BasinModelingFlow: Agentic AI for Interoperability in Petroleum System Modeling

Petroleum systems modeling workflows are often fragmented across different tools and teams, making them difficult to trace and reproduce. BigLoop addresses this challenge by integrating data, simulation, calibration, and interpretation in iterative cycles, but its adoption is limited by the lack of integrated environments to support these workflows. In this context, we present the BasinModelingFlow Platform, an agentic AI system that provides an integrated environment for creating, executing, monitoring, and comparing geoscientific simulation workflows for petroleum systems modeling. Its core innovation is a set of AI agents based on generative models supporting workflow creation, validation, and interpretation: they interpret natural language commands, recommend compatible simulators and configuration combinations, check parameter consistency, and explain executed steps and results, enabling governable automation of in silico experiments. The agents are backed by a memory component that stores provenance on parameters, configurations, and results from previous runs, which supports recommending execution plans, generating documentation, and comparing runs, with recommendations considering the user's context and similarity to previously studied analogous scenarios. This supports scenario reassessment and continuous improvement of geological models, sharpening the characterization of geological uncertainties and strengthening knowledge retention within the company. The platform also incorporates strategies to improve interoperability between simulators and software packages, whose challenges span three dimensions — technical, semantic, and organizational — relating respectively to the data formats exchanged, the consistent interpretation of their meaning across tools, and the alignment of objectives, business processes, and organizational structure. The system will adopt RESQML as the base representation format for petroleum systems data, extended with constructs for geochemical, lithological, and geohistorical information; agents will use existing RESQML-based connectors to transfer information between applications while preserving provenance, and where automation is constrained or impossible, they will advise the user on how best to move information across applications. By combining AI agents with these interoperability strategies, the platform represents a strategic advance: automated geoscientific simulations, traceable runs, and workflows and results shareable across teams, raising the technological maturity of petroleum systems modeling and fostering more robust, efficient, and reproducible workflows. As a case study, we are developing an integrated workflow that couples a commercial petroleum system simulator with an in-house geological process simulator to account for salt tectonics and restore diapiric deformation across geological time, improving the representation of basin evolution from its formation to the present day.