SonarA11y

Not enough for a Devpost submission centered on DigitalOcean. It explains the product well, but it underplays that Gradient is the core of the architecture.

Use this version instead:

About

SonarA11y is an accessibility remediation platform built around the DigitalOcean Gradient AI Platform. It scans websites and PDFs, detects WCAG and PDF/UA issues, and uses Gradient-powered AI workflows to turn those findings into clear, actionable fixes.

Inspiration

I’m a Drupal engineer, and I spend a lot of time working in the accessibility world. I’ve seen how often teams can identify accessibility issues but still struggle to turn audit results into practical remediation steps. That gap inspired me to build something that helps bridge scanning and fixing, especially for public-facing digital experiences.

How I Built It

I built SonarA11y as a Gradient-first system:

• A web scanning layer built with Playwright and Axe to inspect live pages and capture accessibility violations.
• A Python remediation pipeline that uses DigitalOcean Gradient as the inference layer for analyzing findings and generating fixes.
• A routing workflow that sends different issue types through the right remediation path, including PDF review and structured accessibility guidance.
• A frontend dashboard that presents issue-by-issue remediation steps, suggested values, and downloadable reports.

DigitalOcean Gradient is the core of the project, not just an add-on. It powers the AI reasoning layer that transforms raw scan output into remediation guidance teams can actually use.

What I Learned

I learned how important it is to design around structured AI output instead of just asking a model for generic advice. I also learned how much product value comes from combining accessibility scanning with a strong remediation layer. On the platform side, I learned how to build around DigitalOcean Gradient as a central AI service and shape the workflow so model output is useful, consistent, and demo-ready.

Challenges

The biggest challenges were:

• Making model output specific, readable, and actionable enough for real remediation work.
• Handling PDFs in a meaningful way, since PDF accessibility is much harder than HTML accessibility.
• Preserving enough context for good fixes without making the system too slow or expensive.
• Designing the workflow so DigitalOcean Gradient was deeply integrated into the actual product experience rather than just used as a generic LLM call.

Built With

• axe
• digitalocean
• fastapi
• gradientai
• langgraph
• playwright
• python