Retrieval-Augmented Large Language Models for Real-Time Supply Chain Disruption Intelligence and Decision Support

Abstract

Large language models offer unprecedented analytical capability, but their knowledge is frozen at the last training date — rendering them unusable for organizations whose mission depends on emerging, timely information [1]. This paper argues that retrieval-augmented generation is the missing mechanism that converts LLMs from static reasoners into realtime supply chain disruption intelligence systems: retrieval supplies the current evidence, grounding supplies the facts, and agentic orchestration supplies the decision loop. We propose RAGENT-SC, a retrieval-augmented agentic framework coupling (i) a continuously updated multi-format knowledge layer (news streams, contracts, supplier records, operational KPIs, and a supply network knowledge graph), (ii) hybrid vector–graph retrieval with sublinear scalability, (iii) grounded generation with provenance metadata, (iv) agentic orchestration of monitoring, analysis, planning, and audit agents, and (v) a governance layer with hallucination verification and human-in-the-loop escalation.