I build the harnesses that let AI agents do real engineering work unattended, and the checks that keep that work honest. I care about software that proves itself: every change exercised in a real browser or terminal before it ships, every claim backed by a receipt. I'm a software engineer at Convenia, and I run 30labs, a studio that ships systems and sites in weeks, not months.
~/workshop
Projects
otium
Runs unattended Claude Code sessions across many projects: picks up approved work, ships pull requests, validates its own changes in a real browser, and hands off before its context fills.
Outcome: 4,000+ pull requests merged across 30 projects in its first two months.
organon
Claude Code plugins that close the gaps around a session: gated pushes, grounded handoffs, parallel specialist review, and hands-on validation in a real browser. I designed and maintain all of them.
Outcome: 17 plugins, 200+ merged pull requests.
ethos
One versioned set of operating rules for AI coding agents, installed once and imported by every tool I use.
Outcome: Governs every agent session across 30+ projects.
~/studio
Studio
30labs is my studio in São Paulo. We build systems, sites and audits with a fixed scope and a fixed price, delivered in weeks, not months, and consult on growth, brand and AI.
Systems
software that automates work and connects the tools a company already uses, including agents that answer on its behalf
Sites
fast, phone-first, found in search, easy to change later
Audits
a written diagnosis of a system someone else built: what is exposed, whether it holds up, and what it costs to keep