Agent memory for systems that store everything and know nothing. 10+ years of backend and AI. Two of these running in production today.
The brain layer is what I'm building right now. It sits on ten years of everything else.
Not demos. Agents with paying users, error rates I can quote, and architecture decisions I reversed in public. I built VTEX's Copilot with RAG and multi-agent orchestration, and I've run two of my own in production since.
Appcitas has been live since 2015. WalioAI is my 2026 bet. Real customers, two currencies, two markets, and pricing that had to survive contact with a Colombian small business.
Delivery routing for 400+ stores, checkout that survives Black Friday, and a production AI stack that costs under two dollars a month. None of it matters if it pages you at 3am.
Products people pay for. Not prototypes.
My 2026 bet. 25,232 leads captured so far. Your AI agent responds, qualifies, and emails you when they're ready to close. Pay per qualified lead.
Personal brand site for an architect and content creator with 60k+ followers. Built with Next.js and Tailwind, scoring 97+ on Lighthouse.
107,750 appointments booked and counting. An AI WhatsApp agent for healthcare clinics across Latin America, running since 2015 on Node.js and AWS.
Technical deep-dives. No fluff, no hot takes.
Every night at 2:30am my appointment system consolidates what it learned, then forgets on schedule. The four steps to give any product a memory.
Every new session with Claude Code, Cursor, or Windsurf starts from zero. Agent OS fixes this by capturing your team's conventions as injectable context.
Building an AI copilot for enterprise e-commerce: from a chatbot that couldn't handle B2B complexity to a custom RAG system with multi-agent orchestration.