Enabling Grounded Answers Through Knowledge Graphs and Retrieval Augmented Generation

In modern defense manufacturing, achieving technological superiority hinges on both rapid decision-making and unparalleled precision engineering. Advanced machining systems, such as 5-axis CNC machines, play a pivotal role by enabling the production of intricate, free-form geometries with micron-level accuracy. However, these advances often necessitate deep domain expertise for optimal tool selection and machining parameter configuration. This paper introduces GraphLLM, a model-agnostic approach that integrates structured knowledge graphs with large language models (LLMs) to enhance the accuracy and reliability of technical responses. By automatically extracting domain-specific entities and relationships from documents, GraphLLM mitigates LLM hallucinations and improves performance, especially in technically challenging or out-of-distribution queries. Experimental evaluations across various LLaMA models demonstrate significant uplifts of 25%, highlighting the framework’s potential to provide grounded answers for decision-making in advanced manufacturing.