Adding Retrieval Augmented Generation to the MOSAIC Framework

This paper presents a concept for adding Retrieval-Augmented Generation (RAG) features to the MOSAICframework. MOSAIC enables web search in segments ofthe Open Web Index (OWI), in order to establish a special-purpose search engine. An extension, MOSAIC-RAG, hasbeen developed that adopts a RAG approach. It is designedas a modular framework that has integrated a set of process-ing modules built on generative AI models, such as a modulefor re-ranking the search result, a module for summarisingthe full texts of the search result, or a module for summaris-ing all search results. These modules can be ordered in anarbitrary sequence, in order to configure an overall processto improve the search result. Such configurations can beadapted for specific purposes and saved for later reuse.