Back to portfolio
Accessibility·2026·Inclusive App Craft
Webscan — an AI-assisted accessibility testing platform
I designed and built Webscan, a platform that finds accessibility barriers across a website, helps teams fix them, and proves the fixes hold over time — pairing automated WCAG testing with human judgement and AI-assisted remediation.
The challenge
Automated accessibility checkers are fast but noisy, and they stop at detection — they don't help you fix anything or confirm it stayed fixed. I wanted a single workflow that runs from finding a barrier all the way to a verified fix, without losing the human judgement that real accessibility work needs.
How I approached it
- Rendered every page in a real browser and ran the axe-core rule set plus custom checks, so findings reflect what users actually meet.
- Added a human 'verdict' step and deterministic prioritisation so real barriers rise above false positives — with an optional per-organisation model that learns from those verdicts.
- Built AI remediation that drafts fixes from cited evidence only, requires human approval, verifies each fix in a sandbox, and opens a GitHub pull request rather than touching live code.
- Extended coverage beyond public pages: authenticated scans, goal-directed agentic exploration, scripted user journeys, and checklist-based manual WCAG audits.
- Captured real user sessions through a cross-browser extension, with sensitive input redacted in the browser before anything is sent.
- Closed the loop with scheduled monitoring, regression alerts, and branded HTML/PDF/DOCX reports.
Outcomes
- Live in production at webscan.inclusiveappcraft.com.
- One continuous workflow: detection → verdict → issue → AI-assisted fix → retest → monitoring.
- Safety-first by design — reads anywhere, restricts data-changing actions to non-production, and keeps every fix behind human approval.
- Backed by a comprehensive automated test suite (300+ tests).