That Time When I Built an AI-Powered Audit Process to Boost Versium.com’s Accessibility and Performance.

Project Context

Problem

Versium needed a solution to ensure accessibility compliance, SEO optimization, and site performance at scale. The manual process was time-intensive and error-prone.

Solution

Used AI to write a Python script that automatically runs PageSpeed checks, records baseline scores, and logs results in a spreadsheet for tracking.

Audience

Users who benefit from accessible and high-performing web experiences.

Project Type

Full-time Professional Experience

Constraints

AI Model Limitations

Time Frame

3 weeks

My Role

User Experience Researcher
User Interface Designer
Developer

Team

Myself

Project Breakdown

01

Problem Context

Versium.com needed to improve accessibility and performance for a better user experience and to pass Core Web Vitals. Early manual checks suggested color-contrast issues, slow load times, and inconsistent SEO practices. Relying on one-off tests made it difficult to track changes effectively, highlighting the need for a more systematic approach.

Figure 1: Shows the Google PageSpeed Insights interface prompting an audit for Versium.com.

02

Research Process

To capture both accessibility and performance metrics, I chose two core tools:

  • Pa11y: Flags accessibility issues like missing ARIA labels, improperly nested headings, and color contrast violations.
  • Google Lighthouse: Generates scores in Performance, Accessibility, Best Practices, and SEO, plus highlights problems affecting Core Web Vitals.

With Claude AI and Replit, I wrote a Python script that orchestrates both Pa11y and Lighthouse audits simultaneously. This setup produced consistent reports for each of Versium.com’s pages, avoiding the guesswork of manual testing.

Synthesis

Early reports revealed a cluster of issues cutting across both performance and accessibility:

  • Color & Layout: Some UI elements had low contrast ratios, complicating readability.
  • Structure: Heading tags were missing or out of sequence, and certain buttons lacked clear ARIA attributes for screen readers.
  • Load & Interaction: Above-the-fold content took too long to render, causing a fail in Core Web Vitals.

By grouping and prioritizing these findings, I concluded that many accessibility improvements would also boost speed. For instance, compressing images not only trimmed page-load times but also benefited users on low-bandwidth connections or with visual impairments.

Figure 2: A Lighthouse report from Sep 11, 2024, showing a failing Core Web Vitals assessment along with mid-range scores in Performance, Accessibility, Best Practices, and SEO.

03

Design

Based on these insights, I implemented multiple targeted changes:

  • Accessibility Tools: Introduced a user-facing widget (see Figure 3) to allow text resizing, grayscale mode, and link underlining.
  • Contrast & Structure: Revised color palettes to meet WCAG thresholds; reorganized headings so screen-reader users (and search engines) could parse content more easily.
  • Performance Optimizations: Compressed images, deferred non-critical scripts, and minimized CSS/JS to reduce page weight, speed up rendering, and address Core Web Vitals concerns.

Each iteration was validated by running a fresh round of automated audits, letting me see exactly which fixes were most effective.

Figure 3: Shows the newly integrated “Accessibility Tools” widget, enabling users to adjust text size, toggle grayscale, and underline links as needed.

04

Deliverable

At the end of the project, I provided:

  1. Automated Audit Pipeline
    I delivered a Python-based system combining Pa11y and Lighthouse reports, allowing the team to regularly check Versium.com’s accessibility and performance status.
  2. Updated WordPress site
    Compressed and lazy-loaded images, reducing page weight and improving time-to-first-interaction.
  3. Accessibility Features
    The newly implemented widget and revised design elements established a more inclusive experience, particularly for users with visual impairments.
Outcome

By December 14, 2024, Versium.com passed its Core Web Vitals assessment and saw notable score improvements:

  • Accessibility gained roughly 5%.
  • Best Practices jumped by about 40%.
  • SEO remained high.
  • Performance’s raw metric fluctuated slightly, yet real-world load times improved overall.

Ultimately, Versium.com became a faster, more accessible platform, while the automated audit process provided a scalable solution for ongoing monitoring and enhancements.

Figure 4:The final Lighthouse report confirming a “Passed” Core Web Vitals assessment and more balanced scores across all categories.