AI engineer and technical lead. I build AI agents, MCP tooling and self-hosted inference, and I'm currently building Quites.
Before that I led AI products for the Brazilian public sector, from citizen services over WhatsApp to public-security intelligence.
- Agent orchestration with LangGraph, MCP and A2A.
- LLM infrastructure with FastAPI and GPU-hosted models.
- Data engineering on Databricks, entity extraction, fraud detection and domain ontologies.
- Product development with Python, TypeScript and React.
- wa-fake: local WhatsApp Cloud API simulator with signed webhooks, encrypted Flows, a browser inspector and TypeScript/Python SDKs.
- shadcn-govbr: unofficial shadcn/ui kit themed with Brazil's gov.br design system.
- e2e-langgraph-mcp-example: LangGraph coordinating two isolated e2e MCP browser sessions, with failure and cancellation cleanup.
- fastmcp-skill: Claude Code skill for building MCP servers with FastMCP v3.
- ui-moodboard-agent: agent that turns a design brief into a curated HTML moodboard with cited references.
- Data + AI World Tour São Paulo 2026 (Databricks): presented the BNBO ontology use case for SENASP/MJSP. It turns the free-text narratives of police reports from across Brazil into a graph on Databricks, extracting entities such as PIX keys, phone numbers and online profiles, so that scam campaigns spread across states and months become visible.
- e2e #1761: fixed ACP step cancellation in the e2e testing framework.
- Orca #40950: merged fix in Orca, the coding-agent fleet runner.
- MCP Brasil #33: contribution to the MCP server for Brazilian public data.




