Decoding Document Details Similar to the Human Eye
The accessibilityInsight™ platform includes three solutions. AccessibilityCheck is used for robust accessibility compliance reporting and accessibilityFix is used to remediate PDFs for compliance. The third offering, accessibilityReview, allows companies to perform validation across all document types to comply with WCAG 2.1 standards, report on document accessibility levels, and review and perform interactive remediation as needed.
The accessibilityInsight™ platform derives structure and intelligence from documents using a two-pronged approach. The approach is modeled on how the human eye generates visual images of a document, which are then passed on to the brain to decode information. In accessibilityInsight™, an image processing module generates the visuals and works in tandem with the AI-based core engine, which extracts structure from the document using purpose-built machine learning (ML) models. The structure that the AI-based engine decodes is then used to transform the document into a format that is compliant with accessibility standards.
Scaling to Process Large Documents
codemantra has been extensively using Amazon Web Services (AWS) since 2015. Its initial ML models ran on
Amazon Elastic Compute Cloud (Amazon EC2) instances. However, engineers were manually provisioning Amazon EC2 instances, which was not only time-consuming but also costly for compute-heavy processing jobs. Documents that were several hundred pages long required over-provisioning of instances to ensure uptime during processing.
“We started looking for on-demand provisioning alternatives that would give us the scale and cost efficiency required to run our ML models,” says Murali Rangarajan, vice president of engineering at codemantra.
Cuts Document Processing Time by 95%
codemantra explored different approaches to run its ML models, including serverless functions and containers, before learning about the Batch Transform feature on Amazon SageMaker. With Batch Transform, codemantra’s engineers can concurrently manage thousands of document processing jobs every day. “AWS is phenomenal in their support, always taking initiative when we’re trying to adjust our technology as the business grows,” Rangarajan shares.
A 1,000-page document used to take 167 minutes with the threat of timing out during processing, but now the same job takes just 7 minutes. Timeouts are no longer an issue, and there is no limit to the number of pages that can be processed simultaneously. This has proven practical for the digitization of vast backlists of academic publications at pace. In July 2021, codemantra worked with the University of Westminster to make 12,000 archived pages accessible for people with disabilities in just 8 days.
Halving Costs to Run ML Models
codemantra also switched to Amazon SageMaker Multi-Model Endpoints to easily deploy several custom ML models on a single endpoint. This approach has improved endpoint utilization and reduced deployment overhead.
Since the transition, the company’s engineers no longer need to create unique endpoints for each model, and cloud resources auto-scale according to size and type of workload. By leveraging Batch Transform and Multi-Model Endpoint features on Amazon SageMaker, codemantra has decreased its processing cost per document by at least 50 percent.
In addition to ML services on AWS, codemantra uses
AWS Lambda to invoke batch transformation and processing, and
Amazon DynamoDB as a serverless database to maintain different workflow states. It also uses
Amazon Simple Storage Service (Amazon S3) as a data repository for documents and ML predictions. codemantra leverages advanced features within Amazon S3 such as cross-region replication to replicate data across
AWS Regions. It deploys intelligent tiering to optimize storage costs and object versioning to automatically track versions of digital assets.
Ensures Data Security and Compliance
Since launching accessibilityInsight™, codemantra has served many customers from the public sector, particularly universities and government agencies. The business emphasizes data security and compliance on both a product and a strategic level. codemantra is SOC 2 compliant and ISO certified for quality management.
The company leverages AWS to protect customer data and secure the accessibilityInsight™ platform against vulnerabilities and attacks. codemantra appreciates that the AWS platform and services are compliant with global regulatory standards and subscribe to security best practices. Rangarajan concludes, “Our pace of innovation is accelerated because we can iterate our models without worrying about infrastructure or security.”
Learn More
About codemantra
codemantra is an intelligent document processing company that leverages artificial intelligence to automate workflows and achieve digital accessibility compliance. It serves more than 300 public and private sector customers across the globe.
Benefits
- Cuts processing time for a 1,000-page document by 95%
- Halves processing cost per document
- Accelerates pace of innovation
- Complies with SOC 2 and several ISO standards
- Quickly iterates ML models
- Increases document throughput
AWS Services Used
AWS Lambda
AWS Lambda is a serverless, event-driven compute service that lets you run code for virtually any type of application or backend service without provisioning or managing servers.
Learn more »
Amazon SageMaker
Build, train, and deploy machine learning (ML) models for any use case with fully managed infrastructure, tools, and workflows.
Learn more »
Amazon Simple Storage Service
Amazon Simple Storage Service (Amazon S3) is an object storage service offering industry-leading scalability, data availability, security, and performance.
Learn more »
Amazon DynamoDB
Fast, flexible NoSQL database service for single-digit millisecond performance at any scale.
Learn more »
Get Started
Companies of all sizes across all industries are transforming their businesses every day using AWS. Contact our experts and start your own AWS Cloud journey today.
/Codemantra_logo%402x.450dc785095e2b565fbc571682f5b02abebd5f9d.png "codemantra")
