Session Overview
Querying enterprise data lakes with hundreds or thousands of tables often breaks traditional text-to-SQL workflows. As context grows, LLMs struggle with table discovery, multi-hop reasoning, and domain semantics.
In this talk, learn how PuppyGraph bridges the gap by turning existing Apache Iceberg tables into an ontology knowledge graph without data movement. Discover how combining fast graph queries with streaming architectures enables autonomous AI agents to analyze complex enterprise relationships—like supply chain risks and disruption exposure—with full accuracy and auditability.
What You'll Learn
- The core pitfalls of raw text-to-SQL when scaling to thousands of tables
- How to automatically build an ontology graph schema from enterprise catalogs
- Querying existing Apache Iceberg tables directly without migrating to a separate graph database
- Live demo: Generating graph ontologies and uncovering semiconductor supply chain risk
- Verification and auditing: Grounding LLM insights with transparent, executable graph queries
Chapters
- 00:00 — Enterprise data challenges: Moving beyond basic text-to-SQL
- 01:21 — Context loss, wrong table joins, and multi-hop reasoning limits
- 02:27 — System architecture: PuppyGraph MCP and fast sub-second graph queries
- 03:06 — Zero-copy graph queries on Apache Iceberg data lakes
- 04:04 — Understanding ontologies vs. live knowledge graphs
- 05:11 — Automated graph generation & human-in-the-loop validation
- 06:33 — Querying knowledge graphs: Scoping domains and mapping business logic
- 07:54 — Multi-hop query examples for supply chain dependency
- 09:10 — Demo: Generating an ontology from 23 Iceberg supply chain tables
- 10:28 — Demo: Automated supply chain risk analysis & slide generation
- 11:42 — Evaluating risk reports: Quantifying disruption exposure
- 12:42 — Verifying agent queries, auditability, and data reliability
