Staff AI Engineer · System Architect · Beatmaker São Paulo, Brazil · @clipboardhealth · marcelscog.vercel.app · cal.com/marcelxv

8+ years building production systems — from event-driven backends to AI automation pipelines. Currently architecting an AI-driven document validation platform at Clipboard Health that processes ~1,000 documents daily at 70%+ automated approval rates.

Background in journalism gives me an edge most engineers don't have: I document systems clearly, communicate across disciplines, and treat writing as a first-class engineering skill.

A DSL and runtime that sits between LLM reasoning and real-world actions. Rune policies encode decision logic as finite state machines — same inputs, same terminal. Always. Built for teams that need auditable, version-controlled business rules inside probabilistic AI systems. TypeScript DSL AI infrastructure

CLI tool that brings LLM reasoning into the development loop — conventional commits, branch summaries, diff explanations. Opinionated and fast. Shell AI tooling DX

High-throughput document processing in Rust. 2.8× faster and 50–70% cheaper than the Node.js equivalent it replaced. The production case study behind Rune's ingestion pipeline. Rust Document processing Performance

Educational project demonstrating a production-grade RAG pipeline backed by a relational database. No vector DB required. JavaScript RAG pgvector AI

MCP server implementation for exposing tools to Claude and other MCP-compatible agents. TypeScript MCP AI agents

Production-ready boilerplate for teams running multiple Next.js apps under one domain. TypeScript Next.js Microfrontend

Languages

Frontend

Backend & Data

AI & Automation

Infra & Cloud

I publish on Substack — AI integration, system architecture, automation, and dev practice. Mostly in Portuguese, some in English.

software engineer · adventurous entrepreneur · beatmaker