Equalify PDF to Semantic Markup Converter

The Challenge

Academic PDFs are inaccessible by default. They’re closed-source, visual-first, static formats where accessibility is bolted on as an afterthought within Adobe’s walled garden.

Manual remediation costs approximately $100 per page and takes weeks. UIC’s Digital Accessibility Services needed a scalable solution for hundreds of documents destined for Canvas LMS.

The Approach

Instead of trying to fix broken PDFs, we extract content and semantics into MDX — a format that’s human-readable, AI-readable, and program-readable.

The pipeline:

PDF Document
    ↓
[Extraction Layer - Docling/OCR]
    ↓
Raw text + layout information
    ↓
[Semantic Translation - LLM Agents]
    ↓
Structured MDX with semantic components
    ↓
[User Agent Rendering - Canvas/Web]
    ↓
Adaptive presentation based on user needs

Multi-agent architecture:

• Heading structure agent (infers hierarchy from visual layout)
• Data table agent (extracts semantic table structure)
• Chart agent (extracts data + generates descriptions)
• Image agent (context-aware alt text generation)

Each agent is narrow, testable, and auditable — a trusted procedure.

The Innovation

We don’t just convert PDFs. We liberate content into adaptive, user-agent-negotiated experiences.

The “last mile delivery” insight:

PDFs own the entire rendering stack — hostile to accessibility. Semantic markup lets user agents handle the last mile. Same content, different presentations based on user needs:

• Tables become sortable, interactive, sonified, or spoken
• Charts get extracted data + alt text + long descriptions + sonification
• Content becomes truly multi-modal and personalized

Skill-ready design:

The pipeline is architected to become a packageable skill — a trusted procedure any organization can deploy. Each agent in the chain is a governed, testable procedure.

Results

• Cost reduction target: 95-99% vs manual remediation
• Output: Permanent, citable, semantically rich accessible content
• Integration: Canvas LMS native
• Beta release: March 2026

Speaking

Presenting this work at CSUN Assistive Technology Conference 2026: “Escaping to Semantic Freedom”