Llamarine: Open-source Maritime Industry-specific Large Language Model
Annual Conference of the Japanese Society for Artificial Intelligence, 2025
Why This Matters
Maritime navigation is a safety-critical domain where errors can lead to collisions, environmental disasters, and loss of life. Navigators and maritime operators currently lack AI tools that understand the specialized rules, regulations, and physics of ocean navigation. Llamarine is the first open-source LLM built specifically for the maritime industry, providing expert-level knowledge in navigational principles, collision avoidance, route optimization, and regulatory compliance.
The Problem
General-purpose LLMs have demonstrated potential in complex reasoning tasks, but their broad training leaves them lacking in specialized maritime knowledge. The maritime industry has unique requirements around navigational principles, international regulations (such as collision avoidance rules), route planning, and risk assessment that generic models handle poorly.
No publicly available, domain-specific language model existed for maritime navigation, meaning maritime professionals could not leverage AI assistance for critical decision-making tasks without risking inaccurate or unreliable outputs.
Approach
Llamarine was developed through continued pretraining and fine-tuning on a high-quality corpus specifically curated for the maritime domain. The training data comprises maritime textbooks, research publications, and web text from Wikipedia, selected from authoritative sources to ensure depth and reliability in the model's knowledge base.
- Curation of a comprehensive maritime dataset from authoritative sources.
- Development of a foundational model capable of reasoning about complex navigational challenges with greater accuracy than general-purpose LLMs.
- Establishment of a benchmark to evaluate performance in maritime-specific decision-making tasks.
Key Results
- Llamarine outperforms both general-purpose and commercial LLMs in critical navigation-related tasks, such as trajectory planning, risk assessment, and compliance with maritime regulations.
- The model outperforms GPT-4o-mini, GPT-4o, Claude-3.5-Sonnet, and opensource models such as Llama3.1 8B, Llama3.1 70B, and Llama3.3 70B.
- First open-source domain-specific LLM for the maritime industry.
Impact
Llamarine has been adopted by Furuno for a navigation assistant achieving 100% accuracy on regulation-compliant actions. By providing an open-source foundation model trained exclusively on high-quality maritime literature, Llamarine paves the way for AI-driven advancements in maritime safety, efficiency, and operational decision-making. The paper was accepted at JSAI 2025 (Annual Conference of the Japanese Society for Artificial Intelligence).
Citation
@article{nguyen2025llamarine,
title={Llamarine: Open-source Maritime Industry-specific Large Language Model},
author={Nguyen, William and Phan, An and Kimura, Konobu and Maeno, Hitoshi and Tanaka, Mika and Le, Quynh and Poucher, William and Nguyen, Christopher},
journal={arXiv preprint arXiv:2503.00203},
year={2025}
}