Mike Phipps presents semantic infrastructure that connects a curated lakehouse and Neo4j graph to existing AI clients through MCP. The design makes business meanings, joins and reporting standards explicit instead of expecting a language model to reconstruct them.
Mike Phipps discusses preparation work including chunking, deduplication, descriptions of figures, sensitive-data masking and entitlements. Funding relationships and management hierarchies need different structures, while shared entities can connect otherwise separate information silos.
Mike Phipps emphasizes evaluating query accuracy and stability with rules, human judgment and model-based checks. His presentation frames governed organizational knowledge as reusable infrastructure and discusses future federation rather than claiming that graph storage makes answers inherently correct.
Watch on YouTube
