Framework for AI-Assisted Subsurface Data Access

This article presents a technical report detailing a framework for AI-assisted access to subsurface data, focusing on explicit data representations, agent-based workflows, and efficient information retrieval. The report outlines the large-scale conversion of SEG-Y archives into self-describing MDIO v1 datasets, along with a case study that illustrates agent-driven reconstruction of seismic metadata from legacy text headers. Another case study evaluates embedding-based retrieval methods across acquisition and processing reports, demonstrating the effectiveness of vector quantization and graph-based indexing for low-latency, relevance-driven search. The framework integrates these capabilities into an interactive, multi-agent system that facilitates natural-language analysis and coordinated access to both structured and unstructured subsurface information. The report addresses challenges faced by energy organizations in managing vast amounts of seismic data, emphasizing the need for a consistent, machine-readable structure to enhance operational efficiency and decision-making processes.