Skip to main content
Kyle Johnson
← All work
LiveCase study 2 of 2

This portfolio

The site you're reading: a Sanity-driven portfolio that refuses to publish an unsourced number, tested for speed and accessibility on every push.

Role
Designer and developer
Stack
Next.js 16, TypeScript, Tailwind CSS v4, Sanity v6, Vercel, Playwright, Lighthouse CI
Timeline
Nov 2026 – present
Outcome
100 Lighthouse mobile performanceLighthouse CI, production build · Nov 2026

Visit the live siteView the repo

Proof · Lighthouse mobile · Tested Nov 2026

Performance
100Pass
Accessibility
100Pass
Best practices
100Pass
SEO
100Pass
The portfolio home page: the headline "Websites your team can run, and customers can find.", buttons for the work and the resume, and the first case study card below.

Problem

My resume says I build fast, accessible, CMS-driven websites that teams can update themselves. A resume can only claim that. I needed a site that shows it: real scores, a real editing workflow, and code a reviewer can open and read.

Constraints

  • WCAG 2.2 AA and a Lighthouse score of 95 or higher in every category, checked automatically.
  • No third-party scripts besides Vercel Analytics and Speed Insights.
  • Vercel's free Hobby plan, so most images resize on Sanity's CDN instead of Vercel's.
  • Fixed content types instead of a page builder, so every case study reads the same way.

Key decisions

Make the CMS refuse unsupported claims

Chose
Validation that blocks publishing when a number has no source or date, an image has no alt text, or a client hasn't given permission
Instead of
Trusting myself to remember the rules
Because
The site's whole pitch is evidence. A rule in the Studio still holds on a busy week.
Cost
Case studies can't go live half-finished. They wait for real numbers.

Drive the whole design from one token file

Chose
A tokens.json file that generates the CSS, with Tailwind limited to those tokens
Instead of
Tailwind's default palette and spacing, adjusted component by component
Because
Colors, type and both themes stay consistent, and an off-system color class produces no CSS at all.
Cost
A one-off style needs a new token first, and changing a token means regenerating the CSS.

Keep editing tools off visitor pages

Chose
Load Sanity's live preview and editing code only in draft mode, and refresh published pages with a webhook
Instead of
The starter setup, which shipped the editing code to every visitor
Because
Visitors never edit. That code was 583 KB of JavaScript on every page and held the home page at 83 on Lighthouse.
Cost
Visitors don't see a change the second it's published. The page refreshes on their next visit.

Outcome

The site scores 100 in every Lighthouse category on mobile and has no axe violations in light or dark mode. Every page, case study and resume entry is edited in Sanity, not in code.

Every push runs Playwright with axe and Lighthouse on each page in the sitemap, so a new case study is tested the moment it's published, and a failing check blocks the merge.

I built it with Claude Code as a pair programmer. I wrote the rules it works under, the design system and the content model, and reviewed every step before it shipped.

The principles, tokens, components and accessibility rules are public: read the design system.

  • 0 axe accessibility violations (WCAG 2.2 AA), light and dark mode, desktop and phone · axe-core in Playwright, run in CI · Nov 2026
  • 83 → 100 home page Lighthouse mobile performance, after moving editing code off visitor pages · Lighthouse CI, production build · Nov 2026
Side by side: the home page wireframe and the built home page. Layout, type and copy match; the built version drops the link underlines in the navigation.
The home page wireframe (left) and the built page (right).

What's next

Publish the other case studies as their numbers come in, and move the site to its own domain.