From Fraud Alert to Defensible Action: Building an Agentic Fraud Investigation System with TigerGraph

TigerGraph and Hacker House Goa developed SentinelGraph, an agentic fraud investigation system designed to transform suspicious alerts into auditable actions. By combining a temporal knowledge graph, GraphRAG, and deterministic policy controls, the AI agent analyzes transactions to determine if they are fraudulent. The system processed 590,742 transactions and 5,565 closed cases, achieving a 0.914 memory-model AUC. SentinelGraph identifies evidence, admits knowledge gaps, and recommends next-best actions for human sign-off to optimize fraud team efficiency.