Engineering Manager · 15 years · Barcelona
Fifteen years building software, still writing code. I lead teams shipping reliable systems across AI infrastructure, maritime operations, and healthcare — clear product decisions, understandable architecture, and code that survives production.
Skills
Same skills, same words as my LinkedIn profile — the work below is the evidence.
Work
AI infrastructure
Engineering manager, hands-on. I work on execution infrastructure for AI products: orchestration, model routing, retries, observability, and the practical pieces that make model calls usable in production.
Maritime software
Led product engineering work across vessel planning, itinerary, port-call, and onboard workflows. The work sat between product decisions, backend services, frontend delivery, and team execution.
Healthcare workflows
Built healthcare interfaces for nurse-facing and operating-room workflows, where clear information, timing, and data boundaries mattered more than visual noise.
Notes
A deterministic geometry pipeline for five flower shades: differential growth, shell constraints, derived print limits, and validation-gated STL export.
· Open source · Teaching projectPatchQuest — self-learn backend developmentA self-learn backend course for DDD and hexagonal architecture in Node.js, Go, and Bun. Build in your own workspace with source-linked questions and local progress.
· Open-source setupAn agent-first macOS workbenchThe sanitized macOS setup I use for agent-first development: cmux and Ghostty for visible agent work, terminal-native review and navigation tools, deterministic output for agents, and a safe installer with privacy checks.
· Internal presentationAgent Harness at each::labsHow I use ergonomics, fast tools, specifications, sequential implementation, independent verification, and measurement to improve daily agent-assisted engineering work.
· LinkedIn articleWe timed our coding agentsA measured look at where coding-agent work actually happens: framing, verification, and review before the implementation step.
· LinkedIn articleMaking fifty AI providers behave like one productA practical note on reducing many AI providers into one product surface without hiding the operational differences that matter.
· MWC Barcelona 2024Ace the Technical Interview with AIA practical workshop from Talent Arena on using language models to prepare for technical interviews, practise with feedback, understand code, and verify the result.
More notes will be added when they are useful.
Method
The pattern behind my AI work and the notes above: an agent proposes a change, a verifier hunts for the failure mode, and a harness turns the finding into a repeatable check. People stay in charge of what ships.
Worked with
Contact
Short context, practical questions, and real work are welcome.