AI EngineerVideoSource Linked
AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j
Graph-shaped context can expose relationships and document structure that vector search or Text-to-SQL misses. Treat it as an additional retrieval surface, then benchmark it against your own data.
AI Engineer · Jul 23, 2026
Source Summary
The workshop builds graph representations over lakehouse tables and documents, letting an agent inspect metadata, traverse linked document hierarchies, and detect document communities. Tool calls are exposed through **MCP history** in the demonstrated setup.
Practical Implication
Give data agents several context shapes rather than one universal query path. Use a metadata graph to plan SQL, linked views for document neighborhoods, and **community detection** when thematic grouping helps investigation.
Agent-Ready Context
The workshop builds graph representations over lakehouse tables and documents, letting an agent inspect metadata, traverse linked document hierarchies, and detect document communities. Tool calls are exposed through **MCP history** in the demonstrated setup. Give data agents several context shapes rather than one universal query path. Use a metadata graph to plan SQL, linked views for document neighborhoods, and **community detection** when thematic grouping helps investigation. The workshop reports **no accuracy benchmark** for the demonstrated system. Its structured and unstructured graphs are also unlinked, leaving the agent to join extracted identifiers at runtime unless builders add deterministic connections.
Context Map
contextdata#context-engineering#retrieval#mcpUncertainty
The workshop reports **no accuracy benchmark** for the demonstrated system. Its structured and unstructured graphs are also unlinked, leaving the agent to join extracted identifiers at runtime unless builders add deterministic connections.