Skip to content
Nexagen.
Insights

Technical perspective on applied AI inside accredited systems. No reposted vendor decks; real bylines from named engineers.

May 2026 · Applied AI

Applied AI

Self-hosting open-weight models on cleared hardware with vLLM and TurboQuant

For cleared environments the deployment story matters as much as the model. vLLM's V1 engine plus TurboQuant's calibration-free KV cache compression gets serious open-weight models onto on-prem GPUs without phoning home.

Read

May 2026 · Applied AI

Applied AI

MCP is the integration layer federal AI has been waiting for

The Model Context Protocol turns one-off API integrations into a standardized tool surface. For DoW systems with hundreds of internal services, that's a structural change.

Read

May 2026 · Systems Engineering

Systems Engineering

Million-token reasoning models change what RMF documentation looks like

A 1M-token context window lets a single reasoning pass load an entire ATO package, the control catalog, and the system telemetry, then answer questions across all three.

Read

May 2026 · Applied AI

Applied AI

Agentic RAG over operational doctrine, past the chat-with-PDF era

Naive RAG over doctrine PDFs gives confident wrong answers. Agentic RAG plans the retrieval, verifies the citation, and refuses to answer when the doctrine is silent.

Read

Apr 2026 · C5ISR

C5ISR

Running reasoning models on tactical hardware

Quantization and distillation have crossed a threshold. A 7B-parameter reasoning model fits in 5GB and runs at useful tokens-per-second on the kind of compute a tactical vehicle already carries.

Read

Apr 2026 · Cybersecurity

Cybersecurity

What an AI red-team report needs to look like before acceptance review

DoW acceptance review for an agentic system is structurally different from a static-system review. Here's the artifact set we've found actually works.

Read

Mar 2026 · Applied AI

SATCOM

Multi-agent orchestration for SATCOM link reconfiguration under contested RF

How we sequenced sensing, deconfliction, and reporting agents on a tactical SATCOM testbed, and why narrow agents composed beat one wide agent.

Read

Mar 2026 · Systems Engineering

Systems Engineering

Agent-driven RMF artifact generation: collapsing the ATO documentation cycle

ATO packages are document-shaped work. The artifacts are templated, the inputs come from systems we already have, and the human review steps are mostly verification, not authoring.

Read

Feb 2026 · Quality & Process

Applied AI

CMMI Level 3 as an AI evaluation moat

Most AI deployments fail not at the model layer but at the change-control layer. Process maturity is the unfair advantage in responsible-AI delivery.

Read