Retrieval-Augmented Generation (RAG) has shown significant potential in improving the performance of language models by incorporating external data into the generation process. However, traditional RAG systems often depend heavily on unstructured data obtained through standard retrieval methods lacking explicit relationships or structured connections between elements and, which can lead to problems in generating accurate information. Specifically, this paper aims to address these challenges by facilitating relationship mapping and improving the linking of entities, and examining how knowledge graphs can be integrated into current RAG approaches leading to more grounded content generation and minimize hallucinations. Thorough evaluation of our proposed research methodology shows that RAG systems such as sentence window and parent-child, when powered by knowledge graphs, can provide greater consistency and reliable results across key evaluators like correctness, relevance, and faithfulness.