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 build AI-native tools, browser extensions, and micro-SaaS products.

Currently, I operate as a Full Stack Product Engineer, using AI agents as force multipliers to ship code rapidly without sacrificing architectural quality or maintainability.

I am actively building products with AI Agents, studying their impact on Developer Experience (DX), and testing with self-hosted infrastructure.

In DX, I'm developing and testing a 4-step AI-assisted workflow to automate many parts of the development workflow. The goal is to achieve better results faster while reducing the cost of development.

An automated collection and notification system built for B2B.

• Why it matters: Demonstrates the ability to solve concrete business problems using event-driven workflows and the simplest tools that allow one to validate the idea: the MVP was built using n8n and the Evolution API (running on my custom infra), with no custom code or complex deployments. Once the idea is validated (we are onboarding our first client), we can consider the next steps.

A browser extension for custom CSS/JS modifications, featuring an integrated AI panel for real-time DOM manipulation.

• Why it matters: Honestly, I'm just building this to scratch my own itch using tech I already know well. The fun part is bridging the gap between browser extensions and LLMs: instead of me having to inspect elements and manually write Greasemonkey-like scripts, the AI gets direct access to the DOM to write and apply visual or functional patches on the fly.

A system-wide keyboard remapping utility that allows typing Brazilian Portuguese and English on a physical US-layout keyboard without having to change the system layout. No more win+space!

• Why it matters: I built this leveraging AI to deploy complex, low-level system logic in languages I had zero prior experience with, proving that strong fundamentals translate across any syntax.
  • v1: Rust. A functional desktop tray application built in just a couple of days with zero prior Rust experience. The goal was to test if my ideas would port well to a strictly typed language using a spec-driven AI workflow. Built using Kiro (kiro.dev). Code here.
  • v2: AutoHotkey v2. The complete version that achieves my goal. I use it every day, all the time, and it saves me from a lot of frustrating typing mistakes. Brazilian coders will understand.

Custom VPS setup orchestrated with Coolify.

• Why it matters: I set this up to learn more about infrastructure and, eventually, have the actual knowledge to decide if it is worth moving away from PaaS or not. I had never managed a server before and it has been an interesting experience, especially because it opens many self-hosting options that I didn't know about. I learned the practical basics of server management and Docker orchestration. It currently hosts my LLM router (LiteLLM) and n8n pipelines.

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.

• Adapta AI Challenge: Top 3 Finalist in Brazil's largest AI hackathon. Built an OOH analytics MVP using Next.js and AI APIs in ~24h.
• devmagic: A devcontainer setup that solves the "works on my machine" problem.
• Viralei: Hackathon MVP built in 4 days that turns legislative bills into engaging videos.
• To be made open source when I finish the initial write up and websites:
  • Fair Pricing Works: Initiative advocating for Purchasing Power Parity (PPP).
  • Do I Need A License? Project that I created when answering that question for myself.

I choose tools that mix personal preference, velocity, and leverage.

• Core Languages: TypeScript, JavaScript, Python.
• Main Stack: React, Next.js, Node.js, Tailwind CSS, shadcn/ui, Zustand, Zod, Extensions API.
• Main Dev Tools: Cursor, Zed, OpenCode/CLI agents, VS Code + GitHub Copilot, DevContainers.
• Infra & Tooling: Docker, Podman, Coolify, n8n, LiteLLM, PostgreSQL/Supabase, OpenRouter, Tailscale, Cloudflare.

• 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.

Footnotes

1. Widely regarded as one of Brazil's top engineering institutes (Top 1%), known for its extremely rigorous entry and high academic standards. ↩