Multi-Agent Geophysical AI Workflow for Automated Reservoir Characterization

M Quamer Nasim , Paresh Nath Singha Roy, Tannistha Maiti

Traditional geophysical workflows like reservoir characterization are driven in a collaborative manner where teams of geoscientists share their individual analyses to inform key decisions made by executives. However, these workflows are repetitive, time-consuming, prone to human error, and introduce subjective bias. While researchers have used automation to address these limitations via deep learning models for specific interpretation tasks, the overall complex workflow remains manual; specialists still select, run, and process model outputs, which proves to be a bottleneck and has the potential to introduce inconsistency and human bias. This paper introduces a novel, agentic AI framework, driven by a Large Language Model, that automates the geological analysis workflow, from initial data discovery to the generation of a final, multi-modal technical report. Our approach mimics the collaborative nature of a human team through a collaborative, event-driven multi-agent system built on a microservice architecture. The system comprises multiple agents, each specializing in a set of tasks. Manager Agent, that initiates the geophysical workflow, a suite of specialized worker agents (Data Finder Agent, Geological Analysis Agent, Reporting Agent) that perform discrete tasks, and a shared workspace that facilitates communication between different agents to allow for collaboration. To validate this framework, we present a case study of an end-to-end lithology analysis on data from the Athabasca oil sands area. The proposed framework successfully took a geoscientist’s query, autonomously located the correct well data, executed the lithology analysis model, and generated a multi-modal technical report. We conclude that this agentic approach represents a promising framework for efficient and autonomous scientific workflows in the geosciences.