← Back to work

Personal project · carolinetoscano.com · 2026

Redesigning my portfolio with AI

I rebuilt my portfolio with Claude and Cursor in 8 days. AI did the making. I kept the decisions, caught its mistakes and turned the process into a brief format I can bring to any team.

Product Design · AI-assisted workflow · Design System · Front-end · Copywriting

The carolinetoscano.com home page on a MacBook and an iPhone

TL;DR

My portfolio was pretty, but it didn't show how I think. I redesigned the case studies around my decisions, created one standard for every project, added Portuguese, and shipped it myself. The real work was directing the AI and reviewing everything it made.

Project
carolinetoscano.com
My role
Product design, content, front-end with AI
Timeline
8 days, Nov 2026
Tools
Figma · Claude · Cursor
Built for
Recruiters and design leaders hiring senior product designers

Context

A portfolio that showed screens, not thinking

I was looking for a new senior role and getting few replies. When I looked at my site the way a recruiter would, the problem was clear: the case studies talked about the companies and showed final screens, but it was hard to see what I did and why.

Research on what hiring managers look for pointed the same way. They skim, they want your role and the result in seconds, and they want to see your reasoning, written in your own voice.

The challenge

Rebuild everything, alone

No developer and no extra time. I needed to redesign the content and the UI, and ship it, while job hunting.

  1. 01

    Eight case studies with different structures and depth

  2. 02

    Copy that sounded generic and focused on the company

  3. 03

    A site only in English, with recruiters in Brazil too

  4. 04

    No engineering support to build the changes

How I worked

One loop between Figma, Claude
and Cursor

  1. 01 · Define

    Audit and plan

    I reviewed each case as a recruiter would and used Claude to compare my structure with strong case studies and hiring research.

    Me + Claude

  2. 02 · Design

    Prototype the standard

    I picked the Shell Box case as the pilot and explored the new layout, copy and data sections as quick prototypes before touching code.

    Me + Claude

  3. 03 · Build

    Ship with Cursor

    I wrote structured prompts for Cursor to apply the layout, keep my copy untouched and reuse the same components across pages.

    Me + Cursor

  4. 04 · Review

    Check and refine

    I reviewed every page, fixed what felt off and turned each problem into a rule for the next prompt.

    Me

Repeated for each case study, with the approved case becoming the reference for the next.

Before and after

The AI draft
vs. my version

AI wrote fast first drafts. They read well, but they sounded like any portfolio, and some of them described work that wasn't mine. These are real examples from the Shell Box case.

Shell Box · hero

AI draft

“Turning every refuel into progress.”

My version

“Clube Shell Box: new loyalty program”

Why

A slogan makes the reader decode the project. A plain title says what it is, and the subtitle right below says what I did.

Shell Box · process title

AI draft

“From business rules to a shippable MVP.”

My version

“How I worked”

Why

Every title followed the same “From X to Y” formula. Repeated on every section, it is the clearest sign of AI copy. Simple titles let the content do the work.

Shell Box · driver profiles

AI draft

“We mapped who refuels with Shell Box across two journeys…”

My version

“The business team had mapped who refuels with Shell Box… I used those profiles to test every screen decision.”

Why

The draft implied I created the profiles. I didn't. Saying where the input came from, and what I did with it, is more honest and shows my actual role.

Me and the AI

Who did what

Using AI well is mostly about knowing which decisions stay with you. I treated Claude and Cursor like fast collaborators, and kept the judgment calls.

Me

  • Defined what a senior case needs to show

  • Set the visual standard and the section patterns

  • Checked every fact, number and claim

  • Rejected anything that looked or sounded generic

Claude

  • Critiqued the old case studies and suggested structure

  • Drafted copy that I rewrote in my own voice

  • Built prototypes, diagrams and device mockups

  • Wrote first versions of the Cursor prompts

Cursor

  • Applied the new layout to every case study

  • Created reusable components for results, lists and images

  • Added the EN / PT switch without page reload

  • Built the interactive bookshelf on the About page

What I never delegated

  • What goes in a case

    Choosing what to show and what to cut is the core of a portfolio. AI adds; it rarely cuts.

  • Which data to show

    Only real data goes live. A missing number stays a visible placeholder until I have it.

  • My voice

    Recruiters read the case to hear how I think. A generic tone hides exactly that.

  • The final yes

    Nothing shipped without me reviewing the page on desktop and mobile.

Prompting as design

Writing prompts like a design brief

My first prompt to scale the Shell standard was one line. Each review showed me a gap, and each gap became a rule. The final prompt reads like a brief: a reference, what not to touch, how much freedom to take and what to deliver.

prompt → cursorV1
Use the Shell Box page as the standard and apply the same layout to my other projects: Origin, 77Sol, Gamers Club, Civi, Mobbi, Menina Shoes and iPharma.
The gap: nothing stops the copy from being rewritten, and nothing says what to do when a project has no research or no data. Every case would become a Shell clone.
prompt → cursorFinal
ReferenceThe Shell Box case is the visual standard.Reuse its components instead of creating new ones.
Adapt, don't copyNot every project is the same. Choose the sections that tell each story best.Where there's data, use the Shell results pattern.
Do not change the copyKeep every existing text exactly as it is.// Missing content: leave a [placeholder] and list it.
OutputThe page, the structure you chose and why, and the files you changed.
What changed: each rule came from a real problem in review. The output section makes the AI explain its choices, so I review decisions, not just pixels.

Where AI got it wrong

What I caught in review

AI is confident even when it's wrong. Most of my time went into review, and every mistake turned into a rule I still use.

  1. 01

    What happened

    It gave me credentials I don't have

    How I caught it

    The first About page draft listed a UX postgraduate degree and a design system course I never took.

    The rule

    Facts about me come only from my CV. AI never fills a biography.

  2. 02

    What happened

    It wrote about the company, not about me

    How I caught it

    Reading the Shell draft out loud, I couldn't find one sentence that said what I did.

    The rule

    First person, and a “What I did” block at the top of every case.

  3. 03

    What happened

    Every title used the same formula

    How I caught it

    “From X to Y” on section after section. Each one sounded fine; together they sounded like a machine.

    The rule

    Simple, descriptive titles. If it sounds like a slogan, cut it.

  4. 04

    What happened

    The image tool kept changing my face

    How I caught it

    I only asked for a new outfit and background for my profile photo. Each version came back as a slightly different person.

    The rule

    Open the prompt with what must not change, then list the only things that can.

  5. 05

    What happened

    Visuals that looked fine in the chat, not on the site

    How I caught it

    A cover showed my hero mid-typing animation; a phone mockup cut content and leaked past its corners.

    The rule

    Check every asset in place, on the black background, at real size.

Design decisions

What shaped the new site

  1. 01

    One case as the standard

    I perfected the Shell Box case first and used it as the template. It was faster and kept every page consistent.

  2. 02

    A structure, not a template

    Each case follows the same patterns but keeps its own order, because not every project has research, a flow or a design system.

  3. 03

    My role and results first

    A TL;DR and a project sheet at the top, so a recruiter skimming for a few seconds still gets what I did and what changed.

  4. 04

    Copy in my own voice

    I rewrote AI drafts in first person, used simple section titles and cut anything I couldn't explain in an interview.

  5. 05

    Two languages, one click

    A PT / EN switch that changes the page instantly, with translations written by hand instead of automatic ones.

Try it

Two languages,
no reload

The switch swaps every text on the page in place. Recruiters in Brazil read the case in Portuguese; everyone else gets English. Try it on this hero.

Hey, I'm Carol a
Product Designer

I use behavioral science to design products people actually want to come back to.

Same interaction as the live site.

Results

What changed

8

days from first audit to the live siteMy estimate without AI: [x weeks]

8

case studies moved to one standard

2

languages, switching without reload

Numbers in brackets are placeholders for real data.

What I'd bring to a team

A brief format for
working with AI

The site is the output. The process is what scales. I turned what worked into four rules I now use for any AI task, from copy to code.

  1. 01

    Give a reference

    Point to one approved example. “Make it like this” beats any description.

  2. 02

    Say what not to touch

    Copy, data, components that already work. Constraints protect quality more than instructions do.

  3. 03

    Set the freedom

    Say where to adapt and where to follow strictly, so you get judgment instead of a clone.

  4. 04

    Ask for the reasoning

    Request the choices and the changed files, so review is about decisions, not only pixels.

  1. →

    Prototype in code

    Test ideas as working pages in hours, so the team discusses something real instead of static frames.

  2. →

    Prompts as specs

    A brief with reference, constraints and output works as a handoff that both engineers and AI tools can follow.

  3. →

    Design systems AI can read

    Clear tokens, named components and written rules make AI output consistent with the product from the first try.

What I learned

What I'd take forward

  1. 01

    AI speeds up the making, but quality still comes from the brief and the review. The clearer my constraints, the less I had to fix.

  2. 02

    Designing one excellent example first and scaling it saved more time than planning everything upfront.

  3. 03

    Shipping code myself changed how I design for engineers: I now think about components and reuse from the first sketch.