A U.S. small business. A veteran-led team. Engineering roots in multi-party adversarial decisioning for kinetic defense — the original bytes-moving-atoms problem. Every line of the architecture was shaped in rooms where a machine’s recommendation ends a life and a human has to sign. That engineering lesson became the company.
CompositeApps didn’t start as an AI company. It started as a twenty-year argument about what it takes to make a machine’s recommendation something a human can sign for. Missile defense taught the question. Federal healthcare battle-hardened the answer at FedRAMP High. Generalizing that answer — to any domain where bytes have to become atoms defensibly — is what we do now.
Long Nguyen started on multi-party adversarial decision systems for missile defense — systems where the machine’s recommendation can end a life and a human has to sign for it. The engineering lesson from those rooms was narrow and permanent: the decision needs a witness, and the authority cannot be outsourced. Every control on the CompositeApps platform traces back to that lesson.
The Claims Genomics Model — CompositeApps’ first production deployment — sits inside federal healthcare because it is the most adversarial multi-party decision environment we could find. Patient, provider, payer, regulator, inspector general — five parties with conflicting incentives, all with standing to challenge any single call. Every determination the system ships carries its reasoning with it.
Today that engine runs at FedRAMP High under a VA National ATO at nation-scale. The architecture you see operating in the three loops on the homepage is the same architecture that went to Defense Cross-Domain review and the same architecture that will adjudicate a studio reshoot or a utility reroute tomorrow. Battle-hardened in the hardest adversarial environment, then generalized.
A composite material is not one substance — it is heterogeneous components bonded into something stronger and more resilient than any single constituent. Kevlar. Carbon fiber. Concrete with rebar. Each component is imperfect on its own. Bonded, they carry loads no single part could hold.
That is our architecture. No single model, no single vendor, no single judgment. We don’t build models — we compose applications of specialized AI agents, bonded into one governed workflow under your authority. Swap a model underneath; the workflow and the evidence chain stay intact. Your intelligence is the asset. Everything else is interchangeable.
From an adversarial-decision engine built for the worst-case missile-defense scenario, to an adjudication platform running today under a VA National ATO, to the general-purpose composite layer for industry AI. Same architecture. Different missions.
Founder origin in multi-party adversarial systems for kinetic defense — decisions where an output has to survive its strongest counter-case before it advances. This is the architectural DNA of everything CompositeApps has built since.
The Claims Genomics Model is deployed into U.S. federal healthcare claims adjudication — a live, adversarial, multi-party environment where every determination must survive audit by regulators, payers, providers, and a Federal Inspector General.
CompositeApps achieves FedRAMP High authorization. The affiliated platform stands up a VA National ATO ahead of schedule. 80M+ claims processed. $380M of structural waste surfaced with per-dollar provenance.
Five active controls — Behavior Monitor, Provenance Ledger, Glass-Box, Adversarial Triad, Authority-Scoped Fleet — are codified as the Active AI Risk Management Framework. Defense Health Agency evaluation (AOI #06) begins.
Active AI-RMF is positioned as a model-agnostic composite layer between industry AI and sovereign decisions — built for the Army, Navy, federal civilian agencies, the intelligence community, and regulated enterprise. The architecture returns to the domain it was conceived for.
Small, deep, fielded in federal program environments. Missile-defense engineering, decades of federal program execution, and time in uniform — so the distance between “what’s possible” and “what you can defend” is one we have walked before.
Engineer and architect. Origin in multi-party adversarial-decision systems for missile defense. Architect of the Claims Genomics Model — the adversarial adjudication engine running today at FedRAMP High under a VA National ATO. Sovereign systems is the only thing Long has ever built.
Retired U.S. Army Lieutenant Colonel. Decades of field and headquarters experience in C2, Cross-Domain Solutions, and federal IT program execution. Translates operator reality into architectural requirements; carries the conversations that matter with HQDA, the service components, and the NCDSMO.
Federal strategy, teaming constructs, and the integrator relationships that make sovereign AI adoptable inside existing acquisition pathways. Anchors the go-to-market so adoption fits Army and federal mechanisms, not around them.
Every value on this page is enforced by the architecture itself — not by a policy PDF or a culture deck. We hold them because the only way to keep them is to build them in.
Your data, your evidence, your authority. Our runtime comes to your boundary; it does not pull you to ours. Zero egress is a measurement, not a marketing line.
Every consequential determination ships with its regulatory basis, factor weights, and confidence — signed, time-stamped, and on your hash chain. A reviewer verifies reasoning, not just answers.
The AO’s confidence updates on every evidence packet. Drift, anomaly, or adversarial signal scopes down authority automatically. No ticket. No cycle. The framework defends the posture itself.
We do not lock you into a foundation model. Open-weight, fine-tuned, or commercial-under-license — any model plugs in under the same controls. The governor is the constant.
The Adversarial Triad makes single-perspective determinations architecturally impossible. Every advancing output has already lost to its strongest counter-case before it leaves the runtime.
U.S. small business. U.S.-sovereign data. U.S.-citizen workforce. Zero foreign components. Every line of code is eligible for the workloads we serve.
We work with Authorizing Officials, mission owners, program managers, and the integrators who serve them. Come with a workload; we’ll come with a framework and a one-page work statement within a week.