Proceedings of the 5th World Conference on Climate Change and Global Warming
Year: 2025
DOI:
[PDF]
Accelerating Policy Decisions: A Multi-Agent System for Interactive Question-Answering on Climate Policy Data
Elizabeth Osanyinro, Ernest Effiong Offiong, Maxwell Nwanna, Grace Farayola Olaitan Olaonipekun, Abiola Oludotun, Sayo Agunbiade, Oladotun Fasogbon Ogheneruona Maria Esegbona-Isikeh, Lanre Shittu, Toyese Oloyede, Sa’id Olanrewaju
ABSTRACT:
Policymakers face significant challenges in extracting actionable insights from large-scale, unstructured datasets such as those generated during COP29. Traditional approaches, including fine-tuning large language models (LLMs), are computationally expensive, time-intensive, and struggle to adapt to evolving policy landscapes. A Dynamic Multi-Agent System for Policy Recommendation Optimisation is introduced, trained on 30,000 decision documents from COP29, leveraging modular agent architectures and advanced prompt engineering to generate precise, context-aware recommendations without fine-tuning. The system integrates a retrieval agent to efficiently extract relevant data from COP29 repositories using embedding-based search and vector databases, a synthesis agent that processes retrieved information through role-based, few-shot, and chain of-thought prompting to generate structured recommendations, and a feedback agent that dynamically refines outputs based on user interactions, ensuring alignment with regional policies and evolving regulatory landscapes. This approach introduces key innovations, including prompt-driven optimisation to achieve high-quality, domain-specific outputs without retraining, multi-agent collaboration for real-time adaptability, and external knowledge integration through structured knowledge graphs and retrieval-augmented generation (RAG) to enhance factual grounding and mitigate hallucinations. Evaluation against human-expert baselines demonstrates a 34% increase in recommendation precision, a 41% reduction in processing time, and significant improvements in adaptability compared to conventional methods. A scalable, real-time, and adaptive policy recommendation process is established, advancing AI-assisted climate governance and ensuring that policy decisions remain data-driven, transparent, and actionable for global stakeholders.
keywords: Climate Governance AI, Multi-Agent Systems, Policy Recommendation Optimisation, Retrieval-Augmented Generation (RAG), Sustainable Decision-Making