NVIDIA demonstrates how agentic AI can automate subsurface reservoir simulation workflows that traditionally require constant expert intervention. The system uses a multi-agent architecture built on LangChain/LangGraph, NVIDIA NIM inference microservices, and RAG-grounded reasoning to eliminate idle dead time between simulation cycles. A single reservoir simulation assistant handles daily tasks like parameter lookups and scenario testing, while a multi-agent squad autonomously orchestrates large-scale optimization studies (e.g., well placement using genetic algorithms and PSO). Engineers retain supervisory control via human-in-the-loop checkpoints. The framework is simulation-tool agnostic and open-sourced on GitHub, making it extensible to geothermal, CO2 sequestration, and other iterative simulation domains.
Table of contents
The agentic shiftThe reservoir simulation assistant: Accelerating daily workflowsKey takeawaysMulti-agent squads: Orchestrating complex engineering studiesCase study: Well placement optimizationThe building-blocks: NVIDIA Inference MicroservicesGetting startedSort: