2026Record

Portfolio Frontend

The site you are reading. One server-rendered page — no client-side router, no state library — and every project on it comes from the C# API in the same repository.

TypeScript
Next.js 16
React 19
Tailwind v4
Vercel

Decisions

01Server rendering with a 60 second cache

Why: The project list changes a few times a year. Rendering it on the server keeps the page fast and indexable without shipping a data-fetching layer to the browser.
Cost: A newly published project can take up to a minute to appear, and nothing on the page can be personalised per visitor.

02Retry the project list for 54 seconds

Why: The API sleeps on a free tier and takes about a minute to wake. A single 8 second attempt failed during that window, and the page rendered as if there were no projects at all.
Cost: With the API genuinely down, the first render blocks for the better part of a minute before falling back to the empty state.

03No admin area

Why: An authenticated CRUD would be the largest part of this codebase, used a handful of times a year by one person.
Cost: Publishing a project means writing the document by hand in the Firebase console.

04The home is a showcase, the reasoning is one click away

Why: An earlier version put every trade-off on the home page. The first screen became long, and a visitor had to read three of them before learning which projects even existed.
Cost: The strongest argument on the site now costs a click. Each card carries its decision count to make that click worth taking.

05Container width and reading width are separate

Why: The layout pinned everything to 64rem centred, which on a large monitor is a narrow column with empty space either side. Two classes now own it: one for the page container, one capping prose at 65 characters.
Cost: Two global classes in a codebase that otherwise uses inline utilities only. In exchange, width stopped being decided in eight different files.