MeteoChat: A Fine-Tuned and RAG-Based LLM for Semi-Automatic Report Building in Environmental Monitoring

In environmental monitoring, producing high-quality reports is crucial for timely intervention in critical situations such as natural disasters and emergencies. However, generating these reports often requires a significant amount of time and resources. This article presents MeteoChat, a system that automates the creation of environmental reports by combining Large Language Models (LLMs), fine-tuning techniques, and Retrieval Augmented Generation (RAG). The goal is to maintain high report quality while reducing the time and resources needed for their production. MeteoChat operates in two phases. In the first phase, an environmental expert defines a set of generic key questions, context, and corresponding answers independent of specific data. These question-context-answer tuples are then used to fine-tune the LLM. In the second phase, the fine-tuned LLM is integrated into an RAG-based chatbot system, which combines specific environmental data. The environmental expert can interact with MeteoChat through an intuitive web chatbot and download the final report.