Leveraging Knowledge Graphs and LLM Reasoning to Identify Operational Bottlenecks for Warehouse Planning Assistance

Rishi Yash Parekh, Saisubramaniam Gopalakrishnan, Zishan Ahmad, Anirudh Deodhar

Keywords: Warehouse Planning, Knowledge Graphs, Large Language Models, Agentic AI, Warehouse Digital Twin

TL;DR: KG and LLM agent supported Warehouse Planning Assistant for analysing simulation data

Abstract: Analyzing large, complex output datasets from Discrete Event Simulations (DES) of warehouse operations to identify bottlenecks and inefficiencies is a critical yet challenging task, often demanding significant manual effort or specialized analytical tools. We propose a novel framework that addresses this challenge through a powerful integration of Knowledge Graphs (KGs) and Large Language Model (LLM)-based agents. Our framework first transforms raw DES output data into a semantically rich KG, which uniquely captures the intricate relationships between simulation events and entities (e.g., suppliers, packages, workers, equipment), overcoming unstructured log limitations for robust analysis. On this KG, our novel LLM-based agent employs a sophisticated iterative reasoning mechanism that interprets complex natural language questions by generating insightful, interdependent sub-questions sequentially. Each sub-question is formulated one at a time, crucially conditioned on the evidence from answers to previous ones. For each individual sub-question, a multi-step process then generates precise Cypher queries for KG interaction, extracts relevant information, and performs crucial self-reflection to identify and correct potential errors. This adaptive, iterative, and self-correcting reasoning progressively pinpoints operational issues and diagnoses root causes, mimicking human investigative analysis. We evaluate our approach using an example warehouse DES setup, systematically introducing typical bottlenecks like equipment breakdowns and supplier arrival irregularities. For operational bottleneck questions, our proposed agent with its iterative reasoning framework demonstrates significantly higher pass rates compared to traditional baseline methods, achieving near-perfect performance in identifying key inefficiencies. Furthermore, for more complex investigative questions, we qualitatively showcase our framework's superior diagnostic capabilities through three case studies, highlighting its proficiency in uncovering subtle and interconnected inefficiencies often missed by traditional methods. This work attempts to bridge the gap between simulation modeling and advanced AI-driven data analysis informed by the broader advancements in KG+LLM, offering a more intuitive and potent method for extracting actionable insights from simulation outputs, thereby dramatically reducing time-to-insight and paving the way for automated, intelligent warehouse inefficiency evaluation and diagnosis for industrial data analysis.

Submission Number: 28