Web Design

Dekel

A site for an independent fashion designer in Florence, designed and coded in a week through an end-to-end AI execution stack.
Solo design + build · 2026 · Client project · Live
Visit the live site ->Back to work

The brief

Klil Dekel is a fashion designer, a Polimoda graduate based in Italy, and he didn't have a site.

He sent five references. One of them was buildable at the scope he had, and I said so in the proposal rather than after the build.

The first version, and why it never got made

A year earlier I had already designed this project, and given it away.

It was called Klil's Collections: a single-page spatial experience built around full-screen runway video, with non-linear navigation and a custom-orders route reachable from anywhere. I did it unpaid, for my own portfolio as much as for his, and it was supposed to get built.

It didn't. The reason is the part worth writing down.

It never got built because the brand it was designed for stopped being that brand. He changed direction, and the design was left describing something that no longer existed.

I don't think the 2025 design was wrong. It was scoped for a brand that then became a different brand. What made the 2026 version happen wasn't a better design - it was a real budget, a fixed scope, and a week to do it in.

2025 · Prototype · Never built

What I built instead

An archive. Three influences, held against each other:

- Structure from Busygina - an asymmetric grid that never resolves into symmetry, so the eye keeps moving.
- Restraint from Dior - monochrome, generous space, nothing decorative.
- Typographic tone from Heliot Emil - the one thing worth taking from a store I couldn't build.
Four pages: home, about, archive, contact. Fourteen images. No CMS, no catalog, no checkout. Built in Astro, hosted on Cloudflare Pages.

From spec to build

Most of the work happened before the build: the asset pipeline, the type scale, the grid rules, the image budget, the accessibility requirements. All of it was specified in a living brief and held in plan mode until the plan stopped changing. The build was short because the deciding was already done.

The specification is the part that's mine. Execution is fast now, and it is also confidently wrong on a regular basis, so the work moved to verification.

The clearest case was the hero video, and a score that said everything was fine.

The score that wasn't a score

Lighthouse gave the site 100 for accessibility. Automated contrast checks skip any element sitting on a video - they can't sample a background that changes every frame, so they don't sample it at all. A 100 there means not tested.

The number moves

object-fit: cover scales the clip until it fills the screen, then crops from the centre. A desktop viewport keeps 91% of its width; a phone keeps 26%. The two are not looking at the same picture, so there is no single number to check.

Where it stands

It's live at dekel-studios.com, and it's his. Four pages, fourteen images, four requests, about 220ms to first paint.
What's still open:
He can't be found by his own name.

The <title> is "Dekel", and "Klil Dekel" appears nowhere visible on the site - only inside a meta description. The Person structured data we specified never shipped. For a site whose entire job is to be a business card, this is the one that costs him something real, and it's the next thing I'm fixing.

What I'd do differently: I'd have specified the SEO layer as a build requirement instead of a follow-up. It's the difference between a site that works and a site that gets found.

Let’s build something great together.

I'm currently exploring new opportunities in Product Design and UX Research.

My LinkedIn profile