Optimizing Reactor Plant Maintenance: The Case for Shipboard SLMs

Summary

LT P.J. Greenbaum and LT Vince Freschi discuss how shipboard Small Language Models (SLMs) combined with Retrieval‑Augmented Generation (RAG) can overhaul nuclear propulsion plant maintenance by turning static manuals and logs into a live, searchable knowledge base. They explain that small, air‑gapped models avoid the bandwidth and security issues of large LLMs while eliminating hallucinations, and that a continuously updated Reliable External Knowledge (REK) database can instantly surface relevant procedures, past failure reports, and sensor data. The episode outlines a practical rollout plan—including digitizing legacy manuals, building a vector database, defining hardware specs, and instituting a Human‑in‑the‑Loop certification—to empower technicians to diagnose and repair equipment faster and more accurately.