My name is Marcelo. I'm a Computer Engineer from ITA¹ and a former Google engineer (Search, Data Pipelines / Dataflow, Chromium) with 15+ years of experience. I quit Big Tech to explore entrepreneurship and am now building AI-native applications and tools.

I use AI agents (Cursor, OpenCode, Roo Code, etc.) as force multipliers to ship high-quality code rapidly, while retaining control over system design and architecture.

I am actively building with AI Agents and studying their impact on Developer Experience (DX).

A high-performance keyboard remapper written in Rust.

• Why it matters: I built this functional Rust desktop tray application in just a couple of days with zero prior Rust experience. The goal was to test if my product and system design ideas would port well to a strictly typed language using a spec-driven AI workflow. Built using Kiro (kiro.dev).

An infinite mind-map generator powered by OpenAI's OSS models.

• Why it matters: This started as my first attempt at hosting LLMs locally. It proved unfeasible, as my hardware couldn't handle even the 20b model efficiently. This constraint pushed me to discover OpenRouter. Today, I use OpenRouter daily to power not just apps like this, but many parts of my toolchain (Zed, OpenCode, Roo Code, Continue).

My conceptual framework for AI-assisted development, designed to optimize the "Engineer in the Loop" workflow.

• Why it matters: It allowed me to develop a 3-Phase Workflow that matches model capability to task complexity, leveraging Frontier Models, Top Models and Cheap/Weak Models.

A VS Code extension bundling quality-of-life improvements.

• Why it matters: My first hands-on experience using the VS Code Extension API to optimize my own workflow.

• devmagic: A devcontainer setup that "just works," solving the "works on my machine" problem.
• Adapta AI Challenge: Top 3 Finalist in Brazil's largest-ever hackathon. Built an OOH analytics tool in ~24h.
• Viralei: Hackathon Project. Turns legislative bills into engaging videos. Built the MVP in 4 days.
• Fair Pricing Works: Open Source initiative advocating for Purchasing Power Parity (PPP).

I wrote my first line of code in 2010 (in C). Today, I choose tools that mix personal preference, velocity, and leverage.

• Core: TypeScript, React, Next.js, Node.js, Tailwind CSS.
• Systems & AI: Rust, Python (Automation/Pipelines), C++ (Google background).
• AI Workflow: Zed, Cursor, OpenCode/Aider, VS Code (with Roo Code & Continue.dev), Gemini CLI, Google Antigravity, and Google Jules.
• Tooling: OpenRouter, Anthropic API, Google AI Studio, Firebase Studio, Chrome Extensions API, Docker/Podman, DevContainers.

• Solid foundations matter: Algorithms and system design are essential to use AI effectively. How will I know the AI gave me a good solution if I don't understand the bounds?
• AI as a teammate: LLMs are multipliers, but human judgment on architecture is non-negotiable.
• Ship fast, maintain well: High velocity should not come at the cost of messy code.

Feel free to read my full story or check out my learnings.