CGIP AI Assistant
Answers from the orders themselves.
Citation-backed inspection preparation — AI that answers from the directives, not from memory.
The challenge
Preparing for a Commanding General's Inspection Program (CGIP) evaluation means reconciling a unit's programs against a large, shifting body of official directives. Staff sections build checklists by hand, cross-referencing MCOs and NAVMCs line by line — slow, repetitive work where a missed update or stale citation carries real consequences for the unit's inspection results.
The approach
RIG built an AI-enabled preparation tool that generates dynamic, functional-area checklists with direct citations from the official directives — automated, modular, and designed from the start for DoD network realities.
- Source-grounded by design: outputs cite the governing MCO and NAVMC language directly, so every checklist item is verifiable against the original order — not a paraphrase the user has to trust.
- Open-source tech stack: a React single-page application (Material-UI, mobile-capable PWA) with LLM-driven natural-language processing accessed via API, and a serverless execution layer.
- Modular architecture: checklist generation is structured by functional area, so the tool extends to new inspection areas without a rebuild.
The outcome — and the lessons
The assistant was delivered with citation-backed checklist generation working end to end. Just as valuable is what the build proved about deploying AI tools inside DoD networks: restricted cloud environments and blocked serverless runtimes within AFPIMS shaped the architecture in ways no commercial deployment would surface. Those constraints — documented, not hidden — now inform every DoD-facing tool RIG builds, including the local-first, no-backend patterns used in our current in-development projects.