The Barcelona Supercomputing Center is developing Public AI Infrastructure focused on language technologies including text processing, speech recognition and machine translation. This infrastructure provides open and reusable AI models that can be adopted by different ecosystems, from public institutions to small and medium-sized companies. The models are trained using the MareNostrum 5 supercomputer and built on the expertise of the Language Technologies Lab within the BSC AI Institute.

Problem

Most major language models are built and trained primarily in English, giving  little support to  Spanish co-official languages. This causes challenges for public services, businesses and startups who require solutions that are compatible with local languages, cultural contexts, and legal requirements. In addition, many organitzations lack access to affordable AI models adapted to their needs that can be deployed locally.

Solution

The Public AI Infrastructure develops large language models (LLMS) specifically trained with carefully curated data in Spanish and co-official languages. The training process places strong emphasis on reducing bias, addressing ethical concerns and ensuring proper representation of social and cultural values. These models can be integrated into existing systems and deployed either locally or online, enabling broader and more responsible adoption of language AI across different sectors.

alia.gob.es / projecteaina.cat / proyecto-ilenia.es