Frederick, Md. – September 06, 2026 -- ITHAKA, the nonprofit behind JSTOR, has achieved a 98% accessibility check pass rate while cutting remediation costs by more than 97% through an on-demand PDF pipeline built on Amazon Web Services using PDFix structural tagging, according to a new AWS Public Sector Blog article highlighted by PDFix-US.
ITHAKA's 20 million PDFs would cost up to $624 million to remediate manually
ITHAKA's digital corpus spans approximately 20 million PDFs and 156 million pages, with materials dating back to 1550. The AWS article estimates manual remediation of the full collection would cost between $156 million and $624 million, a scale that made conventional, upfront remediation financially impractical.
ITHAKA processes documents only when users request accessible versions
Rather than remediating its entire archive in advance, ITHAKA built a pipeline with AWS that processes PDFs on demand. Each remediated file is stored for future requests, letting the accessible collection expand based on actual usage rather than a fixed, front-loaded project.
PDFix tags structure while Amazon Bedrock writes image descriptions
Within the workflow, PDFix applies structural tagging—headings, paragraphs, lists, and tables—to individual pages. Amazon Bedrock generates alternative-text descriptions for images and charts, and AWS Step Functions orchestrates the process. The reassembled document is validated against PDF/UA requirements using tools including veraPDF.
Processing costs $0.026 per page, down more than 97% from manual rates
AWS reported an estimated processing cost of $0.026 per page and a production deployment roughly two months after the initial AWS-ITHAKA workshop. A human-review fallback handles documents that fail automated accessibility checks, directing specialized remediation staff only to files that require manual intervention.
Modular architecture let ITHAKA swap in PDFix without rebuilding the pipeline
Because the system is modular, ITHAKA replaced the structural-tagging components of its original open-source solution with PDFix and veraPDF without reconstructing the full architecture. That same modularity is expected to let ITHAKA integrate future accessibility and generative AI tools as they mature.
"This implementation demonstrates what becomes possible when PDF accessibility is treated as a scalable technology and workflow challenge,