AI and platform architect. I design agent systems and the infrastructure that keeps them running in production.

Ten years building and leading infrastructure teams: CI serving 300+ engineers and 20k builds a day, global on-call across 20+ teams and 100+ applications, keyless identity migrations, and platform work on Google Cloud. These days most of my attention goes to production AI systems: multi-agent orchestration, MCP servers, RAG pipelines that ground agents in real context, and the observability needed to keep any of it honest.

Structure first, tools follow. Based in Valencia, Spain.

vault-mcp · MIT A production MCP server that gives any agent direct access to an Obsidian vault, with no plugin and no Obsidian process required. Hybrid keyword and semantic search (SQLite FTS5 plus embeddings), a capture and triage pipeline, OpenTelemetry tracing, and health and ready probes.

excalidraw-agent-skill · MIT A Claude Code skill that generates architecture diagrams through isolated agents, so thousands of JSON tokens never reach your main context. Offline renderer, validation, visual critique loop, and optional animated SVG export.

personal-ai-os-starter A local-first starter kit for running your daily knowledge work through agents. Markdown files, no servers, no SaaS. Built for the Build Your Personal AI OS workshop at GenAI Summit EU 2026.

Multi-agent framework work that is not public yet: a DAG workflow engine running 20+ workflows with orchestrator and worker execution, an MCP-over-HTTP server that exposes workflows as tools, sandboxed agent execution in Docker and git worktrees, and a multi-model code review loop across Claude, Codex and Gemini.

Speaking: GenAI Summit EU 2026, Valencia. Hands-on workshop, Build Your Personal AI OS.

I write about AI-assisted engineering at thinkink.io · LinkedIn