About me
I am an AI engineer and researcher based in Lausanne. I completed my PhD in Computer Science at the University of A Coruña in June 2026, and was a Visiting Researcher at LMU Munich and EPFL. Before that, I studied Physics (BSc, Complutense University of Madrid) and Language Technologies (MSc, UNED).
Most of my research is about evaluating language models: how they handle perturbed input, dialects, and languages they rarely saw in training, and how LLM-generated text differs from human writing.
I am looking for roles in Switzerland or remote. Available immediately; EU citizen.
Projects
2026
-
jobfacts: job-posting pipeline with measured fact extraction
Collects AI and data job postings daily from about 170 employer job boards into PostgreSQL. A classifier model fills a fixed schema per posting (work mode, remote from Switzerland, language level, seniority), each value with the sentence from the posting that supports it, so every fact can be checked. On a held-out set of 50 postings in English, German, and French, the extracted fields are 90–100% correct. Built with coding agents under my direction.
-
Reading Cross-Lingual Semantic Differences from Apertus
Scores each word of parallel Swiss government web pages (English vs. German, French, and Italian) for how much its meaning differs from the other version, using Apertus 8B as a frozen reader. Middle-layer hidden-state alignment and contrastive surprisal, combined by a logistic regression head, reach 37.4 Spearman on the held-out test, vs. 17.1 for the best released baseline.
Publications
Selected
-
Contrasting Linguistic Patterns in Human and LLM-Generated News Text
Artificial Intelligence Review, 2024. 267 citations; cited in The Guardian.
Compares human news text with the output of six LLMs. Negative emotions rose with model size, and five of the six models amplified the sexist bias found in human text.
-
Evaluating Pixel Language Models on Non-Standardized Languages
COLING 2025.
Pretrains a tokenizer-free pixel language model on German and tests transfer to Swiss German, Bavarian, Alsatian, and Low Saxon. It beat token-based models on POS tagging, parsing, and intent detection, by up to 26 points in some settings.
2025
-
Nested Named Entity Recognition as Single-Pass Sequence Labeling
Findings of EMNLP 2025, Suzhou, China.
-
Evaluating Pixel Language Models on Non-Standardized Languages
COLING 2025, Abu Dhabi, UAE.
-
On the Impact of Syntactic Infusion for Gender Categorization Across Contextual Dimensions
Revista de Procesamiento de Lenguaje Natural, 2025.
2024
-
Contrasting Linguistic Patterns in Human and LLM-Generated News Text
Artificial Intelligence Review, 2024.
-
Demographic Background Prompting Does Not Affect Linguistic Features on LLM-Generated News Texts
XoveTIC 2024.
2023
-
Assessment of Pre-Trained Models Across Languages and Grammars
IJCNLP-AACL 2023, Bali, Indonesia.
-
LyS A Coruña at GUA-SPA@IberLEF2023: Multi-Task Learning with Large Language Model Encoders for Guarani-Spanish Code Switching Analysis
IberLEF 2023, Jaén, Spain.
-
Another Dead End for Morphological Tags? Perturbed Inputs and Parsing
Findings of ACL 2023, Toronto, Canada.
2022
-
Parsing Linearizations Appreciate PoS Tags – but Some Are Fussy about Errors
IJCNLP-AACL 2022, Online.
-
Cross-lingual Inflection as a Data Augmentation Method
3rd Workshop on Insights from Negative Results in NLP, Dublin, Ireland.
2021
-
Not All Linearizations Are Equally Data-Hungry in Sequence Labeling Parsing
RANLP 2021, Online.
Education
-
Doctorate (PhD) in Computer Science. University of A Coruña, 2021–2026.
Thesis: "Encoding, Representation, and Interpretation of Linguistic Structure Across NLP Paradigms". Thesis record.
-
MSc in Language Technologies. Universidad Nacional de Educación a Distancia (UNED), 2020–2021.
-
BSc in Physics. Complutense University of Madrid, 2014–2019.