In this talk, we'll examine the synergy between knowledge graphs, semantic search and retrieval-augmented generation. We will investigate how integrating an automatically generated knowledge graph derived from a corpus of documents enhances RAG capabilities. Specifically, we will look at how this integration can overcome common limitations of standard RAG approaches, enabling the extraction of "global" trends and insights or those that stem from combining pieces of information from several documents (i.e. ones that are impossible to answer by simply injecting information from a number of unconnected retrieved documents). As tools we will use Neo4j graph database as the knowledge graph combined with vector indices, Langchain for business logic flow.