S. Kulkarni, L. Toba
Idaho National Laboratory; Critical Materials Innovation Hub,
United States
Keywords: knowledge graphs, supply chain disruption, large language models, rag, retrieval augmented generation, artificial intelligence, supply chain
Summary:
The COVID-19 pandemic demonstrated the interconnectedness and complexity of global supply chains for goods. Subsequent events such as Suez Canal obstruction, wars (e.g. Russia Ukraine), and escalating tariffs have further emphasized the importance of supply chains for businesses and customers to control costs while maintaining reliability. These supply chain related issues are further stressed for critical materials. Therefore, the mounting supply chain complexity has reduced the effectiveness of ‘traditional’ supply chain analytics techniques. Current methods fall short in addressing (1) the immense and fragmented body of knowledge that is dispersed across numerous sources—or, in some areas, entirely lacking—and (2) the increasing complexity of supply chains and technological systems, which demand a multidisciplinary understanding that spans geopolitics, engineering, economics, and social behavior. Therefore, a tool that could assess risk from current events, competitive research, or market dynamics on a particular critical materials value chain segment would be of immense value. We propose a novel graph-augmented retrieval, with agentic framework that addresses these limitations through superior structural knowledge representation and causal reasoning capabilities. Unlike traditional semantic retrieval approaches that rely on vector embeddings and chunking strategies, our system constructs a dynamic knowledge graph where nodes represent discrete events gleaned from news articles, journal papers, market reports or other private documents and edges encode causal relationships between these events. This graph-based architecture enables multi-hop reasoning and relationship traversal that semantic similarity search alone cannot provide—critical for understanding cascading supply chain effects. This approach would allow the agentic framework to have relevant information to answer specific questions regarding supply chain disruptions. The entire framework is incorporated into a simple and intuitive, user-friendly interface that provides quantitative and qualitative insights. While focused on critical materials, the system and its architecture is adaptable to other supply chains, offering broad applicability for strategic planning, risk mitigation and informed decision making.