Team Lead for Evaluation and Annotation at the Language Technologies Unit (BSC). His and his group’s work is aimed at creating a comprehensive and efficient evaluation setup for multilingual large language models, generating high-quality evaluation multilingual datasets, and improving the human side of model evaluation, among others. Javier's current research lines and interests are connected with benchmarking efficiency (in collaboration with Hewlett Packard Enterprise), improving reliability and validity of LLM-as-a-judge, and evaluation methods for the biomedical and health domains (in collaboration with BSC's Social Link Analytics Unit, IBM Research Europe, and CERTH). He is also a core member in the Skill, Safety and Trust Evaluation working group of the Trillion Parameter Consortium, and a member of the Trust & Safety and Open Foundation Models & Datasets working groups of the AI Alliance.

He holds a PhD in Psycholinguistics, an MSc in Natural Language Processing and an MSC in Language Education. He has taught and lectured in graduate and postgraduate courses in universities such as The University of Edinburgh and Universidad de Zaragoza. He has also worked as external quality assessor for Madri+d and AQU Catalunya, and co-chaired and delivered invited lectures at multiple international events and conferences.