Based in Brazil (GMT-3), working remotely with US teams. Over 18 years I went from hands-on full-stack and blockchain engineering to leading a 25-engineer team across the US and LatAm, and I still work hands-on in architecture and code.
- 4 AI agent systems in production at Oak, instrumented for cost, observability and usage.
- An agentic development culture across the team: spec-driven delivery, a separate agent reviewing every PR, and AI coding measured by cost and usage per developer through self-hosted Claude Code telemetry (OpenTelemetry, Prometheus, Loki, Grafana).
- Engineering for a crowdfunding protocol with 1M users, from architecture to infrastructure.
- ai-workflow: a spec-driven, trunk-based delivery workflow for Claude Code. You decide what to build, agents build it from the spec, a separate agent reviews it, you merge.
- cdp: Credential Delegation Protocol (work in progress). An open protocol and Rust reference implementation so AI agents get authenticated access without ever seeing the credentials.
- spec-kit-preset-claude-ask-questions:
a Spec Kit preset that turns
/speckit.clarifyand/speckit.checklistinto Claude Code's native question picker, with a recommendation and its reasoning. - claude-code-mods: function-hook plugins for Claude Code, such as an English coach, a toolbar and a status line.
- ecies-bls12381: a Go package for ECIES encryption over the BLS12-381 curve.
- tailscale-exit-node-terraform: a self-hosted Tailscale exit node on AWS, deployed with Terraform.
- We instrumented Claude Code for our team. Here's what the data showed
- Leading with heart: what 18 years taught me about managing engineers
- Languages: TypeScript, Go, Python. Also Rust and Solidity.
- AI: LLMs, agentic systems, multi-agent orchestration, Claude Code tooling
- Backend and web: Node.js (NestJS), React, Next.js, PostgreSQL, API and SDK design
- Infrastructure: AWS, Kubernetes, Docker, Terraform, CI/CD, OpenTelemetry, Grafana
- Blockchain: EVM/Solidity, Solana, protocol design
Security shapes how I design, though it is not my day job: CEH, threat modeling and secure system design. I publish independent research as 0xrafasec: CVE deep-dives, malware analysis and blockchain security.
Conversations about AI engineering, technical leadership and running AI systems in production.
Portuguese (native), English (professional).




