AI Infrastructure for a Defense Contractor Preparing for CMMC 2.0
Frontier-grade AI productivity across 20 years of business-development content — without a single controlled document, or its derived index, leaving the perimeter.
Defense contractor. Name redacted at their request.
The Problem
A defense contractor needed to bring AI into business development — not engineering. Contract negotiation, marketing content, and 20 years of institutional knowledge sat in a proprietary CRM they couldn't risk exposing to third-party APIs. They were mid-preparation for CMMC 2.0, meaning third-party auditors — not internal self-attestation — would soon verify every control. Anything built had to be audit-ready from day one, not retrofitted later.
Internally, they lacked the GPU/inference expertise, Kubernetes fluency, and LLM tooling knowledge to scope, let alone build, a compliant solution. Prior consultants offered strategy decks, not deployable infrastructure.
The Solution
DPLYD deployed a fully managed, on-prem AI appliance — inference and infrastructure-as-a-service, repeatable and consistent, running inside the contractor's perimeter. The core engineering:
Bridged Linux ↔ Windows environments
Custom services spanning SMB shares so inference could reach content wherever it lived.
Inherited access control
Existing document-level ACLs were ingested and enforced at the inference layer, so search results always honored pre-existing security profiles.
Crawl-walk-run rollout
Phased deployment so the organization adopted capability at a pace matched to its own operational maturity.
The Outcome
The contractor got frontier-grade AI productivity without sending a single document — or its derived index — outside their boundary. Security posture was inherited, not bolted on, keeping them CMMC-ready throughout rather than requiring a rebuild before audit.
Security posture was inherited, not bolted on — CMMC-ready throughout, not rebuilt before audit.
Why This Matters Broadly
Defense contractors face a shared bind: real AI value sits in unstructured, decades-deep, proprietary content — but frontier models and cloud indexing are structurally incompatible with controlled environments. The fix isn't a thinner AI tool; it's moving inference and the index inside the perimeter, inheriting the access controls and audit posture the organization already built. That's the model DPLYD now productizes for the sector.