Backend engineer working in Go and TypeScript, with a focus on AI agents, context engineering and self-hosted infrastructure. Based in São Paulo.
I build systems for real operations: sales-planning tooling that reconciles data across ERPs, portals and spreadsheets; a Go + Svelte order and quoting system integrating shipping carriers and supplier stock; and agent infrastructure: a coding-agent harness, MCP servers, and context engines with permissions, provenance and audit built in.
Daedalus: coding-agent harness in Bun/TypeScript, originally derived from Pi. Custom context compaction, schema-validated executable plans with dependency and file-conflict checking, task packets for subagents, managed worktrees, local GUI. Currently paused.
MercadoLivreMCP: read-only MCP server and Codex plugin for the Mercado Livre seller API. OAuth with PKCE, rotating refresh tokens, tested.
Competitive-Intelligence-Classifier: dual-model text classifier (AutoGluon) that turns ambiguous import records into competitor product intelligence. Built for BASF's BI team; cut a 14-day manual analysis to minutes at ~95% precision, with human review for low-confidence predictions.
Archon: fork where I'm working on workflow and worktree identity for deterministic AI coding runs.
Client and venture code (a commercial-planning system in production, an AI operator for marketplace sellers) is private.
Go · TypeScript / Bun · Svelte / SvelteKit · Python · PostgreSQL · Docker · Linux (CachyOS + Hyprland) · Tailscale · GitHub Actions
Agent orchestration, durable memory and temporal state for agents, bounded autonomy, retrieval (vector, hybrid, graph), coding-agent harnesses.
Marketing student at USP; I like understanding the business before building the system.