Zach Blumenfeld uses a fictional vehicle-repair workflow to connect warehouse tables with manuals and recall documents. A graph of schema relationships helps an agent choose SQL join paths, while links to source documents preserve the distinction between structured facts and supporting evidence.
Zach Blumenfeld builds document containment, sequence and cross-reference relationships, then demonstrates scoped traversal and full-text search. Structured parsing can capture these relationships deterministically; lexical search and query expansion complement navigation rather than proving that vector retrieval is unnecessary.
Zach Blumenfeld explores community detection as another context shape, using graph structure to discover document themes and useful source groupings. Resolution, connectivity and representation choices affect these groups, and richer labels can require additional interpretation.
Zach Blumenfeld combines these approaches in an agent demonstration, discussing data drift, security boundaries and the limits of the demonstration's retrieval strategy. The workshop provides implementation detail, not a controlled comparison of production-scale accuracy.
Watch on YouTube
