I reverse-engineer hardware protocols, build tooling for AI coding agents, and run ISP infrastructure.
Based in Argentina (Córdoba / Eduardo Castex, La Pampa) · Open to remote work
Most of my work sits in three places that rarely overlap in one person:
Hardware nobody else supports. I capture the USB/HID traffic of proprietary Windows drivers, decode the protocol, and reimplement it in userspace for Linux and macOS. That means reading raw byte offsets until a display lights up — figuring out that buf[9] carries CPU temperature rather than the per-thread maximum, or that you must select an HID interface by usage_page or every write returns EPIPE.
Tooling for AI coding agents. Not prompt tinkering: spec-driven orchestration, persistent memory, verification workflows, MCP servers, and configuration harnesses that survive real projects.
Networks that have to stay up. I have worked with FTTH and ISP infrastructure since 2017 — fibre, servers, and the operational side where downtime is measured in angry phone calls, not error budgets.
The web and mobile work sits on top of that foundation, not instead of it.
Python · USB HID · Linux / macOS / NixOS
Userspace drivers for PC hardware with no official support: AIO cooler displays, Gigabyte RGB Fusion (ITE8297), USB lighting controllers. Includes the complete protocol analysis, so the findings are useful even without the code. Ships systemd units, a macOS LaunchAgent, PKG/DMG installers and Nix flakes.
Rust · Tauri · systemd
Open-source driver and web panel for the Redragon SS-550 Stream Deck on Linux. Multi-page layouts, custom icons, autostart, and installers for Arch, Ubuntu and Fedora — for a device the vendor never supported outside Windows.
Python · MCP · CLI
Spec-driven orchestration for AI-assisted delivery: persistent context, verification workflows, CLI tooling and an optional read-only MCP server. Built to stop agents from drifting halfway through a feature.
PowerShell · Agent configuration
Configuration harness for OpenCode and OhMyOpenCode — agents, hooks, skills and templates. My most-starred project, and the one people found on their own.
Python · RAG · Qdrant · Next.js
Domain-expert AI agent for Argentine kinesiology: RAG over a curated knowledge base, DeepSeek as the LLM, Qdrant as the vector store, deployed with Docker Compose behind Nginx.
Dart · Flutter · Android / Windows
Encrypted, offline-first password manager with biometric unlock, autofill and signed updates.
Shell · Agent configuration
Enterprise Claude Code setup: 8 specialised agents, 20 invocable skills, 7 context rules, 8 automated hooks and 19 MCP server integrations, with an installer.
| Area | Tools |
|---|---|
| Systems & hardware | Linux (Arch, CachyOS, NixOS, Fedora), Rust, Python, USB/HID, systemd, reverse engineering |
| Backend | Python (FastAPI), Node.js, NestJS, PostgreSQL, Docker |
| Frontend & mobile | TypeScript, React, Next.js, Astro, Tailwind, Flutter |
| AI engineering | MCP servers, agent orchestration, RAG, vector stores, spec-driven development |
| Infrastructure | NixOS, GitHub Actions, Nginx, FTTH / ISP operations |
- TecnoDespegue — my agency: internal software and process automation for LATAM clients
- YouTube · @tecnodespegue — project walkthroughs and tooling deep-dives, in Spanish
Open to remote roles and contract work — full-stack development, developer tooling, Linux/DevOps, or hybrid engineering and infrastructure positions. Also available on-site in Córdoba and La Pampa.
Interested in how I work? See COLLABORATION.md.