AI engineer
Career
Open to AI engineering roles and to freelance projects.
I build systems where AI agents do the work and people keep control: permissions a person sets, a review before anything ships and a record of every step.
- Based in
- Porto Alegre, Brazil
- Work
- Remote, or relocation anywhere, including abroad
- Languages
- Portuguese (native), English (fluent)
- Focus
- AI agent systems with permissions, review gates and audit trails
AI engineer track
Six pieces of work. Each one says what I built and what I led, and links to the page that shows it.
A model gateway and a human gate
Icarus is an app factory run by AI agents. Each agent role reaches its model through one gateway (Claude, Gemini, GPT, Codex or a local model), the discovery, spec and judging steps must answer in JSON that the code validates against a schema, and spending caps and a stop switch apply. A person approves each build at a gate that refuses one that failed its checks.
Structured output and slides drawn in code
Loop is a marketing team made of AI agents. The model answers in fixed JSON shapes that schemas check before anything is drawn, a renderer written in code draws every slide so the look never drifts, and a separate reviewer call checks the copy for invented claims. Nothing posts without a person.
Agent permissions and an audit trail
Anima is a hub for running several companies, where coding agents call an HTTP API as users of their own. The database checks every call against the permissions a person set per workspace and module, tokens are stored only as hashes, risky actions start switched off, and every agent write records the person, the agent, the token and the model the agent reports.
Readiness checks and claims for agent work
Tempo is a task board shared by people and coding agents. A ticket is ready for an agent only when it has acceptance criteria, a way to verify the result and no placeholders left. Agents are meant to claim tickets on leases that expire, so a crashed agent never blocks the work for good.
A review gate and signed CI receipts
The delivery tools that let AI coding agents write code at Lucentive. Two independent model families review each pull request, bound to the exact commit being merged; CI receipts are signed and checked again at merge; isolated worktrees and ownership rules keep parallel agent work manageable.
Specifications that coding agents build against
For ANANSI, a private messenger, and Acesa, a 3D property showcase, I write the specification and direct the AI coding agents that write the code. In ANANSI a ticket closes only with source, commands, results and an independent review recorded. In Acesa an agent turns written art briefs into Blender scripts, and a checker rejects models too heavy for a phone.
What is shipped, and what is not
The delivery tools were built and operated at Lucentive, and IAS was built with the Lucentive team; an earlier version is archived at ias.dev. Anima, Tempo, Icarus and ANANSI are in development and not published, Loop is a private prototype and Acesa is a demo. Each project page says what works today and what is still a plan.
Experience
Feb 2026 to presentRemote
Co-founder & AI Engineer
Lucentive
Co-founder of Lucentive and product owner of IAS, a platform where fleets of AI agents turn business requests into production software, under policies, gates and a full audit trail.
- Built and ran the agent-fleet workflows behind the company’s day-to-day development.
- Co-built the review gate where two independent model families must approve the exact commit, and built the service that runs it unattended.
- Built the CI runner that picks the checks each change needs and signs a receipt that is verified again before merge.
- Built ALMA’s security gateway, which opened a live multi-tenant product to outside AI agents: machine identity, signed requests, scoped permissions and rate limits.
- Moved the company’s work record off a third-party tracker into a self-hosted ticketing system shared by people and coding agents, without losing an issue.
Apr 2024 to Jan 2026Porto Alegre
Systems Developer
Freelance
- Replaced small businesses’ spreadsheet routines with LLM-powered reporting and client communication.
Sep 2023 to Apr 2024
Pedagogical Coordinator
Centro Profissionalizante Ennet
- Set up a new school network’s student retention from zero: the playbook, an online teaching library and on-site training for the Belo Horizonte team.
Jan 2021 to Feb 2023
Pedagogical Coordinator
Enjoy Inglês Profissionalizante
- Improved student retention with automations built from SQL and spreadsheets.
- Built an enrollment program that was adopted across the school network.
Oct 2020 to Jan 2021
Instructor (English, IT & Digital Marketing)
Enjoy Inglês Profissionalizante
2015 to 2017
Early career
Polo Marista de Formação Tecnológica
- Junior front-end developer, as an apprentice (2015 to 2016).
- Volunteer in computer maintenance and educational projects (2016 to 2017).
Skills
AI systems
- Multi-agent orchestration
- Context engineering
- Tool calling and MCP integration
- Structured output with schema validation
- Provider failover and token telemetry
- LLM-as-judge and regression evals
- Scoped agent permissions, human in the loop, audit trails
- LoRA and QLoRA fine-tuning (personal projects)
Product engineering
- TypeScript
- Python
- Node.js
- React and Next.js
- SQL and PostgreSQL
- Supabase
- Three.js and WebGL
Tooling
- Claude Code, Codex and Gemini as coding agents
- CI/CD
- Git worktrees
- Playwright
- Vite
- Cloudflare Workers
Education and languages
Education
- BA, English Language and Literature, Uniasselvi (5th semester, on hold)
- Informatics with emphasis in Robotics, Polo Marista de Formação Tecnológica, 2014 to 2015
Languages
- Portuguese (native)
- English (fluent)