Combine GDS and Agentic AI for Automatic Report Generation

Session Track: Data Intelligence

Session Time: 08:30 am - 09:00 am November 6

Session description

Organized crime networks—from local fraud rings to complex international syndicates—operate through intricate webs of relationships that traditional analytical approaches often fail to uncover. In this session, the presenter will demonstrate how combining Neo4j's GDS library with agentic AI architectures creates a powerful platform for law enforcement intelligence analysis. The session will explore a complete workflow that transforms raw criminal data into actionable intelligence reports through two complementary approaches. First, attendees will learn how to leverage Neo4j GDS algorithms, including community detection (Louvain, Label Propagation), centrality analysis (PageRank, Betweenness), and other graph-specific techniques to identify criminal organizations and key players. Second, attendees will discover how to implement LangGraph-based agentic systems that automatically recognize temporal and geographical patterns as well as activity evolution, and generate comprehensive, human-readable reports from these analytical findings. The presenter will showcase two distinct agentic architectures: a controlled parallel workflow where specialized agents handle specific analytical tasks (group demographic analysis, temporal and geographical evolution, threat assessment, report generation) through coordinated but well-defined responsibilities, and a fully autonomous approach where agents dynamically determine their own analysis paths. Through live demonstrations using real co-offending network data, you will see how these systems can automatically identify threats and generate executive briefings that transform complex graph analytics into clear investigative guidance. By the end of this technical session, you will understand practical implementation details including Cypher queries for data modeling, GDS algorithm selection and tuning, LangGraph workflow orchestration, and prompt engineering strategies for generating accurate, contextually-aware reports. The session will also cover critical considerations for deploying such systems in sensitive law enforcement contexts, including data privacy, algorithmic transparency, and human-in-the-loop validation.