The GRAD project aims to develop an innovative architecture designed to overcome the limitations faced by traditional RAG (Retrieval-Augmented Generation) systems in managing complex data. While current RAG methodologies rely primarily on simple vector proximity search—often missing broader contexts and deeper correlations between sources and entities—GRAD combines structured knowledge graphs with advanced data discovery mechanisms. In doing so, it maps indirect relationships across heterogeneous sources, revealing hidden information patterns. The research aims to standardize this architecture to create a reliable retrieval system, ideally suited for managing corporate and scientific knowledge.