~/about
Fernando Valdez
AI software engineer in Barcelona, born in Buenos Aires. I came into this trade by automating my own job, without a computer science degree, and I have been putting systems into production since 2021.
What I do
where the last four years wentFor the last four years I have worked at ioBuilders on a digital trust platform for the legal sector: qualified electronic signature, certified multichannel notification, evidence sealed on TSP and blockchain networks, and digital identity. Around fourteen multi-tenant Micronaut services running on AWS and Oracle Cloud.
I work the whole lifecycle: refining requirements with product, writing the Groovy and Java that implement them under DDD and hexagonal architecture, and tracing a production incident through dead-letter queues, Loki logs and Prometheus alerts until the root cause turns up.
Agents that do real work
read-only by defaultI founded the shared standards repository my team works from and wrote its corpus: machine-readable architecture rules distributed as a git submodule across sixteen repositories, so people and agents follow one playbook. On that base the team built the twelve command-line skills we run daily for incident triage, dead-letter queue review, merge request review and release notes.
One rule I do not break: agents are read-only by default. Nothing mutates production without a human approving it.
What I take on
by project, one at a timeBackend from zero to production
The first commit, the domain model, the services around it and the deploy that puts it in front of users. I design under DDD and hexagonal architecture, with events where the domain asks for them, and I carry it until it holds real load.
Agents that do real work
AI agents inside the daily operations of a regulated backend team: incident triage, dead-letter queue review, merge request review, release notes. A standards corpus split into atomic files so people and agents follow one playbook. Read-only by default; nothing mutates production without a human.
Data with legal weight
Data where every access is audited and the model carries legal consequences: qualified electronic signature, certified notification, evidence sealed on TSP and blockchain networks, digital identity. In my own products the same idea, smaller: who can read what, decided in the database.
Products end to end
My own products, from the landing page to the app store: Next.js and Supabase, Capacitor or Expo for mobile, Stripe for payments, three languages from day one. The same person decides the schema and the button.
Keep it running
Production is where a system proves itself. I trace incidents through dead-letter queues, Loki logs and Prometheus alerts until the root cause turns up, and I still maintain what I shipped for clients.
How I got here
Business administration degree, then admin and IT support work. I started automating my own job with scripts, which turned out to be the door in.
First developer job, building interfaces in React for client projects.
Joined on the frontend, moved to the backend team after four months, and stayed.
Two products of my own, both live: Futer, on my own, and Luckybites with a co-founder. AI agents in the daily operations of my team, and the search for the person who takes Futer further than one engineer can.
Stack
Groovy, Java 21, Micronaut, RabbitMQ, PostgreSQL, Spock.
Kubernetes, ArgoCD, Kustomize, AWS, Oracle Cloud, Grafana, Loki, Prometheus.
TypeScript, Next.js, NestJS, Supabase, Astro, Expo.
Claude Code, MCP, and firm opinions about permissions.
Contact
Write to me if you are building something that needs a backend taken from nothing to production, if you want to compare notes on agents doing real operational work, or if you think you are the co-founder Futer is looking for. I read everything and I reply.