Hybrid RAG-LLM Framework For Intelligent Supplier Risk Assessment In Global Supply Chains

Abstract

Traditional supplier risk scoring systems rely on structured data and predetermined rules, often failing to capture emerging threats embedded in unstructured sources such as news media, regulatory filings, and environmental disclosures. A hybrid framework combining Retrieval-Augmented Generation (RAG) with Large Language Models (LLMs) addresses this gap by retrieving contextually relevant documents and synthesizing evidence-based risk evaluations. The framework integrates enterprise resource planning data with external intelligence streams to assess geopolitical, financial, compliance, and sustainability risks across supplier networks. Real-time adaptation capabilities enable procurement organizations to respond dynamically to evolving threat landscapes. Each risk decision includes transparent provenance metadata, ensuring auditability and regulatory compliance. The framework addresses critical challenges, including algorithmic bias, interpretability requirements, and human oversight through structured governance protocols. By combining retrieval precision with generative reasoning capabilities, this system represents a transformative advancement in procurement risk mitigation, delivering proactive intelligence while maintaining enterprise-grade transparency and ethical deployment standards for responsible artificial intelligence integration in supply chain operations.