Case Study: R.A.A.D.

I conceptualized, designed, and built R.A.A.D. (Robust Automated Accessibility Diagnosticator): an end-to-end accessibility operations platform that combines automated axe audits, multi-profile Watchdog runs, AI-assisted remediation guidance, manual review, Figma design checks, and automatic code-fix branches behind a purpose-built operator experience.

🧩 Challenge

Accessibility work breaks down when the tools are fragmented. Teams bounce between scanners, spreadsheets, Figma comments, Jira tickets, and one-off scripts. Findings lose context. Remediation stalls. Design issues show up too late. Authenticated journeys get skipped because the audit cannot sign in.

I saw that gap in day-to-day accessibility leadership and decided to build the system I needed: one product that could own discovery, testing, review, design critique, and fix generation without handing the story off between five tools.

R.A.A.D. dashboard showing apps tested, URLs in scope, average score, automated audits, open issues, and a testing reach chart — Organization dashboard: reach, scores, open issues, and active runs

🗺️ Approach

I am the sole product owner, UX designer, accessibility engineer, and full-stack developer. I started from the operator workflow, not from a feature list. An auditor needs to add an app, define URLs and authenticated journeys, connect source repos, run tests, review ambiguous cases, critique designs before code ships, and turn confirmed issues into code changes.

That drove a few product decisions:

One application model. Each app owns inventory, auth, journeys, repos, and history so context never scatters.
Hybrid assurance. Automation finds volume. Manual review protects judgment. AI accelerates remediation language without pretending to replace auditors.
Shift-left into Figma. Measured design checks catch contrast and target size before engineering. AI drafts cover reading order, color-only meaning, and missing labels, clearly marked as drafts.
Fixes as reviewable diffs. Autofix only attempts criteria where the markup itself determines the correct change. Everything else stays in a manual queue. Branches are staged locally so a person still owns the commit.
Design system for the tool. I built RAAD’s own tokens and components so the operator UI stayed consistent, keyboard-friendly, and fast to extend.

R.A.A.D. application page for Carey Estes with navigation for settings, run test, manual test, design check, and reports — Application shell: run, manual, design check, reports, settings

R.A.A.D. application settings basics step with display name, slug, group, and inventory type — App settings wizard: inventory, journeys, repos, and access

💡 Solution

R.A.A.D. is not a thin wrapper around axe. It is a full product: information architecture, interaction design, a local design system, Flask APIs, Playwright automation, AI integration, authentication handling, report generation, and an autofix pipeline that turns finished audits into reviewable git branches.

Automated accessibility audits: axe-based scans across URL inventories, organization and app-level scores, and Watchdog multi-profile coverage (keyboard, zoom, forced colors, reduced motion, and more).
AI checking: optional suggestions grounded in failing markup and criterion, helping auditors move from “what failed” to “how to fix it.”
Manual testing: human-in-the-loop review with a headed browser bridge to flag defects or dismiss false positives.
Figma design reviews: measured contrast and target-size checks, AI draft findings for what only a picture can show, and node-anchored comments posted back into Figma.
Automatic code fixing: finished reports open staged fix branches via Cursor agents for mechanically resolvable criteria. Nothing is pushed without a human.
Bespoke authentication: operator login for the dashboard, plus per-app credentials and an auth pause/resume handshake for gated staging journeys.

R.A.A.D. Design Check form with fields for design name, auditor, Figma URL, and actions to fetch review or post comments to Figma — Design Check: fetch a Figma frame and draft review findings

R.A.A.D. Manual Test screen for reviewing accessibility issues with human judgment — Manual Test: human review integrated into the same product

R.A.A.D. accessibility report showing 22 unique issues, 203 occurrences, score of 26, executive summary, and disposition table — Audit report: WCAG 2.2 AA summary, unique issues, occurrences, and score

🧰 Roles in this work

I built R.A.A.D. end to end. These are the hats I wore across product, accessibility, design, and systems.

Full-stack product builder

From Flask APIs and Playwright runners to Docker packaging, CI sync helpers, and a custom front-end design system.

Accessibility engineer

Operationalizing WCAG 2.2 AA testing across automation, assisted review, multi-profile coverage, and remediation pipelines.

UX designer

Turning a complex audit lifecycle into a coherent operator experience with clear modes, wizards, and report storytelling.

Systems thinker

Connecting Figma, source repos, authentication, AI, and issue tracking into one accountable workflow instead of a pile of scripts.

🎯 Impact & Results

R.A.A.D. grew out of the accessibility program I lead at OneMain Financial. These are outcomes from that OMF work: faster remediation, broader coverage, and stronger organizational pull.

472%

Increase in issue remediation velocity year over year

26

Applications brought into the accessibility testing program

1,500+

Automated and manual accessibility tests across the portfolio

71%

Issue reduction following the 2025 Accessibility Awareness Campaign

Back to Home OMF Accessibility Leadership