ITHAKA provides researchers, students, and educators access to an extensive collection of scholarly content through platforms that include JSTOR. Its digital corpus encompasses approximately 20 million PDFs and 156 million pages, including materials dating back to 1550.
Remediating a collection of this size through conventional manual methods would present a significant financial and operational challenge. According to the AWS article, manual remediation of ITHAKA’s full collection could cost between $156 million and $624 million.
Rather than attempting to remediate its entire archive in advance, ITHAKA worked with AWS to create an on-demand pipeline that processes documents as users request accessible versions. Each remediated PDF is then stored and made available for subsequent requests, allowing ITHAKA’s accessible collection to grow organically based on actual user demand.
Within the pipeline, PDFix applies structural tagging—including headings, paragraphs, lists, and tables—to individual PDF pages. Amazon Bedrock generates alternative-text descriptions for images and charts, while AWS Step Functions orchestrates the processing workflow. The completed document is reassembled and evaluated against PDF/UA accessibility requirements using validation tools that include veraPDF.
Early Results Reported by AWS
- A 98% accessibility check pass rate
- An estimated processing cost of $0.026 per page
- A cost reduction of more than 97% compared with manual remediation
- Production deployment approximately two months after the initial AWS and ITHAKA workshop
|
“This implementation demonstrates what becomes possible when PDF accessibility is treated as a scalable technology and workflow challenge.” — David Herr, President, PDFix-US |
ITHAKA combined PDFix with AWS services, artificial intelligence, validation tools, and appropriate human oversight to create a practical solution for an extraordinarily large and diverse document collection. It is an excellent example of how organizations can move beyond an exclusively manual approach and begin addressing accessibility at scale.
The architecture also incorporates a human-review fallback for documents that do not satisfy the required accessibility checks through automated processing. This combination of automation, validation, and human intervention allows ITHAKA to direct specialized remediation resources to documents that require them while processing many other documents efficiently.
Because the pipeline is modular, ITHAKA was able to adopt PDFix and veraPDF in place of the structural-tagging components used in the original open-source solution without rebuilding the complete architecture. The same modular approach will allow ITHAKA to evaluate and integrate new technologies as PDF accessibility and generative AI capabilities continue to evolve.
The AWS article presents the ITHAKA implementation as a replicable model for universities, libraries, government agencies, nonprofits, and other organizations managing large collections of inaccessible or untagged PDFs. It also demonstrates how PDFix can serve as the structural-tagging engine within cloud-based, enterprise, and on-premises document workflows.
Read Also: Kim Bearden Named Recipient of 2026 Women in Supply Chain Forum™ Award
