RAG-Powered Geoscience Assistant

Abstract

The increasing volume of geoscience-related data from research papers, environmental reports, and satellite observations has created challenges in efficient knowledge retrieval and interpretation. Traditional keyword-based search systems fail to provide context-aware and precise responses for complex queries. This paper presents a Retrieval-Augmented Generation (RAG)-Powered Geoscience Assistant that integrates vector-based retrieval with large language models (LLMs) to generate accurate and context-aware answers. The system processes user-uploaded geoscience documents, converts them into embeddings, and stores them in a vector database for semantic retrieval. A Streamlit-based interface enables interactive querying, while the backend utilizes embedding models and LLM APIs to deliver grounded responses. Experimental evaluation demonstrates improved accuracy, reduced hallucination, and enhanced user experience compared to traditional search systems. The proposed system provides an efficient and scalable solution for domain-specific knowledge retrieval in geoscience.