Mike Phipps

Mike Phipps is a lead AI engineer at the Gates Foundation working on its Strategic Intelligence Platform, which gives AI systems access to the foundation’s operational knowledge. His defining conviction: enterprise AI becomes durable when it understands an organization’s data, relationships, and institutional practices—not when it owns the latest model or chat interface.

From particle physics to enterprise AI

Phipps began his technical career in experimental physics, contributing to particle-detector software and beam-test documentation at Florida Institute of Technology. He later joined the University of Illinois Urbana-Champaign’s ATLAS heavy-ion research program at CERN, participated in detector-development work, and received the university’s 2017 Maurice Goldhaber Research Scholar Award in Nuclear Physics.

At the Gates Foundation, he applies a similar systems-level approach to institutional data. The Strategic Intelligence Platform integrates operational records and documents within a cross-system semantic knowledge graph built on Neo4j, allowing AI agents to navigate funding, investments, management structures, meetings, and organizational hierarchies.

• Institutional knowledge is the enterprise AI moat. Effective AI depends on field definitions, reporting conventions, access restrictions, and procedural knowledge held by data owners—context that cannot be recovered reliably from disconnected records alone.

• Different organizational relationships require different models. Funding structures form a directed acyclic graph, while investment-management hierarchies need derived relationships that capture indirect responsibility. Shared entities link these structures with personnel records and documents.

• Governance belongs inside the data pipeline. Semantic document sections and structured metadata are paired with sensitive-data classification, user-specific access controls, and personally identifiable information masking.

• Retrieval evaluation should follow live institutional data. Phipps compares agent responses against current graph queries, measuring first-answer accuracy and consistency; failures inform improvements to schemas, domain rules, and the data model.

Through the Model Context Protocol, the platform delivers this organizational context inside existing tools such as Claude. Phipps is also developing more constrained agent workflows and connections between the central enterprise graph and specialized team datasets. His AI Engineer World’s Fair presentation outlines the architecture.