DYNAMIC DECISION SUPPORT SYSTEMS: CUSTOMIZING RAG KNOWLEDGE BASES FOR SPECIFIC MARITIME MISSIONS

Abstract

Maritime operations of long-haul transoceanic voyages frequently face a connectivity desert, where leveraging cloud-based Artificial Intelligence is prohibited due to the absence or high expenses of satellite internet. This paper discusses a new Mission- and Polar-Specific Offline Retrieval-Augmented Generation framework that aims to provide crews with high-fidelity, localized decision support. Unlike generic AI models, the proposed system will be dynamically ingesting a curated corpus comprising vessel-specific technical manuals, cargo-specific Material Safety Data Sheets, and route-specific maritime regulations. The architecture leverages Edge AI by deploying quantized Large Language Models and vector databases on ruggedized onboard hardware, ensuring sub-second inference latency without any external connectivity. To solve the challenge of data freshness during long isolation periods, we have proposed a Hybrid Synchronization Mechanism that enables "delta-vector" updates-transmitting only critical new embeddings through low-bandwidth satellite links to keep local knowledge bases current with minimum data overhead. We show evidence that this framework significantly reduces the cognitive load on officers and engineers and increases operational safety and autonomy. This study concludes that mission-tailored RAG systems are a crucial step toward the digital transformation of resilient and autonomous maritime transport.