Backend engineer specialized in Node.js, TypeScript, and Go, with deep focus on distributed systems, developer tooling, and production-grade LLM applications.

I design APIs and platforms that survive real traffic — clean boundaries, strong observability, and the kind of architecture that lets teams ship fast without breaking things. Lately, much of my work sits where backend engineering meets AI: agent workflows, retrieval pipelines, and LLM systems wired into production codebases.

I value pragmatism over hype, fewer abstractions over clever ones, and tools that get out of the developer's way.

Open Source

• 🐹 gos — Install and switch Go versions in seconds. One Bash script. Zero dependencies. macOS · Linux · Windows.  
• 🎭 electron-stagewright — Drive Electron apps the way Playwright drives browsers — an MCP server built agent-first for AI coding agents (Claude Code, Cursor, Codex, Cline). Launch apps, query the accessibility tree semantically, assert with retrying expect_* primitives, and replay session traces with per-tool token budgets. Docs  
• 🖼️ vitrine — Turn code into beautiful images, straight from your macOS menu bar. Native, instant, fully local — no account, no network. Website  
• 🌐 lingua-marketing — Public marketing site for Lingua (linguacode.dev). Built with Astro 6, Tailwind v4, and deployed to Cloudflare Pages.  
• 🍺 homebrew-gos — Homebrew tap for distributing gos across macOS and Linux.  

Currently Building (private)

• 🤖 Janusly — An AI operator for business workflows: a DAG runtime where AI is part of the loop, not glued on top. The differentiator is the failure-recovery loop — AI patch suggestions with self-rated confidence, sandbox replay before commit, cluster apply across DLQ entries sharing a failure signature, and one-click rollback. Generic workflow execution (durable retries, decision engine, RL adjustments, NL run explanations) is the table-stakes layer underneath. With an Anthropic key, it becomes end-to-end: prompt → workflow → execution → decision → learning → recovery → rollback → conversational explainability. Without one, every deterministic path still works.  
• 🛒 Puntovivo — Full-stack POS platform (v3). TypeScript + tRPC, multi-vertical (retail, restaurants, delivery), offline-capable, with AI assistants (invoice OCR, in-app copilot), thermal-print receipt rules, and a custom design system for operator-grade UI.  
• 🧠 Lingua — linguacode.dev · Multi-language desktop code runner — JavaScript, TypeScript, Python, Go, and Rust in one offline-first, Monaco-powered app. Combines Monaco Editor, a project file tree, inline console output, and language-specific execution backends for both desktop and web builds. The multi-language answer to RunJS: same "open, write, run" ergonomics, but with Go, Rust, and Python as first-class citizens instead of JavaScript-only. The marketing site is open source; the core product is private and in active development.  
• 📋 Gancho — Smart Clipboard for the Apple ecosystem: clipboard history + a searchable snippet library for Mac, iPhone & iPad. Private by design — your data stays on your devices.  

Areas of ongoing client / contract work (generic)

• High-throughput Node.js / NestJS APIs and queue-driven workers with strong observability.
• Go microservices where concurrency, latency, and footprint actually matter.
• Production LLM systems — Claude / OpenAI agents with tool use, structured outputs, and grounded retrieval.
• Internal developer tooling — CLIs, scripts, and platform glue that compresses friction across engineering teams.

Languages

Backend & APIs

AI & LLM Engineering

Data & Storage

Infra, DevOps & Observability

Drive Electron apps the way Playwright drives browsers — but designed for AI agents, not adapted for them.

An MCP server that lets AI coding agents operate real Electron desktop apps. Agent-first from the primitive level up: errors carry recovery hints and similar-ref alternatives, every response reports its own token cost, composite wait_for_state predicates collapse multi-call chains, and retrying expect_* tools replace read-compare-retry loops. Ships replayable session traces with per-tool token budgets.

claude mcp add electron-stagewright --scope user -- \
  npx -y --package @electron-stagewright/core --package playwright electron-stagewright

Install and switch Go versions in seconds. One script. Zero dependencies.

A single Bash script that replaces the manual pain of managing Go installations. Auto-detects OS and CPU architecture. Available via curl | bash, Homebrew, and more.

gos latest        # install latest stable Go
gos install 1.21  # pin a specific version
gos current       # show what's running

Turn code into beautiful images, straight from your macOS menu bar.

A native macOS menu-bar app that turns snippets into share-ready images. Native, instant, and fully local — no account, no network, nothing leaves your machine. Visit the website →

• Boring tech, sharp execution. Pick proven tools. Spend the novelty budget where it actually matters.
• Observability is not optional. If you can't see it in production, you don't own it.
• Small interfaces, composable parts. Most complexity comes from premature abstractions, not missing ones.
• Type the boundaries. TypeScript and Go aren't religions — they're contracts that catch bugs before users do.
• AI is a system, not a prompt. Real LLM apps live or die by retrieval, evals, guardrails, and observability — not the model name.
• Ship, then sharpen. A merged PR teaches more than a perfect plan.