From physics to generative AI, by way of energy and data.
Physicist by training. I take generative AI solutions from experimentation to production, and I teach how to build them.
NowGenerative AI at Capgemini Engineering · Teaching at EBIS
I work where generative AI, software engineering and the cloud meet. I take solutions built on LLMs and agents from experimentation to systems that are actually used in production. Alongside that I teach, trying to bring the technology closer from two angles I see as inseparable: understanding its foundations and knowing how to turn them into real solutions.
I trained as a physicist. Before generative AI I worked in energy, with a master's in renewable and nuclear energy and a paper on floating wind turbines, and in data, from solar plants to MLOps. Physics still shapes how I approach problems: reduce them to their essential principles before adding complexity.
Outside work I keep building out of curiosity, from apps and small tools to electronics. This site is where I keep it all.
«The simplest solution is the right one.»Background
Generative AI Technical Lead · Capgemini Engineering
I lead generative AI solutions technically, from proof of concept to production. I design and build systems with LLMs, RAG, multimodality and agents on AWS, Azure and GCP, and I make sure they are production-ready, with evaluation, security and observability. I also support the team through technical mentoring.
Generative AI Professor · EBIS Business Techschool
I teach in three master's programmes and in specialisation courses. I teach people to understand the technology and turn it into solutions, from choosing a model to deployment.
Data engineering and data science
Before generative AI I worked as a data scientist and data engineer on solar plants, ETLs, dashboards, machine learning models and MLOps. I haven't left that world behind, because generative AI also stands on good data.
A copilot for teaching. It recalculates the schedule live so the class ends on time, and I follow it from my phone.
In use
A nostalgic platform of social mini-games, with duels between friends by code or link. It grows one game at a time.
Live and growing
A chain-reaction puzzle. Slide a brick, let the rest fall and break lines before the tide rises.
Published
A visual browser of the classification of every vascular plant in the world, built on open data from Kew. You explore it by zooming in, from the major groups down to each species.
Published
Pedro Texeira's 1656 map of Madrid, fitted onto today's aerial imagery. You compare the two with a curtain, a fade or a magnifier, and the notes tell what became of many of its buildings and streets.
Published
Access to data on remote devices for people who don't write SQL, with visual tools and plugins.
Working, private local use
Where to place dampers on the platform of a floating wind turbine to reduce its vibrations. Published in Energies.
Published
A film social network to follow what's on and upcoming releases, and talk about movies with friends.
Temporarily unavailable
An app to run programming challenges in a group, with sessions, Markdown challenges, submissions and a leaderboard.
Plug and play, not currently deployed
A temperature and humidity monitor, an internet radio and an LED hourglass.
· Criterio Tech · EBIS annual forum
With agents, code gets produced faster than we are able to understand it, and that gap is comprehension debt. I explain why it has to be paid, since what we deliver is no longer code but responsibility for the system, and how to keep it under control by verifying every task and matching oversight to the impact of each part.
In Spanish. Playing it loads YouTube.