· 7 min read
Building My Portfolio as if It Were a Product
My first portfolio copied someone else's design. This one has a design system, CI quality gates and privacy built in, all next to a full-time job. Here's why.
My first portfolio was a copy of someone else's design, built in two days. This one I designed myself and built the way I'd build a product: a design system, quality checks on every change, privacy done properly, and a list of things I chose not to build. I did it next to a full-time job.
The reason is simple. When someone asks for my resume, I send this link. So the site can't just describe what I can do; it has to show it. This post shows what that means in practice, for anyone wondering how far a personal site is worth taking.
Why treat a portfolio like a product?
Because it's the one thing I send instead of a resume, and it has to prove the work rather than list it. A resume says "builds accessible, fast web apps". A site can be one, and anyone can check.
It also has real users, and they want different things. Recruiters want the experience and the projects in under a minute. Engineers want to see the code, the components and the scores. Friends just want to look around. A product has to serve all of them without getting in anyone's way.
What was wrong with my first portfolio?
It wasn't mine. I built it in two days on Brittany Chiang's layout, a layout many developers copy: a sticky sidebar, a glow that follows the mouse, sections that light up as you scroll.
It looked good, and it said nothing about me. Then it sat untouched for eight months. A site that shows you can follow a layout is not much of a showcase, so I started again from the design up.
What does "product grade" mean for a personal site?
To me it means four things: a design system, automatic quality gates, privacy that works, and being findable by people and AI.
A design system, not a set of pages
Every piece of the site is a component, from a single tag up to a whole page frame. I use atomic design: atoms (a button, a badge) build molecules (a project card), which build organisms (the header, the project list) and templates (the page shell).
The building blocks come from shadcn/ui, restyled once through variants instead of at every use. A "nav" button or a "tag" badge is defined in one place, so the look stays consistent and a change happens once.
I designed the look in Claude Design. It proposed three directions; I picked "Broadsheet", an editorial style with a serif for titles, a mono font for labels and thin rules instead of boxes. I chose its ink-blue accent with future pages in mind. Every component has a story in the public Storybook, 69 in all, and the components are synced back to Claude Design, so the design and the code stay one system.
Quality gates on every change
Nothing reaches the live site unless it passes the checks. Each change runs lint, type checks, unit tests, a production build, a Storybook build and Lighthouse on mobile and desktop.
The Lighthouse budgets are strict on purpose:
| Score | Budget |
|---|---|
| Accessibility | 100 |
| SEO | 100 |
| Performance, mobile | 90 or more |
| Performance, desktop | 95 or more |
If a change drops accessibility to 99, it doesn't merge. The source code is public, so you can read every check yourself. The blog posts are checked too: a post with a bad title length or a broken date fails the tests before anyone sees it.
Privacy as a feature
The site only sets what it needs, and the privacy page lists every cookie and storage key. Google Analytics loads only after a visitor accepts. The traffic numbers I see by default come from Vercel's cookieless analytics, and there's a switch to opt out of those too.
I built the cookie banner myself instead of using a library, so I control exactly what it does. That paid off in an unexpected way. The banner first appeared after the page loaded, and on the archive it was the largest thing on screen, so Lighthouse counted it as the page's main content. Rendering it with the page instead cut about 2.5 seconds from that measurement, known as Largest Contentful Paint.
Findable by people and by AI
A portfolio nobody finds is a private one. The site describes me to search engines with structured data, tells AI search crawlers they're welcome, offers an llms.txt summary, and pings search engines when it changes.
I also check whether it works. I asked ChatGPT, Claude, Gemini and Perplexity the same ten questions, from "Who is Tugrul Subekci?" to "Online tughra maker in Orkhon runes", each in private mode so my own history couldn't help:
| Assistant | Names me | Links tugrulsubekci.com |
|---|---|---|
| Gemini | 5 of 10 | 3 of 10 |
| Perplexity | 4 of 9 | 3 of 9 |
| ChatGPT | 3 of 10 | 0 of 10 |
| Claude | 1 of 10 | 0 of 10 |
Perplexity's free limit stopped it at nine questions. They find me by name, mostly through LinkedIn and GitHub, but none of them found the tughra maker or my posts yet, and ChatGPT still linked my old domain. That's the baseline; I'll ask again as the site gets older.
What I left out on purpose
Saying no is part of building a product. Each of these is something portfolios usually have and mine doesn't:
| Left out | Why |
|---|---|
| A PDF resume | I want people to share the link, not a file that goes out of date in an inbox. |
| A contact form | My email and LinkedIn are already on the site. A form adds spam and nothing else. |
| A "now" page | It needs updating every month, and I won't keep up with that. |
| A page per project | Same reason: more content to keep current than it's worth today. |
| Paid tools | Everything runs on free tiers. My only costs are the domain and my AI subscription. |
Each one would have been easy to add. Keeping the site small is what keeps it accurate.
How AI assistants changed the cost
AI made it cheap to build things properly. Claude Design gave me the visual directions, Claude Code built most of the code with me, and Claude's research feature produced the background docs I worked from for the logo, the blog and the SEO plan. How I write the posts themselves is in Blog Posts as Code.
So I stopped asking "is this overkill for a portfolio?" None of it took long: Storybook, the CI pipeline, unit tests and the Lighthouse checks all went in within two days.
What the AI didn't do is decide. I chose the design direction, the turquoise for the logo, a seal instead of "Built by", and building the banner myself.
It was also sometimes wrong. When the project screenshots looked blurry, it told me the images were just too small. I didn't believe it and said the pixels were being changed. I was right: the image optimiser was turning the screenshots into 256-colour files. It also suggested /writing for the blog; I asked why not /blog, and it agreed that was clearer.
An assistant is fast. Knowing when it's wrong is still my job, and that's a skill worth showing too.
Has it paid off?
It's too early to say. The site is new, and AI search has only just started to find it. What I can point to is what it produced and what anyone can measure today:
| What | Today |
|---|---|
| Accessibility and SEO in Lighthouse | 100, enforced on every change |
| Components with stories | 69 |
| Largest Contentful Paint saved on the archive | about 2.5 s |
| AI answers that link the site | 6 of 39, across four assistants |
| Running cost beyond domain and AI subscription | none |
It also became a place to make things. The Tuğrul logo and the tughra maker both started here.
Frequently asked questions
Should a developer portfolio have a PDF resume?
It doesn't have to. If the site has your experience, projects and contact details, the link does the resume's job and stays current. Recruiters can still print the page if they need a file.
Is Storybook worth it for a personal site?
If the site is meant to show how you build, yes. It turns your components into something anyone can browse, and it makes a visual break easy to spot before it ships.
Takeaways
A portfolio is the first thing people judge your work by, so it should be held to the standard of your work. AI assistants make that standard affordable for one person; the judgement is still yours.
The site isn't finished, and that's the point of treating it as a product: it keeps improving, one checked change at a time. If you want to see what I build next, follow me on LinkedIn.