Knowledge graphs (KGs) are essential for Retrieval-Augmented Generation (RAG), allowing AI systems to tap into external knowledge for enhanced reasoning. However, traditional methods for extracting KGs from text, particularly using large language models (LLMs), are computationally expensive and can struggle with information loss when processing long documents. Researchers have introduced a new two-pronged approach called Distill-SynthKG to tackle these issues. First, a multi-step process called SynthKG carefully analyzes documents, breaking them down into smaller chunks and resolving references to maintain context. This ensures more accurate and comprehensive extraction of entities, relationships, and the context describing them (called propositions), forming a high-quality KG. Second, the knowledge embedded within SynthKG is then 'distilled' into a smaller, specialized LLM. This smaller model can generate KGs directly from entire documents in a single step, dramatically reducing the computational cost compared to repeatedly querying a larger LLM. Tests showed that Distill-SynthKG outperformed larger, less efficient models in accurately representing knowledge, improving information retrieval, and enhancing multi-hop question answering—tasks that require connecting information from multiple sentences or sources. This innovative method offers a more efficient way to build knowledge-rich AI systems, paving the way for more scalable and effective applications of KGs in areas like question answering and intelligent agent development.