The modern technological landscape is characterized by the need for applications to interact with content available in various formats. This includes processing complex documents such as insurance claims and medical bills, analyzing user-generated media on mobile apps, and building a semantic index on digital assets that encompass documents, images, audio, and video files. However, extracting meaningful insights from unstructured, multimodal content is no simple task. It requires setting up processing pipelines for different data formats and undertaking multiple steps to retrieve the necessary information. This often involves deploying multiple models in production, which necessitates cost optimization, safeguards against errors, integration with target applications, and regular model updates.
To simplify this intricate process, Amazon introduced Amazon Bedrock Data Automation during the AWS re:Invent conference. This capability of Amazon Bedrock is designed to streamline the generation of valuable insights from unstructured, multimodal content, including documents, images, audio, and videos. Bedrock Data Automation reduces the development time and effort required to build intelligent document processing solutions, media analysis, and other data-centric automation solutions.
Amazon Bedrock Data Automation can be used as a standalone feature or as a parser for Amazon Bedrock Knowledge Bases. This allows users to index insights from multimodal content and provide more relevant responses for Retrieval-Augmented Generation (RAG), a technique that enhances information retrieval by incorporating machine learning models.
As of today, Bedrock Data Automation is generally available, with support for cross-region inference endpoints in additional AWS Regions. This feature allows the use of compute resources across different locations seamlessly. Based on feedback during the preview phase, improvements have been made to enhance accuracy and support logo recognition for images and videos.
How Amazon Bedrock Data Automation Works
The Bedrock Data Automation feature can be accessed through the Amazon Bedrock console. For those unfamiliar with the console, it is recommended to explore the visual demo, which showcases how to extract information from documents and videos. This demo provides a comprehensive understanding of how the capability works and the customization options available. In this post, we will delve into how Bedrock Data Automation can be implemented in applications, starting with a few steps in the console and followed by code samples.
The Data Automation section within the Amazon Bedrock console requires confirmation to enable cross-region support the first time it is accessed. From an API perspective, the InvokeDataAutomationAsync operation now requires an additional parameter (dataAutomationProfileArn) to specify the data automation profile to be used. The value for this parameter depends on the AWS Region and account ID.
The dataAutomationArn parameter has been renamed to dataAutomationProjectArn, reflecting that it contains the project’s Amazon Resource Name (ARN). When invoking Bedrock Data Automation, specifying a project or blueprint is necessary. If blueprints are provided, custom output will be generated. For standard default output, configure the DataAutomationProjectArn parameter to use arn:aws:bedrock:
InvokeDataAutomationAsync is an asynchronous operation. The input and output configurations are provided, and the result is written to an Amazon Simple Storage Service (Amazon S3) bucket, as specified in the output configuration. Notifications from Bedrock Data Automation can be received via Amazon EventBridge using the notificationConfiguration parameter.
Bedrock Data Automation offers two output configurations:
- Standard Output: Delivers predefined insights relevant to a data type, such as document semantics, video chapter summaries, and audio transcripts. This option allows users to set up desired insights with just a few steps.
- Custom Output: Allows users to specify extraction needs using blueprints for more tailored insights.
To see these capabilities in action, a project is created and the standard output settings are customized. For documents, plain text is chosen instead of markdown. These configuration steps can also be automated using the Bedrock Data Automation API.
For videos, a full audio transcript and a summary of the entire video are requested, along with chapter summaries. Blueprints can be configured by choosing the Custom output setup in the Data Automation section of the Amazon Bedrock console. Sample blueprints, like the US-Driver-License blueprint, can be explored for more examples and ideas.
Sample blueprints are not editable, but they can be duplicated and added to a project. This allows for fine-tuning the data extraction process by modifying the blueprint and adding custom fields, which can utilize generative AI to extract or compute data in the required format.
The process of uploading the image of a US driver’s license to an S3 bucket is followed by using a sample Python script to extract text information from the image using Bedrock Data Automation. The script utilizes the AWS SDK for Python (Boto3) to perform this task.
The script’s initial configuration includes the S3 bucket name for input and output, the input file location, the output path for results, the project ID for custom output from Bedrock Data Automation, and the blueprint fields to be displayed in the output.
Running the script with the input file name reveals the information extracted by Bedrock Data Automation, including the name and dates on the driver’s license. Similarly, the same script is executed on a video file to extract insights.
Key Information About Amazon Bedrock Data Automation
Amazon Bedrock Data Automation is now available via cross-region inference in two AWS Regions: US East (N. Virginia) and US West (Oregon). When utilized from these Regions, data can be processed using cross-region inference in four Regions: US East (Ohio, N. Virginia) and US West (N. California, Oregon). All these Regions are located within the US, ensuring data processing within the same geography. Plans are underway to add support for more Regions in Europe and Asia by 2025.
Bedrock Data Automation includes several security, governance, and manageability capabilities, such as support for AWS Key Management Service (AWS KMS) customer managed keys for granular encryption control, AWS PrivateLink for direct connection to Bedrock Data Automation APIs within a virtual private cloud (VPC), and tagging of Bedrock Data Automation resources and jobs for cost tracking and enforcing tag-based access policies in AWS Identity and Access Management (IAM).
While Python is used in this blog post, Bedrock Data Automation is compatible with any AWS SDKs. For instance, Java, .NET, or Rust can be used for backend document processing applications; JavaScript for web apps processing images, videos, or audio files; and Swift for native mobile apps processing user-provided content. Extracting insights from multimodal data has never been easier.
For further reading and code samples, consider exploring the following resources:
- Amazon Bedrock
- AWS SDK for Python (Boto3)
This news is a significant advancement in data automation, providing businesses with the tools needed to efficiently process and analyze complex multimodal data. It opens new avenues for developing intelligent applications that can derive actionable insights from diverse data formats, ultimately enhancing decision-making and operational efficiency.
For more Information, Refer to this article.
