Every decision in this site, and why

Your own portfolio is the hardest client you’ll ever have: you know all the arguments, you have all the context, and nobody can tell you no. To keep myself honest, here’s the self-review — every meaningful decision on this site, what it costs, and whether I’d make it again.

Why it’s built like this

Data-driven content, no CMS. Projects and articles live as typed TypeScript modules. The win: content is refactored with the same tooling and tests as code, and the build fails if a case study is missing its outcome. The cost: writing in a template string is less pleasant than a CMS editor. Verdict for a portfolio’s scale: correct. At product scale, wrong — use a real CMS.

Next.js App Router, static output. Every page here is static or SSG. For a portfolio, static is a performance and resilience gift: it’s fast everywhere and can’t have a runtime outage. A contact form via mailto means there’s literally no backend to operate.

Tailwind with strict tokens. Every colour goes through named tokens — ink, paper, coral, butter, and the dark-night family — audited for contrast. When a design tweak came in from the browser, going through the token rather than a hex keeps the contrast guarantee intact. This is a direct habit from the design-system work at the council: if the system makes the right thing the easy thing, you stop auditing and start designing.

Serif display, characterful accents. The display type carries the personality, with small hand-drawn accents — stars, squiggles — that are aria-hidden and never interfere with the reading experience. The tension here: "characterful" and "professional" pull in opposite directions. The rule I landed on: decoration only in whitespace, never in the reading path, and the colour palette stays consistent across every page.

The accessibility decisions, explicitly

• Skip link is the first focusable element — in the layout, not added later.
• Every interactive element is a real link or button.
• One H1 per page — even when the visual design wanted two headlines.
• The dark-mode toggle stores a theme and applies it before first paint — no flash, no prefers-color-scheme guessing on a site that has an explicit brand mode.
• The reduced-motion media query collapses all transitions to instant — the Reveal component checks it too.
• Decorative SVGs are aria-hidden. Screen reader users get content, not confetti.

These are all small. Their cost in implementation was hours. Their value is that I never have to think about them again — the system defaults to accessible.

What I’d change

• The contact form’s mailto flow. It’s zero-backend, but a mailto handoff is a conversation-killer for anyone without a configured mail client. Mitigated with a copy-to-clipboard fallback, but a real endpoint (a Cloudflare Worker — I already run them elsewhere) is the honest fix.
• The markdown renderer is hand-rolled. The blog renders article bodies with a custom parser. It was a fun exercise, but a battle-tested library would be less code to maintain and would handle more edge cases. The hand-rolled version supports exactly what my four posts use — anything richer and I’m back in parser land.
• Some headings drift toward generic portfolio-speak. Every time I catch a phrase that could appear on any portfolio — anything like "drive measurable results" — I rewrite it toward the specific: what I actually made, who used it, what happened. Specificity is the only moat.

The review that matters

The self-review was a useful exercise precisely because it exposed where I’d cut corners (the mailto, the parser) and where the discipline paid for itself (the token system, the a11y defaults). If you build for a living and haven’t written down why your own site is the way it is, try it. The decisions you can’t defend are the ones that need attention.