about
The tools changed.
The bar didn’t.
I discovered code on a small farm in Brazil around 1998. I was 15, with a second-hand computer, an MS-DOS 6.0 book, and no internet. No friends nearby. No access to technical knowledge. Just a kid who wanted to be an inventor.
One afternoon, between building small electronic devices and a metal Wolverine claw (yes, really), I stumbled upon a file called GORILA.BAS (a QBasic demo). I opened it, saw colored lines of code, and felt something click. I had found a universe that was natural to me.
From satellites to startups
At 19, I landed a research fellowship at INPE, Brazil’s National Space Research Institute. Dream job for any nerd: infinite electronic components, fascinating projects, brilliant people. I wrote Linux kernel drivers for satellite communication hardware with 99.5% accuracy and ranked first among 40 students.
But science in Brazil doesn’t pay. My stipend was ~$50/month. So I moved to the private sector.
At Embraer, I spent 7 years learning to build production systems. I started in aircraft technical publications and quickly began automating processes with bash scripts on Sun Microsystems machines. Embraer gave me time to earn my Computer Engineering degree, study Applied Mathematics, and grow as an engineer.
After 7 years, I felt trapped. I wanted to fly further.
Building, failing, building again
I quit and co-founded a startup. In 2008, before WordPress was mainstream in Brazil, we built a food delivery ordering system from scratch: PHP, MySQL, classic startup spaghetti code.
That system processed over $6 million in transactions, reached every Brazilian state and expanded to Latin America. But we fell into the classic trap: we added features to turn it into an ERP and lost what made it work. Then a player called iFood entered the market.
We kept trying until 2018. Two or three more projects that didn’t work, burning investor money and our own. At one point, I was married, had a son, walking 12km daily because I couldn’t afford gas. Working remotely for Solvay, a Belgian multinational with €10B+ revenue, from 3AM to noon serving their European operation, at the office until 5PM, then teaching at a local university until 11PM. Sleeping 3 hours a night. For two years.
No CS degree covers that. That period did.
30,000 students and a digital business
In 2016, I created my first course, “Programar do Zero” (Code from Zero), teaching non-programmers how to code. Programming was still mainly an academic topic in Brazil at the time. It worked.
Over $1 million in revenue. 30,000+ students. 80,000 YouTube subscribers. During the pandemic, it boomed. At its peak, organic revenue was $10-14K/month, and with paid traffic, $60K/month.
Going global
In 2022, I joined TUI Group, the world’s largest tourism company (€23B revenue), as Tech Lead for the global checkout system. Working remotely from Brazil, I directed a 12-member team across 4 European markets, building a multi-currency transaction system processing 5M+ annual bookings with 25 microservices and 99.9% uptime.
After Solvay and TUI, I had proven I could deliver at enterprise scale across time zones and cultures. Fully remote, from Brazil to Europe.
Then AI changed everything
In 2024, I realized I couldn’t grow further inside my “digital business working 4 hours a week.” The rise of AI coding agents changed what was possible.
But the bar didn’t move: production, or it doesn’t count. So I brought 25 years of engineering discipline to the agents. I write the spec before the code, let agents execute against the contract, then review and ship. I trained 400+ professionals in the method.
Crypto Fintech: 13 apps in 70 days
In late 2025, I built a complete crypto fintech. Solo. In 70 days. Using Claude Code guided by specifications.
The platform includes a payment gateway with PIX integration, Brazil’s instant payment system (3B+ monthly transactions, comparable to India’s UPI), connected to 4 providers via factory pattern with HMAC-SHA256 webhook validation and mTLS client certificates.
It also features an OTC exchange engine with VWAP pricing (24h rolling window), cross-rates from 5 aggregated sources with median and outlier filtering, and asymmetric bid/ask spreads. The exchange settler handles on-chain deposit detection (10s polling), 2-block confirmation, tolerance-based auto-refund, and crash recovery.
The full stack: 13 applications, 3 Fastify APIs, 3 PostgreSQL databases, full OAuth 2.1/OIDC (5 roles, 19 scopes, 18 granular permissions, 2FA), Kubernetes with auto-scaling, Terraform IaC, and Prometheus/Grafana monitoring.
Real money flowing through the system. Production-grade fintech with compliance, not a prototype.
The AI did not build it. The specification did.
What I’m building
I’m building educational content: an ebook, a course, and a YouTube series on building production systems with AI: spec-driven development, harness engineering, and the architecture that keeps them from breaking. AI changed how we write code, not how we think about it.