MeteoChat: Semi-Automated Generation of Environmental Reports with LLMs and RAG

Monitoring natural phenomena and supporting timely decision-making during emergencies or natural disasters is closely linked to a detailed analysis of the available environmental data collected over the years. Due to the large volume of data available, manually producing high-quality reports requires significant time and resources. This paper presents a system, named MeteoChat, designed to automate the creation of environmental reports by leveraging Large Language Models (LLMs), which are optimized through fine-tuning techniques and Retrieval-Augmented Generation (RAG). The system operates in two main phases: in the first phase, an environmental expert defines a set of key questions and corresponding answers applicable to various types of data, such as temperature and precipitation. This information serves as the foundation for fine-tuning the language model to specialize in the analysis and generation of environmental content. In the second phase, the optimized model is integrated into an RAG-based chatbot that combines specific data to generate accurate responses to be included in the reports. Users interact with the system through an intuitive web interface and can download the final report in docx format, containing all the requested information. This approach significantly reduces the time and resources needed for report generation while maintaining high-quality standards.