Intelligent seismic workflows: The power of generative AI and language models

Abstract Advanced seismic data processing involves specialized methods often implemented through various software, requiring extensive expertise and time from geoscientists to execute geophysical workflows. Recently, large language models (LLMs) have demonstrated the ability to understand natural language, reason about domain-specific topics, and assist users through complex tasks. In this paper, we introduce an LLM-based autonomous agent for seismic data processing, focusing on full-waveform sonic data workflows. The proposed agent is shown to reliably understand user queries, select appropriate tools, and execute geophysical tasks such as bandpass filtering, data clipping, and frequency spectrum analysis. Safeguards and guardrails are incorporated into the agent to ensure operation within defined parameters, maintaining data security and integrity. By automating routine processes, the agent allows geoscientists to focus on higher-level decision-making while ensuring accuracy and consistency across geophysical tasks. As generative artificial intelligence evolves, integrating LLMs with other models has the potential to revolutionize seismic data analysis, making advanced geophysical tasks more accessible and scalable for users at all levels of expertise.