Portfolio design · Information architecture · AI-assisted build
How I built this portfolio, and then fixed it
I built Version 1 around everything I wanted to say. It looked polished, and the content was real. But it failed the basic job of a portfolio: people could not quickly tell who I was, what I had done or why they should pick me. Eventually, I could not either.
The challenge
Version 1 contained the evidence but hid the argument across too many pages, paths and repeated explanations.
My role
I made the original structural mistake, designed and built both versions, asked people to test them and changed course when their confusion proved the structure was wrong.
What changed
Version 2 explains who I am, what I have done and how I work on one scannable homepage. The cases are optional depth.
The brief I wrote after V1 failed
I did not begin with these conclusions. I reached them after the first version was already public and people had tried to use it.
The second brief was brutally simple: at first glance, answer who I am, what I have actually done and why somebody should pick me. Everything else must either support that answer or get out of the way.
- 01Make my position clear at first glance
- 02Put the complete short version on one page
- 03Keep the detailed cases as optional evidence
- 04Preserve the full working archive for interviews
One brief, four visual directions
Before Version 1 failed structurally, I used Lovable to compare four visual directions against one content brief. Typography, colour, spacing, composition, texture and image treatment could change. The underlying portfolio could not.
That made the comparison useful. I was judging four visual directions, not four unrelated websites wearing different copy.
The words inside the previews are the original exploration copy. They are not a second version of the current portfolio.
I design products from the user experience inward.
Designer, product leader and technologist working across complex products, AI-assisted design and human-centred systems.
Selected work, 2026
Product design · 2026
Current product concept
A new end-to-end experience designed from user needs before technical architecture.
Read the caseDesign critique · 2026
When AI built the wrong product
A polished interface that exposed why AI execution cannot replace product and UX direction.
Read the caseReal product work
Complex enterprise UX
Simplifying technically demanding workflows under genuine enterprise constraints.
Read the caseSelected archive
Earlier work
UI, UX, graphics and implementation from a career spanning design, engineering and product.
Read the caseHow I think
- 01Experience before architecture
- 02Complexity belongs behind the interface
- 03AI accelerates execution, not judgment
- 04Immediate feedback is part of functionality
Selected
Quiet Editorial
Warm off-white, ink navy, editorial serif
The winner. It looked elegant, current and senior, and it left room for the work.
Editorial survived. The structure did not.
Editorial won because it looked elegant, current and senior without trying to perform seniority in capital letters.
Tactile looked like an ugly blog from 2020. Precise went into the stash for documentation and business presentations. Playful was lovely, but it shouted graphic designer rather than Design Director.
When Version 1 failed, I did not throw away the part that worked. I kept the visual system and rebuilt the information architecture.
V1 failed the first-glance test
The first version split the homepage, work, principles, biography, walkthrough and build story across separate pages. Some of the same ideas appeared in several places with slightly different labels. Each page made sense alone. The portfolio did not make sense as a whole.
One friend expected to understand who I was from the homepage and felt lost. Another found the useful principles only through a tiny link, then discovered a different “How I think” elsewhere and wondered how many versions existed. She also said the portfolio was much bigger than expected.
Then I looked at it as a visitor and realised I would not read the whole bloody thing either. The problem was not missing evidence. It was too much content, too many starting points and no clear argument at first glance.
What I changed in V2
I rebuilt the homepage as the complete short version: photograph, one-line position, immediate proof points, selected case studies, approach, selected experience, biography and contact. The top navigation now points directly to those sections.
I removed the public walkthrough, shortened the copy, stopped repeating explanations and made the case cards genuinely scannable with consistent image crops, aligned titles and brief descriptions. Detailed cases remain for people who want the decisions and the uncomfortable details.
I kept Version 1 privately as a useful archive for interviews. The public site no longer expects a recruiter or hiring manager to navigate it like its author.
What AI did, and what I did
- 01I wrote and edited the content.
The career evidence, opinions, mistakes and decisions are mine.
- 02Lovable explored visual directions.
It produced alternatives against a brief I defined. I chose Editorial and rejected the others.
- 03Coding AI implemented the decisions.
The production portfolio is SvelteKit inside WEBZY, not the Lovable prototype.
- 04People exposed the failure. I redesigned it.
AI made iteration fast. It did not notice that the information architecture was wrong. My friends and I did.