Bruno Amorim — Resume
Use the options below to download the file or view it directly in your browser.
Case Study
Gerador de Simulados
An AI practice-exam generator for medical residency, built for my girlfriend's exam prep — and my way of learning to build a full product with AI tools: APIs, a database, Google login and payments.
30-second read
Problem
The practice-exam software my girlfriend used for her residency prep was expensive and outdated.
Role
Everything, solo — product, UX/UI and the build, using ChatGPT and Codex.
What I built
AI-generated exams in an institution's style
Grading with an explanation for every question
Google login, exam history and a Premium plan
Outcome
Working and online, but never launched. Two students tried it and liked it.
Why
A study tool that shouldn't be this expensive
My girlfriend is preparing for the medical residency exams. Watching her study, I saw how much the practice-exam platforms cost and how dated they felt — and I wanted to learn to build real products with AI, not just prototypes.
So the project had two goals at once: a useful tool for one real user, and a hands-on way to learn to connect APIs, work with a database and add Google login.
Key decision
New questions, never copied ones
Instead of collecting past exams, the product generates new questions inspired by the style of each institution — Albert Einstein, USP, UNIFESP, Santa Casa, Sírio-Libanês. That avoids copying copyrighted exams, and the question bank never runs out.
The rules for the AI became product requirements:
- Never copy real questions or present them as official
- One correct answer, with plausible alternatives
- An explanation of 240–400 characters for every question
- A clear "Generated by AI. Not official." label on every exam
Process
Three weeks from prototype to a working product
The first versions, in mid-April, were quick standalone pages: a simulator shell and a ten-question test exam to see what a good practice exam should feel like. A week later came an MVP with simulated AI; by 10 May, the live version with real generation, login and payments.
Taking an exam
Generated in seconds, answered without distractions
The student picks an institution, a number of questions and a topic. While the AI writes the exam — around 12 seconds for five questions — a loading screen keeps the wait friendly. Then the generator collapses out of the way, and a fixed bar keeps the time, progress and the grade button always in reach.
Grading
Every answer explained, not just scored
Grading shows the score, then walks through every question: the right answer, the student's answer, and why. For exam prep, the explanation is where the learning happens.
History
A record of every exam
Logged-in users keep a history of their exams — institution, topic, date and score — and can reopen any of them to review each question and explanation.
Build
Learning the parts behind the interface
I built it with ChatGPT and Codex, which let me go past the screens into the parts a designer rarely touches:
- OpenAI API (gpt-4.1-mini) generating the questions, from a structured prompt that returns JSON
- Supabase for Google login, the exam history and each user's plan
- Vercel serverless functions for generation and payments
- A free plan — two exams a day, up to five questions — and a Premium plan through Mercado Pago
Outcome
Built, working, not launched
The product works end to end and is online, but I never marketed or launched it: no sign-ups and no payments. Two students used it and gave positive feedback.
Its real result is what I can now do: take a product from idea to a working, AI-powered app on my own — and design with a much better sense of what it takes to build.
Next project