AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j
Neo4j proposes graph representations to give AI agents richer context from lakehouses than queries alone
“context coming in shapes and not necessarily queries”
Neo4j AI research engineer Zach Blumenfeld presented a workshop on augmenting AI agents with graph-based context built from both structured warehouse tables and unstructured documents in a lakehouse architecture. The core argument is that text-to-SQL and vector search give agents data access but not the relational shape of context needed to answer complex questions. The proposed solution is an agnostic graph data model that bridges structured and unstructured data to improve agent reasoning quality.