Novawerke AI was founded on a thesis the defense technology industry has been slow to accept: cloud-dependent AI is a strategic liability in contested environments.
The DoD has spent years adopting AI systems architected for commercial cloud environments — systems that assume persistent connectivity, reliable GPS, and low-latency network access. In peer-adversary contested environments, all three of those assumptions fail simultaneously.
Novawerke was founded to solve this from first principles. Not to retrofit cloud systems for the edge, but to build the inference architecture edge-native from day one — designed for the environment where the mission actually happens.
"The platform is not the hard problem. The cognition under constraint is the hard problem. We build the cognition."
Sean brings an unusual convergence of backgrounds to Novawerke's founding: operational intelligence experience as a Defense Intelligence Agency contractor, capital markets discipline from structured credit derivatives trading at Bear Stearns and Deutsche Bank, and graduate study in AI Engineering at Johns Hopkins University.
Born at Fort Bragg, North Carolina — the son of a Green Beret — Sean's understanding of what it means to operate in austere, contested environments without a reliable support infrastructure is not academic. It is generational.
That background shaped a thesis he began developing years ago: that the DoD's adoption of cloud-dependent AI represented the same category of systemic risk as the structured finance instruments he once traded — elegant models that worked perfectly until the assumptions they were built on failed all at once.
Co-founded Novawerke AI LLC, Virginia — edge-native AI for DDIL environments
Graduate study in AI Engineering — March 2026 panel on cloud AI as strategic vulnerability
Defense Intelligence Agency contractor — operational intelligence in contested environments
Frank holds an MS in Applied Artificial Intelligence from the University of San Diego and is the principal architect of Novawerke's Hub-and-Spoke Hierarchical Reasoning Model. His technical approach synthesizes Knowledge Distillation, Reinforcement Learning, and GFlowNet exploration into a unified training methodology that produces edge-deployable models capable of robust inference under severe computational and connectivity constraints.
Frank's core contribution is the KDRL architecture — a training framework that combines teacher-supervised knowledge distillation with multi-agent generative flow networks to produce autonomous systems that reason effectively from first deployment and improve through operational experience, without ever requiring cloud access.
His work on the airlocked star topology ensures that Novawerke's security posture is architectural rather than procedural — capability isolation enforced by design, not policy.
ODA 595 didn't wait for perfect conditions. They rode into Mazar-i-Sharif with the capability they had, in the environment they faced, and they completed the mission. Ancient platform. Most modern weapons available. No logistics tail. No guaranteed comms.
Every autonomous system Novawerke's architecture powers will carry that same principle: reason with what you have, where you are, when the network is gone.