I had a lot of work, but the collection was not doing enough to explain why each project mattered. A visitor had to commit to a full case study before knowing whether it was relevant to them.

The problem was not a lack of detail. It was pacing. Every project opened with context, then process, then the work. The result was accurate, but it asked for patience before it had earned attention.

I wanted the portfolio to feel more like a conversation: give someone the useful answer early, show the evidence, and leave room for the messy thinking if they want to keep going.

I started with an information and layout brief, not a moodboard. It gave the redesign a job: lead with the result, alternate payoff with explanation, keep the writing human, and make a shorter version of every story possible.

That brief became the design system's first constraint. The visual language could be quiet and expressive, but it could not hide the work or turn every project into the same template.

I did not ask AI to invent the portfolio’s point of view. I brought references, described how the system should feel in motion, defined the reading experience and orchestrated the design-system ideation. Then I created the screens from scratch and used AI to challenge them, explore alternatives and suggest features or components I might push further.

The useful pattern was directed collaboration: I made a design decision, AI helped me expand or implement it, and I reviewed the result against the references and the actual project story. AI increased my range and speed. It did not replace taste, authorship or the final call.

Once I was satisfied with the story and screens, I used AI to help translate them into a small editorial system instead of twelve one-off pages. I reviewed the implementation as it took shape: Astro keeps the output static and fast; the shared shell handles navigation, themes, reading depth and lightboxes; each project keeps its own voice and evidence.

That split was my design decision. AI made the repetitive implementation faster, while I kept the project-specific parts open and reviewed the shared behavior whenever one change affected the rest of the site.

I orchestrated the visual foundation through references, interaction feeling and the needs of the project stories. In Figma I established the cover, logo, light and dark intent, and early layout decisions. I then designed the screens from scratch and used AI to help extend the approved patterns into code-level tokens, shared components, responsive rules and templates in Astro.

The system lives across design and implementation, but its point of view came from my review of every page and flow. Its job is not to make the portfolio consistent at all costs; it gives the work a shared grammar while each project keeps its own rhythm, colour and personality.

There is no meaningful conversion number yet, so I did not invent one. I validated what I could actually observe: can the page be understood, can the story be navigated, and does the build behave at the edges?

AI helped generate checklists, run routes and report failures, but its “passed” result was never the final word. I tested the site myself because automated QA could miss behavior that only appeared after a particular sequence of scrolling, section navigation and Escape-key use.

AI could propose a polished structure before it understood the project, make every section sound equally important, or produce more copy and components than a reader needed. It could also report that QA passed when its test path had not reproduced the sequence where the experience failed.

My work was not simply accepting or rejecting outputs. I supplied the references, made the screens, reviewed every project and flow, and tested the behavior myself. The most important design act remained deciding what was supported, what deserved a larger stage and what should not ship.

The redesign changed how I think about using AI in design work. The gain was not that a machine made a portfolio for me. The gain was that I could direct more iterations, test ideas sooner and move between writing, structure, code and critique without losing the thread.

Speed only mattered because I remained the deciding factor. I brought the point of view, designed the screens, reviewed the evidence and corrected the experience when the tools were confidently wrong.