“Deeds are love; fine words are not.” In print since 1554. The nearest English is talk is cheap — though the Spanish says more than that: I give credit to what a person has done, never to what they say they are going to do. Words do not count. Deeds do.
I architect enterprise AI that runs unattended for months, and I own what happens after it ships. Twenty-five years of engineering across CTO, CEO and founder roles — one practice, three corporate structures, and client relationships that outlasted all of them.
By then the pilot has no owner, the cost curve is unexamined, and nobody is accountable for the day the model quietly stops being right. That is the seat I take: embedded, senior enough to decide, and still there when the consequences arrive.
Not case studies. Running infrastructure — you can open them while you read this.
Multi-tenant conversational AI platform. Sole architect; still writing the architecture.
On-chain market intelligence built on Bitcoin Core nodes I run myself.
Enterprise AI and security engineering for clients across Europe and the Middle East.
Highlighted rows are the same practice under three legal structures. Four companies founded, full P&L in each. The Government of Catalonia and Barcelona City Council are the only client names on this page — both are matters of public record. Everyone else stays unnamed, which is the point.
I have hired, trained and led engineering teams since 1999, and I run a training platform used to develop technical staff. Growing engineers, architects and technical leads is not a new part of the job for me — it is how one practice survived three corporate structures and every change of market.
I advise executive teams and boards on where enterprise AI creates measurable value, where it does not, and what an organization must change structurally to absorb it. That includes technical due diligence, and the judgement to recommend stopping a programme when the evidence says it has no edge.
The layer most enterprises skip — and the reason their AI programs end up hostage to one vendor's pricing and uptime.
A proprietary five-phase security audit with a documented ethical protocol, applied consistently across private and public sector engagements.
Model-driven systems judged on evaluation criteria and failure recovery, not on demo output — with controlled human hand-off where it matters.
MCP servers and agent guardrails released openly, derived from real production failures rather than from a whiteboard.
A multi-provider LLM gateway that moved live agent traffic off single-vendor inference without modifying a single agent.
Google Partner certified · Published writer on AI, cybersecurity and technology leadership, with 5,000+ professional followers · Speaker · Volunteer mentor to entrepreneurs and technology founders through CP'AC
I don't name clients.
Not to impress a hiring team, not in a deck. It is the same discipline a Fortune 500 board expects to be extended to them.
I alert on silence.
A 30-day outage once passed undetected because conventional monitoring only watches for errors. Systems now report that they are working, not just that they failed.
I stop what has no edge.
Including my own systems. The failure mode in enterprise AI is rarely the model — it is that nobody senior is willing to call it.
Based in Dubai, UAE, covering US hours. I already run production work remotely for clients on three continents and need no sponsorship to do it — a remote engagement can start immediately. Relocation to the United States is on the table too, and that is the part that would require sponsorship.
