Two languages growing up. Three countries since.
Where I come from
Every stop added a language, a discipline, or a kind of person to work with — and the route between them matters more than any single line of the CV.
Why the trains: I’m a steam-train enthusiast, and a career runs a lot like a railway line — stations you chose, and a next one you can’t see yet.
IPamplona
2020 – 2024
Spanish and Basque growing up, and a computer-engineering degree at UPNA in Pamplona.
IIBrno
Erasmus+
A term at Brno University of Technology, working in English with classmates from across Europe — where being understood mattered more than being right.
IIIMadrid
2024 – now
A master’s in applied AI at UC3M, predictive maintenance alongside Airbus maintenance engineers — people who know the aircraft, not the model — and client work at WhiteBox.
IVNext station
points not set yet
I don’t know yet — and I’d rather choose it well than fast.
IPamplona
2020 – 2024
Spanish and Basque growing up, and a computer-engineering degree at UPNA in Pamplona.
IIBrno
Erasmus+
A term at Brno University of Technology, working in English with classmates from across Europe — where being understood mattered more than being right.
IIIMadrid
2024 – now
A master’s in applied AI at UC3M, predictive maintenance alongside Airbus maintenance engineers — people who know the aircraft, not the model — and client work at WhiteBox.
IVNext station
points not set yet
I don’t know yet — and I’d rather choose it well than fast.
Selected work
Three problems, three kinds of proof
Agentic RAG for natural-language database querying
Master’s thesis at UC3M: an agentic retrieval-augmented system that answers natural-language questions against complex databases, designed to stay robust when the schema is not.
Public-tender intelligence pipeline
For Mahou San Miguel: real-time RSS ingestion, LLM extraction of ~30 fields from heterogeneous tender PDFs, and a Palantir ontology that turned scattered documents into one queryable foundation. ~300,000 tenders in the initial backfill.
Board Game Assistant
A cited-RAG rules assistant built solo: hybrid dense and sparse retrieval in Qdrant, cross-encoder reranking, LangGraph orchestration, and a hexagonal architecture whose dependency rule is enforced by import-linter in CI. Retrieval recall@5 of 100% on a 100-case golden set I authored.