EBU Neo - a sophisticated multilingual chatbot for a trusted news ecosystem exploration

Abstract

In recent years, Generative AI and Large Language Models (LLMs) have transformed user interaction with digital content, including news. However, concerns remain about the quality and transparency of information provided by commercial chatbots, which often rely on opaque source selection influenced by commercial interests. To address this, we introduce Neo, a news chatbot that delivers reliable and transparent information using content from Public Service Media (PSM) organizations. By leveraging diverse, multilingual sources from European Broadcasting Union (EBU) members, known for their strict editorial standards, Neo provides users with accurate and up-to-date responses. This paper presents the design of Neo, outlining key scientific contributions such as the optimization of a modular Retrieval-Augmented Generation (RAG) pipeline, advanced semantic indexing methods, and support for multilingual input. We also detail the technical implementation, evaluate cost considerations, and share insights from early user trials. Finally, we outline directions for future development and planned enhancements.