AI and Public Service at the Data Technology Seminar 2025
The most innovative and effective technologies for operators and professionals in the radio and television domain, with a special focus on transparency and fairness, ranging from Recommender Systems, news verification, generative AI and LLM: this is the Data Technology Seminar (DTS), held at the EBU in Geneva from 11 to 13 March 2025.
The DTS is at its third edition, and is enjoying growing success and interest, as demonstrated by this year’s approximately 200 participants , coming from all over the world. The event program is available to everyone on the website DTS 2025, while access to the individual presentations is reserved to EBU members.
EBU DTS 2025
The Rai Research Centre (CRITS) contributed substantially to the success of the event through the participation of Paolo Casagranda in the Program Committee, the presentation of three works by Maurizio Montagnuolo, Alberto Ciprian, Stefano Scotta and Lorenzo Canale, and the moderation of four sessions by Alberto Messina and Paolo Casagranda.
Specifically, Maurizio Montagnuolo presented “Revolutionizing video content analysis: unleashing the power of AI Microservices”, illustrating a microservices software architecture based on Artificial Intelligence for the analysis and metadata extraction of multimedia content. Thanks to this architecture, individual analysis functions can be added, updated or replaced without affecting other services, offering several advantages, including scalability, flexibility and adaptability of the system to user needs.
Alberto Ciprian presented “Generative AI in action”, a project created for the IBC Accelerator Program 2024. The aim of the project is to demonstrate how Generative AI can be inserted into production processes. Particularly, a pilot episode of a hypothetical animated TV series was created using Generative AI tools for both the Scriptwriting and the graphics part. The result was appreciated by industry insiders and demonstrated how it can be a valid tool to speed up some phases and support creatives in the classic pipeline.
Stefano Scotta and Lorenzo Canale presented “LLM based framework to evaluate RAG systems”, a method to optimize the choice of components in a Retrieval-Augmented Generation (RAG) system. Given a dataset, thanks to an automatic question generation module and a process of selecting the most relevant contents, the framework evaluates both the accuracy of the answers and the quality of the information retrieved based on the chosen components, facilitating the identification of the most effective solutions to improve the overall performance of the system.
