Advancing Geoscience with Multi-Modal AI: A Comprehensive Copilot

In geology, LLMs (Large Language Models) have been applied to various tasks, including document analysis, geological consultation, workflow planning and reservoir simulation [1, 2, 24]. However, existing applications often focus on single-modality data or specific subdomains within geoscience, limiting their overall effectiveness in addressing the full spectrum of industry needs. As a result, LLM-based agents are not yet fully present in all the stages of the seismic exploration pipeline. The objective of the multi-modal AI copilot extension presented in this paper is to address these limitations by creating a comprehensive, integrated system capable of processing and analyzing diverse geoscience data types. This extension aims to enable seamless interaction between users and their geoscience projects (including but not limited to seismic processing, interpretation and well data analysis) through natural language interfaces, while executing complex commands and analyses on multi-modal data, including seismic volumes in SEG-Y format, through integration with 3D visualization software APIs.