Álvaro Zamorano

AI Systems Architect, Madrid

Autonomyby design.

I design AI systems that run on their own, and the checks that prove they work.

Kit, a fox made of blocks. Break him and he rebuilds himself, then checks his own work.

Building Kit

Selected work

Things I built that you can open and check for yourself.

  1. AgentOS

    Agents that finish

    An agent runtime that finishes what it starts. Each mission carries a definition of done written as code, and a separate verifier decides when it's met. A human only steps in for payments or anything that can't be undone.

  2. LiveKnowledge

    Reports you can question

    Turns a long consulting report into something clients can question. Every answer points to the exact page it came from.

  3. HABLA

    Websites for AI agents

    An audit that tells you whether AI agents can find, read and cite a website. Each check runs as a single command, so anyone can repeat it.

  4. esGEO

    Getting cited by AI

    A method for writing content that AI models quote, turned into tools and a course for Spanish-speaking teams.

  5. This site

    Built from blocks

    Kit is rebuilt from blocks in your browser, live. Every release has to pass the same checks for layout, speed and accessibility before it ships.

AlsoPS2WEB, a PlayStation 2 emulator in the browser, bring your own games.Plano 3D, an architect's floor plan turned into checked measurements and a 3D view.

Built. Checked.

How I build

Starting an agent is easy. Knowing when it's done is the hard part, so that's where I start.

  1. Define done

    Before anything runs, I write down what finished looks like, in a form a machine can check.

  2. Let the agents work

    A team of AI agents does the work with its own tools and memory, without waiting on me.

  3. Check every result

    A separate step checks each result against that definition. Anything that fails goes back and gets redone.

  4. Call a human when it matters

    People step in for payments and for anything that can't be undone. Nowhere else.

Verificationover vibes.

If a machine can't check that something is done, it isn't specified yet.

StudioAZM, my film studio

Impossible scenes.No shoot.

A generative model is a camera without an operator. We bring the script, the optics, the light and the cut, and brands get scenes that would be impossible or too expensive to film.

Studio reelWatch the reelES

Work for brands

Commission a scene

www.studioazm.com, in Spanish

Writing

Kit's log

What building agents taught me, one scene at a time, and where I wrote it up.

  3. A machine-verifiable definition of doneEssay, on GitHub
  4. Most agents are demos in a costumeVideo, on LinkedIn
  6. The moat keeps movingVideo, on LinkedIn
  7. How I buildOn this page
  8. Your next reader is a machineCarousel, on LinkedIn ES

About

I build AI-first. I direct the tools, write the checks, and only trust what passes them.

I'm Álvaro, an AI systems architect based in Madrid.

By day I work on AI for large enterprises. By night I build with it.

Industrial engineer, MSc from Universidad Politécnica de Madrid. Anthropic certifications in Claude: Architect, Developer and Foundations.

Kit is the fox that builds on its own while you sleep. On this page he's also the one who fixes himself.

I also direct AI films at StudioAZM

Let's build something that runs on its own.

Tell me what should happen without you, and how we'll know it did.

Message me on LinkedIn