Gabriel Castro

QA Automation Engineer. Based in Brazil, open to bringing confidence to teams anywhere in the world.

Gabriel Castro

Software breaks. I build the test suites that bring consistency, so teams can ship big changes with confidence.

2,917 testsacross 3 production systems
637 spec files 715 suites 40–50 CI/CD pipelines

About

Five years in test automation, most of it owning end-to-end coverage for a SOC 2 fintech platform in the US debt-sales market: three production systems, nearly 3,000 automated tests, and the pipelines that ran them every day.

I came to QA from web design and development, so I read a UI the way a designer does and a codebase the way a developer does. AI is part of how I work every day, and I treat what it produces the way I treat any code: as something to test before trusting.

How I use AI

AI speeds up every part of my testing. The final call is always mine.

Modern test automation isn’t just about writing test code. It’s about deciding what deserves trust when machines build alongside us.

From acceptance criteria to test cases

I use Claude to turn acceptance criteria into scenarios ready to automate, including the edge cases that are easy to miss.

Review before trust

Every AI-generated test case gets checked against the real requirements, which is also how I find gaps in coverage.

Writing and running tests

Claude Code alongside Playwright, with Playwright MCP to drive the browser. I review and own every line it writes.

Failures and bug reports

AI reads the failure logs with me to get to the root cause faster, then helps draft clear bug reports.

Projects

Things I build for testing and for the people who test.

Frogeye Reporter, Today view: totals, pass rate, failures, flaky tests, new failures, retries and tests by hour
Today: the latest run at a glance.
Frogeye Reporter, History view: runs in history, best and worst run, average pass rate, trend and pass rate per run
History: how the suite evolves from run to run.
Frogeye Reporter, Defects view: product and test defects today, new failures and product versus test defects over time
Defects: product bugs kept apart from test problems.

Frogeye Reporter

A dashboard for Cypress and Playwright runs

A test reporting dashboard for Cypress and Playwright. It reads results straight from CI and keeps the history of every run, so after each one a team can see what needs attention, not only what passed and what failed.

Now
Results from each CI run: pass rate, new failures, flaky tests and retries, with product defects kept apart from test defects.
History
Every run is kept, so you can compare runs and follow how the suite evolves: pass rate, failures and duration over time, the best and worst runs, and the overall trend.
AI
Coming next: every failure explained in plain words, with the likely cause and where to look first. Works with Claude, OpenAI or Gemini.
Debug
Coming next: traces, videos and logs for every attempt, and quarantine for unstable tests without touching the code.
  • Cypress
  • Playwright
  • CI/CD
  • Run history
  • Test analytics
Coming to GitHub soon

Vagalume

A job board built to be tested

A fictional job board I built as a practice target for software testers. Candidates apply to jobs, companies move applicants across a kanban board, and a small course store has a cart, coupons and checkout. Some of it fails on purpose.

UI
Shadow DOM, an iframe, drag and drop, a cookie banner, a broken image, and a data-testid on nearly every element.
API
60 documented operations in OpenAPI 3 with Swagger UI, ready for contract testing in Postman or code.
Data
PostgreSQL 16 in Docker, so tests can check what the screen can’t show.
Failures
Checkout returns 502 about 15% of the time, coupons are rate-limited, and there are slow and flaky endpoints to test against.
Built
With AI assistance, as a practice target rather than production code.
Failure rate, on purpose
Admin action (500) 35%2.5–4 s response
Checkout (502) 15%
Intentional failure rate by flow
ItemValueNote
Admin action (500)35%2.5–4 s response
Checkout (502)15%
Error responses documented in the API
  • 400
  • 401
  • 403
  • 404
  • 409
  • 422
  • 429
  • 502

4xx client errors 5xx server errors

Coming soon

Vagalume is the target. Next come the test suites that run against it, each in its own public repository, so you can watch them catch the failures Vagalume throws on purpose. New versions of Vagalume will keep adding things to test.

  • Cypress suite for Vagalume

    UI and API tests written in Cypress.

    Nov 10
  • Playwright suite for Vagalume

    UI and API tests written in Playwright, with their own set of scenarios.

    Nov 30
  • REST Assured suite for Vagalume

    API tests in Java, driven by the OpenAPI contract.

    TBA

Big changes, nothing broken

Platform-wide changes the test suite carried the product through.

GraphQL to REST

The whole API layer was replaced. The E2E suite was the safety net that confirmed every flow still behaved the same from the user’s side.

In-house component library

Helped design the components, then kept the regression suite green while every screen moved over to them.

Authentication moved to Auth0

Rebuilt login handling across the suite so tests kept running against the real identity provider.

Dark mode

A visual change touching every page. Combined automated checks with design-QA reviews to catch contrast and layout regressions.

Experience

  1. 2021 – 2026

    QA Automation Engineer

    300 Softwares, placed with a US fintech platform

    Owned test automation for three production systems on a SOC 2 fintech platform, from test strategy to the daily CI pipelines, with AI built into the workflow.

  2. 2017 – 2021

    Programmer and Web Designer

    300 Softwares

    Built web applications and WordPress sites from Figma to code, with JavaScript on the front end, and did the functional testing that eventually became my job.

  3. 2014 – 2025

    Co-founder and Tech Lead

    Grupo Eleven

    Ran WordPress projects for clients and did UI quality audits.

  4. 2011 – 2021

    Photographer New website coming soon

    Independent; sports photojournalist for Metro and ESPN Brasil

    Where my eye for detail comes from. Over those years I was also invited to give several talks.

    Lead photographer for the book O Atleticano Vai ao Paraíso, by Fred Melo Paiva Buy

Education

  • In progress: In progress Postgraduate degree in Software Quality Engineering and Testing PUC Minas
  • Completed: 2021 Software Testing / QA certificate Última School
  • Completed: 2009 Creative Enterprise Ownership Fashion Institute of Technology (FIT), New York
  • Completed: 2005 – 2009 B.A. in Social Communication, Advertising PUC Minas

Tools I use

Automation
  • Cypress
  • Playwright
  • JavaScript (ES6+)
  • TypeScript
  • Page Object Model
API testing
  • REST API testing
  • Test data through the API
CI/CD and data
  • Git
  • Azure DevOps
  • Azure Pipelines
  • SQL
  • Auth0
Practice
  • E2E, regression and integration testing
  • Test parallelization
  • BDD (Gherkin)
  • Test data management
  • Agile / Scrum
AI-assisted testing
  • Claude
  • Claude Code
  • GitHub Copilot
  • Gemini
  • Playwright MCP
Performance
  • k6
Design
  • Figma
  • Adobe Creative Suite
  • UI/UX