I build AI agentsthat ship.

I came up through the unglamorous end of the stack. My first job was the integration backbone for one of Colombia's largest retailers — Java, Spring, Project Reactor, WebFlux — the kind of system where a bad decision doesn't surface as a bug, it surfaces six months later as an outage. That teaches you a specific kind of attention to detail that lasts.

Since then I've been CTO and partner of an AI content-processing platform, and I've spent years coaching engineering teams across Colombia, Ecuador and Peru. Between the two, I learned the thing this whole page is about: AI projects almost never fail on the model. They fail on everything around it.

So that's what I build. Agents on the Claude Agent SDK and MCP — a WhatsApp number that sells and books visits on its own, a hiring pipeline a recruiter runs by talking to Claude, a journal that maps what you actually believe. The retries, the state, the guardrails, and the moment the agent confidently does the wrong thing: that's the job. I don't just build these systems — I also teach how to build them properly. If I can't explain a system, I don't trust myself to ship it.