For the past several months, I've been mostly quiet here.
That was intentional. After six months shipping production AI on a company's AI team, I made the decision to step back, prioritize my wellbeing, and be present for my family as we welcomed our daughter. It was the right call — the best work I did was not at a keyboard.
Now I'm back. And I'm building in the open.
What I Was Doing
While I was on the team, I was doing real things with AI — not demos, not experiments. We shipped AI to production. I learned firsthand what breaks at scale: the retrieval that works in testing and fails in the real world, the context that runs out at the wrong moment, the workflow that looked solid until real users touched it.
The thing I kept running into was a harder problem underneath: the AI had no idea how the company actually worked. It could surface a document, but it couldn't reason about the unwritten rules, the informal exceptions, the institutional knowledge that lives in people's heads and doesn't exist anywhere a model can reach. That's not a search problem. It's a continuity problem.
That's the problem I've been thinking about for the better part of a year.
What I'm Building Now
I'm in São Paulo, building an agentic orchestration framework from scratch in Python — event-driven, production-oriented, and designed so you can understand what's happening when something breaks.
The core is a node-based workflow engine: you wire up LLM calls, tools, parallel branches, and routing logic the same way you'd describe a process on a whiteboard. No magic. No framework overhead that obscures the actual behavior. Typed, tested, and designed to stay out of its own way.
On top of that framework, I'm building a multi-agent SDLC harness — a system where AI agents run the software development lifecycle (plan → implement → test → review → document) with real checkpoints and real output. I use it every day to develop the framework itself. The system is building itself.
I'm running all of this on a Mac Mini self-hosted via Cloudflare Tunnel — public face at learn-agentic-ai.com, private face on Tailscale. Low cost, real infrastructure, and the privacy argument for self-hosted AI is one I believe in.
Available for Contract Work
I'm taking on contract work — agentic systems engineering, production AI architecture, and the kind of technical work that's more interesting than it is visible.
What I'm building toward: a solo contracting practice at around 20 hours a week. Enough to do serious work without crowding out family time and the music that make it worth doing. I work selectively, scope carefully, and deliver reliably.
What I'm not doing: sprint-the-clock billable hours, chasing the framework of the week, or building AI features that aren't grounded in a real problem. If you have something real and want someone who has shipped real AI, let's talk.
Reach out at [email protected] or connect on LinkedIn.
Building in the Open, in Two Languages
I'll be documenting the build here at learn-agentic-ai.com, in English and Portuguese. The posts will be concrete: here's what I built, here's what broke, here's what I learned.
If you're wrestling with knowledge fragmentation inside a growing company — information trapped in people's heads, scattered across tools, impossible to onboard new people into — I'd genuinely like to hear how it's showing up for you.