Heterogeneous plies — carbon, glass, resin — bonded into a structure stronger than any constituent. That's the engineering idea this company is named after, and we mean it literally: frontier models are the plies. The arbitration layer is the resin.
A single model is a single point of failure: one vendor, one bias, one black box. A composite argues with itself — one ply proposes, another attacks, the resin reconciles — and the result carries load no single layer could.
"Heterogeneous parts, bonded into something stronger than any constituent. That's how trust gets built."
We didn't learn that from a whitepaper.
We started in multi-party adversarial decision systems for missile defense — engagement recommendations fused from radar, IR, and SIGINT, where the machine proposes and a human signs for the consequence. Long before "human-in-the-loop" was a slide, it was our acceptance criterion.
U.S. ARMY · DVIDSThe architecture went to production in the most adversarial multi-party decision environment in U.S. federal healthcare: 180 million claims a year, five parties with opposing incentives, every determination FOIA-survivable — at FedRAMP High. Glass-Box appeals outperform the industry baseline 72:38.
U.S. ARMY · DVIDSClaims, fire control, grid operations — the loop is identical: signals in, an argument in the open, a human signature out. So we generalized it: CompositeApps, the sovereign layer where bytes become atoms — running inside your boundary, on your chain, under your authority.
U.S. ARMY · DVIDSHere's what one decision actually looks like inside.
{
"artifact": "RA-0141",
"determination": "ACCEPT_WITH_COMPENSATING_CONTROLS",
"advocate": { "proposal": "sentry monitor · VLAN-12 span", "conf": 0.91 },
"defender": { "objection": "east-west modbus uncovered",
"demanded": ["CM-8(3)", "IR-4"] },
"arbiter": { "basis": ["CM-8(3)", "IR-4"], "scope_days": 30,
"evidence_cadence_min": 15 },
"chain": { "prev": "0x7a31…c9e2", "hash": "0x9e44…b1a7" },
"signer": "CDR · CAC/PIV · SAML2",
"egress_bytes": 0, "latency_ms": 412 // pre-decision
}
And the whole thing runs inside a boundary you already trust.

Engineer and architect. Origin in multi-party adversarial-decision systems for missile defense. Architect of the Claims Genomics Model — the adjudication engine running at FedRAMP High. Sovereign systems is the only thing he has ever built.

Retired U.S. Army Lieutenant Colonel. Decades across C2, Cross-Domain Solutions, and federal IT program execution. Translates operator reality into architectural requirements — and carries the conversations 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 — so adoption fits Army and federal mechanisms, not around them.
No autonomous action without a recorded argument. The transcript is not overhead — it is the deliverable.
Vendors can supply intelligence. They cannot supply accountability. The signature stays with your people, on your chain.
A decision you can't reconstruct is a liability with good UX. We ship the reconstruction with the decision.
Enough reading. Watch it run.