156 Million Pages. How ITHAKA Built PDF/UA Compliance on AWS with PDFix

JSTOR runs one of the largest digital libraries in the world – 20 million PDFs, 156 million pages, scholarly content going back to 1550. ADA Title II set the deadline. The scale set the challenge.

The pipeline started from an open source AWS solution originally built with the Adobe PDF Accessibility Auto-Tag API. ITHAKA’s team evaluated it and made a deliberate switch – replacing Adobe with PDFix and veraPDF for structural tagging and PDF/UA validation. The swap happened without rebuilding the pipeline. That modularity was the point.

Manual remediation at $1-4 per page meant a potential bill of up to $624 million for the full corpus. Bulk upfront processing wasn’t realistic for a nonprofit. So ITHAKA and AWS built on-demand: the pipeline remediates a PDF when a user actually requests it and caches it for every user after that.

PDFix handles structural tagging- headings, paragraphs, lists, tables – with veraPDF validating PDF/UA compliance and Amazon Bedrock generating alt text for images. The pipeline runs on AWS Lambda, Step Functions, and Fargate. A PDF comes in, gets processed page by page in parallel, reassembled, validated, and cached.

The results after moving to production:

• 98% accessibility check pass rate
• $0.026 per page – down from $1-4 manually
• 97%+ cost reduction
• Production-ready in approximately 2 months from a 2-day build workshop

The underlying solution is open source. Universities, libraries, government agencies, and nonprofits facing similar compliance deadlines can adapt it for their own environments.

Read the full case study on the AWS Public Sector Blog →