The LLM will utilize a user-friendly interface with tools tailored to different policymaking needs, prompt engineering to augment user input, and Retrieval Augmented Generation (RAG) to fetch and rank data from pre-determined sources.

The development process starts with the User Input Module capturing and directing user questions. The Keyword Extraction Module identifies key phrases, and the Source Selection Module matches them with relevant data sources. The Fact-checking and Relevance Module verifies data accuracy and relevance. The Prompt Tuning Module refines user intent for enhanced response quality.

The Foundation LLM, incorporating models like ChatGPT, Claude 3.5, and Gemini, generates detailed responses. The Response Synthesis Module ensures coherence, and the User Response Delivery Module provides the final answer with source attributions.

The key differentiators of the tools are:

Scientific Integrity: Built on a foundation of high-quality scientific data, the LLM will accurately source-track information from credible reports and journals.

Comprehensive Knowledge Synthesis: The model will convert complex, technical information into concise, non-expert-friendly recommendations tailored for policy use.

Diverse Perspectives: Incorporates a broad range of knowledge, including Indigenous and local community insights, global case studies, NGO and private sector.

Local Policy Customization: Includes local policy knowledge to generate customized policy pathways, ensuring that recommendations are relevant and actionable within national contexts.

Ethical and Transparent: Designed in alignment with UN standards for data handling and AI ethics

The LLM was presented during the Our Ocean Conference in Busan (South Korea) in April 2025, and will be showcased at the UN Ocean Conference in June 2025.

The project is currently in phase 2 until April 2025. In this next phase, IPOS will focus on three key areas:

LLM Development: Collecting and normalizing ocean data, prioritizing sources, and building and testing the LLM to ensure accuracy and reliability.

Real-World Application with the Seychelles: Collaborating with the Seychelles to apply the LLM in real-world scenarios, supporting their journey toward achieving global ocean sustainability targets. This use case will directly inform the tool’s development.

Scientific Oversight: Assembling a scientific committee to guide the LLM's development and ensure its outputs are scientifically rigorous and actionable.