Where language meets technology.

The challenge. Game rules depend on official terms, recurring phrases and multi-word expressions. General machine translation can communicate the meaning while missing the game’s established terminology.

My approach. After taking an introductory Python course, I applied what I had learned to my thesis: I analyzed the game’s language, compiled an English–Spanish terminology and phraseology glossary, and coded a prototype translator based on it. Giving glossary entries priority levels, so that longer fixed structures are matched first, helped preserve the game’s set phrasing.

The result. I compared my prototype with the official card translations and with DeepL, Reverso and Google Translate. The prototype kept terminology and fixed phrasing more consistent, while the general engines produced more fluent grammar. A key takeaway for post-editing: the prototype’s grammar errors are easy to spot, whereas a general engine’s terminology errors can hide behind natural-sounding sentences.

What I learned. Bugs like “graveyard” turning into “cementrenio” (the term “enter” was being replaced inside the word) taught me that programming problems in translation are, at heart, textual problems. I had started learning Python from scratch, so carrying the project through to my thesis defense, graded 9.9/10, was a challenge and a path I’m glad I chose.

The GitHub repository also includes the thesis abstract and conclusions in Spanish and English.

Flotar, directed by Alexia Corrales Rosalén, explores love, acceptance, fear and, ultimately, letting go.

My contribution. I translated from Spanish into English and worked on subtitles in both languages. I loved being part of the project, and found it a valuable introduction to audiovisual translation.

Follow Flotar on Instagram ↗

Selected extracts from several English- and German-to-Spanish legal translation assignments completed during my degree (2025–2026). Personal details from the birth-record sample are omitted.

The problem. Clients may send long PDF style guides and extensive Excel glossaries. Looking up tone, untranslated items, date formats and preferred terms by hand can slow down translation.

Course context. Final project for the Generative AI for Everyone: Build Real-World Applications with Language Models course at Universidad Carlos III de Madrid (October 2026).

The prototype. A chatbot that uses retrieval-augmented generation (RAG) to find the relevant guidance in a client’s uploaded style guides (PDF) and glossaries (Excel) and answer in context.

Example questions. “How does this client want ‘dashboard’ translated?” or “Does this paragraph follow the client’s style guide?”

Why it matters. It explores how external client documentation can be connected to an LLM to support consistent, context-aware decisions in a translation workflow.

For the 16th ECOshow, my team created an interactive map of places and services that could help people from Slavic communities as they settle into Valencia. I was delighted to take part in a project designed around a real community need.

My contribution. I built the interactive map in Google My Maps, organizing the places into categories such as restaurants, shops and groceries so people could find what they needed at a glance.

Our team proposed a concept for connecting tourists with older adults experiencing loneliness through guided tours, creating opportunities for language learning and human connection. The app remained at concept stage and was not developed.

My contribution. I filmed and edited the short video introducing the group’s idea. I was delighted to take part in this project and share our proposal at ECOshow.

This project explored how audio description can make museum content more accessible to visitors who are blind or have low vision. It brought together language, careful observation and clear spoken description in a museum context.

My contribution. I wrote and recorded audio descriptions of artworks for the Museu d’Història de l’Escola.