My portfolio, designed by me and built with AI
The site you're reading. I designed every screen and used prompt engineering to direct the AI that wrote the code, from the CMS to the visual identity. It's how I work day to day, just running in production here.
- Role
- Product Designer · Prompt Engineering
- Timeline
- 2026, ongoing
- Platform
- Web
- Team
- Solo

Overview
A lot of portfolios are a template with a different name on top. I wanted mine to show a practice I've been building: using AI models to turn design decisions into a real product, with a CMS, deploys and an actual content model. Writing code line by line was never my strength. Turning a problem into clear instructions, reviewing what the AI produces with the eye of someone who designs interfaces every day, and pushing until it's right is. That came from courses on the topic and daily AI use at work, and it keeps evolving. This site is also where I try out new tools.
Why build it with AI, not alone in the code
I could have shipped a template in an afternoon. Instead I wanted to see how far prompt engineering takes me when the result has to run in production, with a CMS, deploys and decisions I'd defend in a design review.
Designers spend their careers specifying work for someone else to build. Here I swapped that someone for an AI and took the director's seat: I designed every screen, wrote every prompt, reviewed the generated code and asked for changes until the behavior was right. That changes how you design. You stop proposing things that are expensive to build for no reason, and you start noticing the small interactions that only exist when someone (or a well-directed AI) bothers to write every detail.
I wasn't trying to prove I can code. I wanted to prove I can steer an AI, with precise prompts, to a product I'd sign my name to.
This didn't come out of nowhere. I've taken courses like Claude Design Fundamentals, Claude Cowork, Code: The Practical Guide and Generative AI for UI/UX Design, and I use AI daily at work to prototype and validate ideas before I even open Figma. The site takes that practice to a whole product, and I keep using it to try new tools.
The stack, and how I got there
The site runs on Next.js 15 (App Router) with React 19, Tailwind CSS, Framer Motion and TypeScript end to end. Content lives in Sanity, with the Studio embedded at /studio, so editing happens inside the app itself. Every push to main deploys to Vercel.
I didn't pick this stack because I knew every piece of it. I picked it because I knew, from research and prompting practice, what each tool solves, and then I guided the AI prompt by prompt until the implementation was right:
- Next.js and Vercel: server-rendered pages, a fast first load, a clean URL per case study and zero-config deploys. Shipping is a git push.
- Sanity: a real content model instead of Markdown files that would need a redeploy for every change. Text, images and project order are edited in a visual Studio.
- Framer Motion: restrained motion that does a job, like reveal on scroll and the sliding project panel.
- Tailwind: constrained tokens that keep spacing and color consistent without a separate design system repo.
Designing the content model
The hardest part was the schema, not the visuals. A portfolio is only as flexible as the shape of its data, so I described to the AI how I wanted projects modeled: documents with localized fields, a meta block (role, timeline, platform, team), tags and a list of content blocks.
Each block has a heading, a rich text body and an optional set of images with a chosen layout. Specifying that structure, and reviewing each attempt until it worked, is what lets a case study be expressive without custom code per project. One renderer handles every story.
Bilingual from the model up, no plugin
The whole site exists in English and Portuguese, and language has been part of the model from the start. Every localized value is an { en, pt } pair, resolved by a small language context with a toggle in the nav. Because the content was modeled that way, bilingual support came free at every layer: cards, case studies, even the CMS fields.
Rich text and a real image gallery
Early on, case study bodies were plain strings. That broke the first time a story needed a bulleted list or a pull quote. So I had everything moved to Portable Text, with a renderer that maps the CMS styles to the site's typography: bold, lists, subheadings, and quotes in a rounded callout box.
Images got the same care. Blocks support several layouts: the default gallery, with a main image and thumbnails, plus stacked, side-by-side grid and Pinterest-style masonry. A click opens any image full screen. Every image keeps its original aspect ratio, so nothing gets cropped or distorted, whatever file is uploaded.
A Markdown to CMS pipeline
Entering eight or nine case studies by typing into the CMS is slow and error-prone. So I asked for an import pipeline: I write each case as a pair of Markdown files (one per language), and a script reads the frontmatter and converts the body to the CMS's rich text.
I didn't write that script alone. I described the behavior I needed and reviewed each version until it worked the way I wanted. It's careful on purpose: it imports as drafts so nothing goes live unreviewed, skips projects that already exist so Studio edits are never overwritten, and refuses to touch protected documents. It's a small tool, but it lets me write in a comfortable editor instead of fighting a web form.
What this practice taught me
Guiding the AI through every step made me a better designer, not a programmer. I decide differently when I know what each choice costs, when I can tell a layout is cheap and a custom interaction is expensive, instead of guessing. That judgment comes from practice: the courses, daily AI use at work and the experiments I keep running here.
Designing the thing and guiding the AI that builds it turned out to be the same skill. Doing both closed a gap I didn't know I had.
What I still want to explore: new models and agents as they come out, per-case analytics to see which projects hold attention, a light and dark theme, and a guided "add a project" flow in the Studio so future updates depend even less on me.