D. Brian Letort, Ph.D.

I lead enterprise AI programs at global scale. I build governed platforms, operating models, and data foundations. I publish research on the systems layer that makes enterprise AI reliable. I teach — through Pluralsight and graduate faculty appointments — the next generation of engineers and architects doing this work.

I operate from a simple premise: the AI platforms that earn enterprise trust are the ones governed from day one, and those are the ones used at the top of the house. Everything else eventually stalls inside legal review. That premise shapes how I architect, how I write, and how I advise.

At Northrop Grumman, across a 20-year tenure, I served as Northrop Grumman Fellow, Chief Enterprise Architect, and Chief Data Scientist. I aligned a $1.5B IT portfolio to mission architecture, contributed to $13B in proposal architectures, trained 800+ employees in machine learning, and delivered $4.5M+ in annual automation savings. I hold a Top Secret clearance from that era. The posture I built there — careful, auditable, mission-first — is the operating default I bring to every program I touch.

In my current role, I lead a portfolio of five concurrent enterprise transformation programs at a publicly-traded global infrastructure operator: a $30M annual budget, 120+ matrixed contributors, enterprise AI strategy and governance, platform delivery, Master Data Management, and enterprise reporting across financial and operational domains. The work engages the entire C-suite and the CEO is an active user of the governed AI platform my organization operates.

Across my career I have taught 5,000+ students through graduate and doctorate coursework in computer science, data science, machine learning, and enterprise AI. That teaching discipline is where I formed the conviction that the clarity of the teacher is the clarity of the architect.

My research program asks a practical question: why is enterprise AI so unreliable, and which layer of the stack is missing?

The answer I am writing toward, across four published papers on Zenodo, is Context Compilation Theory — the discipline of treating the assembly of context for a model the way software engineering treats the compilation of source code. Retrieval tells you what to read. Reasoning tells you what to say. Context compilation is the governed, measurable, optimizable layer between them. It is where reliability lives, where provenance is enforced, and where token economics becomes a design surface rather than an accounting line.

This is not an academic side project. It is the through-line of the work. Every platform I build is an implementation of the theory. Every outcome I publish is evidence for it.

Read the research →