<![CDATA[Maritime weather information is critical for shipping safety in archipelagic countries such as Indonesia. However, current dissemination methods at the Teluk Bayur Maritime Meteorological Station rely on one-way infographic distribution through social media, lacking interactive communication capabilities. This study presents a lightweight Retrieval-Augmented Generation (RAG)-based chatbot that integrates Google Sheets as a dynamic knowledge base for real-time maritime weather dissemination through a conversational interface. The system was implemented using Python and Streamlit within a four-layer monolithic architecture and covers six information categories, including early warnings, three-day weather summaries, and vessel safety recommendations across nine maritime regions in West Sumatra for four vessel types. A rule-based decision support module classifies sailing conditions as safe (aman) or cautionary (waspada) based on wind speed and wave height thresholds defined by Meteorological, Climatological and Geophysics Agency of Indonesian (BMKG). Functional testing confirmed accurate responses across all menu features and free-form queries. Quantitative evaluation using 50 queries, scored on a 0–1 scale using ROUGE-L and TF-IDF cosine similarity, showed that the RAG-enabled model achieved an overall mean score of 0.2285, compared to 0.1425 for the non-RAG baseline (+0.0860) demonstrating that structured context grounding effectively reduces hallucination and enhances information reliability for maritime safety applications.]]>
Copyrights © 2026
