Abstract:
Static geomodelling is constrained not only by algorithms, but also by serial handoffs among data conditioning, interpretation, gridding, facies modeling, petrophysical property modeling, volumetrics, and quality control. Each iteration can require repeated software interaction, manual transfer of assumptions, and specialist review. This presentation investigates whether agentic artificial intelligence (AI) can transform that fragmented sequence into a traceable, human-governed modeling loop without displacing geological judgment.
The architecture separates responsibilities: a large language model interprets intent and plans the work; Model Context Protocol (MCP) exposes structured software tools and returns project context; and tNavigator performs the deterministic calculations and stores native project objects. In the demonstrated static-model case, the agent inventories available data and functions, then constructs a workflow for machine-learning-assisted facies interpretation, facies-proportion mapping, horizon interpolation, stratigraphic gridding, well-log blocking, sequential indicator facies simulation, and facies-conditioned sequential Gaussian porosity simulation. It recommends interpolation and variogram settings, invokes the workflow, evaluates intermediate outputs, and assembles spatial and distributional quality-control diagnostics. Human checkpoints retain authority over geological concepts, parameter acceptance, and final validation.
The case demonstrates how workflow-assembly and iteration cycles can be shortened while preserving an inspectable chain from user objective to parameters, software actions, results, and assumptions. More importantly, it reframes AI from a prediction-only tool into an execution layer that can plan, act, evaluate, and revise within defined technical guardrails. Recent tNavigator advances, including a local MCP server, multi-agent collaboration, AI-assisted workflow editing, and an integrated static-to-dynamic environment, provide a practical foundation for extending this pattern to uncertainty ensembles, assisted history matching, reservoir simulation, and field-development scenario screening. The intended outcome is not an autonomous geological truth, but faster generation and testing of technically defensible alternatives, enabling geoscientists and engineers to focus on uncertainty, risk, and decisions.
Biography:
Ivan Praja is Geology Team Lead at Rock Flow Dynamics Southeast Asia, with more than 12 years of experience in subsurface characterization and integrated reservoir studies. His work focuses on geological modeling, geomechanics, uncertainty analysis, and static-to-dynamic reservoir workflows. He holds an MSc in Integrated Petroleum Geoscience from the University of Aberdeen.
