Paper Title
Scara: An Explainable Agentic AI Framework for Real-Time Supply Chain Fraud Detection Using Synthetic Simulation

Abstract
Fraud in buying and selling processes in both public and private sectors is a serious and costly problem, where there are many transactions and not enough supervision at different levels of organizations. This manuscript presents SCARA (Supply Chain Anomaly and Risk Assessment Agent), a smart system that uses fake data creation, various fraud simulations, location-based logistics modeling, and a large language model (LLM) to find and explain five main types of procurement fraud: invoice inflation, ghost shipments, product mismatches, product diversion, and shipment delay. The synthetic dataset was created using a reliable simulation process that included 50 suppliers, 20 buyers, and 100 types of products across ten cities, with location calculations done using the Haversine formula. A controlled 15% fraud injection rate adds marked anomalies based on real-world issues and suggests actions to investigate, like doing audits, checking supplier credentials, and applying stricter monitoring rules to reduce these types of fraud. An interactive Streamlit dashboard allows practitioners to explore and query transaction data in real time. The results show that SCARA gives clear, easy-to-understand, and useful fraud evaluations, creating a solid base for involving humans in AI decision-making for supply chain purchasing. Keywords - supply chain fraud detection, synthetic data generation, large language models, procurement analytics, anomaly detection, explainable AI, LLaMA, Groq inference, Streamlit, and agentic AI.