Building this site
with AI

I wanted a portfolio that was fully mine — not a template — and I wanted to build it the way I actually want to work: AI-first, owning the whole pipeline from idea to live site. I'm a designer, not an engineer. So the real project was learning to orchestrate AI to generate the build, then learning enough of the technical stack to integrate, ship, and maintain it myself.

The site you're reading is the result. Every design decision was mine. AI compressed the distance between an idea and seeing it live, but the direction, the integration, and the call to ship were the work.

My old site sat on a template and always felt borrowed. I wanted something built from scratch, in a voice and structure I controlled down to the line. Four things made that demanding.

The difference between a toy result and a shippable one is how you direct the model. This was closer to working with a very fast collaborator that needed clear direction and honest review, not a button that makes a website.

I orchestrated more than one tool at once — Claude Design as the visual builder, Claude as strategist, coder, and technical guide — moving output between them while holding a single coherent direction. I never took output at face value: I read what came back, caught what was off, and directed the fix. That evaluation loop is most of the actual work. A few deliberate prompting techniques did the heavy lifting.

Technique 03 — Context engineering. Chat sessions hit context limits, so I built a reusable continuity prompt that handed a fresh session everything it needed instantly:

Repo, structure, image URLs, palette, type, and current state — handed over in one block so no session started from zero. Treating context as something to design, not just prompts to write, is what kept the build coherent across dozens of sessions.

Generating a design is the easy part. Making it a fast, hosted, shareable site took hands-on integration and problem-solving — the kind of work that only shows up once you own the pipeline.

Once the homepage felt right, the real design-systems move was to stop treating pages as one-offs. I extracted the homepage into a reusable system — a defined palette, a type scale, and a repeatable section rhythm — and applied it across all six case study pages without altering a word of their content.

That's the same instinct I bring to any product: get the language and the structure coherent once, then let it scale. It's why the site reads as one considered thing rather than seven separate pages, and why I can add a new case study — like this one — in the same system without redesigning anything.