Design and implementation of a local LLM-driven decision support system for AI-enhanced supply chain management

This paper presents a large language model (LLM) decision support system for supply chain management (SCM). Building on a multi-layer architecture originally developed for intelligent manufacturing planning, the proposed artificial intelligence (Al) system is re-engineered to address end-to-end SCM tasks including supplier evaluation, inventory planning, production logistics, and distribution planning. A 14-billion-parameter LLM is deployed on-premise via the Ollama framework, allowing natural-language interaction while ensuring that sensitive SCM data remain within the local computing environment. The system integrates multiple knowledge sources – formal SCM theory, standardised process maps, structured supplier and inventory databases, and empirically calibrated forecasting models – into a single, query-driven decision pipeline. A classification module distinguishes theoretical, analytical, and procedural queries and routes them to appropriate deterministic models and knowledge bases before the LLM generates explanations. Through automated context construction and domain-restricted prompting, the AI agent produces technically validated, context-aware responses suitable for interactive planning. Experimental deployments on realistic SCM scenarios indicate that the system can provide rapid supplier scoring, inventory parameter tuning, DRP-based distribution planning, and policy comparisons with transparent, database-grounded justifications. The results demonstrate that combining local LLMs with classical decision models offers a practical path toward trustworthy, privacy-preserving AI decision support for both industrial SCM environments.