GraphRAG-Powered AI Agents: Transforming Corporate Metadata into Intelligent Knowledge

Session Track: Knowledge Graphs

Session Time: 07:15 am - 07:30 am November 6

Session description

In an era where AI governance demands comprehensive metadata management, enterprises struggle with fragmented data landscapes and semantic gaps between technical metadata and business understanding. This lightning talk demonstrates how GraphRAG-enabled AI agents revolutionize corporate data discovery and intelligence through metadata-driven insights. Loys Belleguie will present a practical framework combining Neo4j knowledge graphs with LLM-based agents to build enterprise knowledge bases from metadata assets. He'll showcase how multi-agent systems leverage graph structures to understand complex relationships between table definitions, column descriptions, data lineage, and contextual documents, creating intelligent proxies for understanding vast data landscapes without accessing the underlying data itself. Attendees will learn a methodology that transforms metadata lakes into intelligence engines. Rather than answering precise quantitative questions, the system excels at inference and prioritization: identifying which suppliers warrant sustainability audits based on their graph centrality and connections to high-risk materials, pinpointing critical data assets through lineage analysis, or discovering hidden dependencies in manufacturing processes through Bill of Materials metadata relationships. The framework answers questions like "Which data assets are most likely to contain sensitive information?" or "What tables should we prioritize for quality improvement based on their downstream impact?" Key takeaways include understanding why metadata is now mandatory for AI governance, implementing graph algorithms for metadata similarity and importance scoring, and designing agent workflows that enrich sparse technical metadata with business context. This approach enables intelligent decision-making about where to focus limited resources—whether for audits, data quality initiatives, or compliance efforts—all derived from the metadata's position and relationships within the knowledge graph.