (e.g., Amazon Nova, Titan Text Embeddings V2; Anthropic Claude 3.x, 4.x). Initial login to the web UI requires a temporary password provided via email post-deployment. Links to the Documentation Site and Deployment Guide are available for detailed instructions.

Highlighted Details

• Fully serverless architecture for high scalability and managed infrastructure.
• Modular, pluggable design supporting state-of-the-art models and AWS services.
• Command Line Interface (CLI) for batch processing, evaluation, and programmatic access.
• Support for few-shot examples to improve model accuracy.
• Integrated Human-in-the-Loop (HITL) system for validation workflows.
• AI-powered evaluation framework for assessing extraction accuracy.
• Web User Interface for monitoring and interaction.

Maintenance & Community

The project welcomes community contributions, with a detailed Contributing Guide available. It specifies linting tools (ruff for Python, ESLint for UI) indicating development standards. While specific community channels or core maintainer details are not highlighted, the presence of a contributing guide suggests ongoing project engagement.

Licensing & Compatibility

The solution is licensed under the MIT-0 license. This permissive license generally allows for broad usage, including commercial applications and integration into closed-source projects, with minimal restrictions.

Limitations & Caveats

Successful deployment and operation are contingent on obtaining access to specific Amazon Bedrock models, which may involve separate approval processes. The solution is tightly integrated with AWS services, requiring an AWS environment for deployment and use. Customization for complex production use cases may necessitate engagement with AWS Professional Services.