GraphSentinel Uses AI Agents and Graph Database to Automate Fraud Investigations

GraphSentinel is an agentic fraud investigation system built for the TigerGraph Agentic Fraud Investigation challenge, designed to go beyond simple transaction classification. The system begins from a risk alert or analyst request and produces a fully traceable investigation record that includes graph evidence, policy citations, recommended actions, and regulatory decisions such as SAR filings. It combines a temporal graph layer, a risk model trained on past investigations, policy-constrained decision-making, and GraphRAG retrieval of fraud patterns and regulatory references. A key design feature is value-of-information scoring, which makes the impact of each piece of gathered evidence visible at every step rather than bundling everything into a single final verdict. The workflow is implemented as an explicit LangGraph state machine, making each investigation stage independently testable and auditable.