Vertical AI models: specialization or generalization?

22 October 2026 | 13.25h to 13:55h

The emergence of large general-purpose models has shaped the recent evolution of artificial intelligence, but interest in specialized vertical models is also growing. These models are developed for specific domains under principles of control, governance, and technological sovereignty. In this session, we will explore two examples that demonstrate how sovereign vertical AI models can deliver value in fields as important as biomedicine and the preservation of linguistic diversity in the digital environment.

Moderator:
TBC

Speakers:

  • Ray G. Butler, CEO and Director of Data Science, Butler Scientifics

  • Aleix Figueres Simó, AI lead, Fluendo

Presenter:

CIDAI | LinkedIn

JOAN MAS

Director, CIDAI

Joan Mas is a Telecommunications engineer (UPC) and MBA (Webster University). He worked for 10 years at the European Space Agency, the Netherlands, in the ENVISAT and METOP projects, Earth observation satellites. From 2000 to 2008 he developed at NTE S.A., equipment for the International Space Station and life support systems for manned space missions. Until 2024 he was Director of the Digital Technologies Division at Eurecat, promoting developments in areas such as Data Analytics and Big Data, Artificial Intelligence, Cybersecurity and others for multi-sector applications. He is currently Scientific Director in the digital field at Eurecat and, from 2021, he is also Director of CIDAI.

Speakers:

Butler Scientifics | LinkedIn

RAY G. BUTLER

CEO and Director of Data Science, Butler Scientifics

Ray G. Butler holds a degree in Computer Science from the University of Las Palmas de Gran Canaria and an Executive MBA from EADA Barcelona. He began his professional career at Abengoa, Giesecke & Devrient, and Panlab. In 2012, he founded Butler Scientifics, where he leads the development of BioConverse, a data exploration platform that helps research teams in hospitals, universities, and pharmaceutical companies gain a better understanding of diseases through their clinical data. Today, with a strategic focus on data science automation, he collaborates with leading medical centers while also teaching at IL3 – University of Barcelona.

BioConverse evolves: from conversations with data to agent-guided biomedical exploration

BioConverse showcases its latest advances as an agentic platform designed to transform biomedical data exploration. This session will demonstrate how the platform orchestrates the different components of its ecosystem to automate every stage of the data science workflow: defining objectives, formulating research questions, auditing data, integrating external information, exploring data with AutoDiscovery, interpreting results, and producing traceable conclusions. Discover how AI can go beyond chat interfaces to become a rigorous, practical, and controllable guide for biomedical research.

FLUENDO | LinkedIn

ALEIX FIGUERES SIMÓ

AI Manager, Fluendo

An aeronautical engineer with a master’s degree in Automation and Robotics from the Universitat Politècnica de Catalunya (UPC), he began his career in the field of unmanned aerial systems (UAS), founding a company focused on precision agriculture and aerial inspections. He later specialized in software development, primarily in computer vision, working in sectors such as automotive and embedded systems. In recent years, he has focused on artificial intelligence, leading teams and projects covering the entire AI lifecycle, from model training and evaluation to deployment in production and real-time environments.

IA·Veu: synthetic Catalan voices with safe and responsible AI

In this talk, we will present IA·Veu, a Catalan voice cloning project created to help reduce the gap in speech synthesis technologies for under-resourced languages. We will explain how we achieved high-quality synthetic voices using recordings from professional voice actors, the challenges specific to the Catalan language, and the limitations encountered in prosody and expressiveness. We will also share the project’s evolution towards generative models with control over intonation, rhythm, and pitch, as well as the results achieved through new training strategies and transfer learning techniques that enable the adaptation of new voices using significantly less data.