AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j

Integrating graph intelligence into lakehouse architectures enhances AI agent reasoning by providing structured context across disparate data sources. A semantic layer approach, utilizing Neo4j, creates an agnostic data model that bridges structured warehouse tables with unstructured document repositories. Three primary graph shapes—connections, hierarchical tables of contents, and community-based themes—enable agents to navigate complex data relationships, perform estate-level analysis, and identify documentation gaps. By moving beyond simple vector or lexical search, this method allows for deterministic data loading and efficient traversal, significantly reducing agent hallucinations. This framework empowers developers to build copilots that reason about data holistically, ensuring accurate repair procedures and identifying patterns in large-scale datasets that traditional semantic search methods often fail to capture.