GURU Generation 2: Autonomous Workflow Automation for HPCMP Users
The Department of Defense (DoD) relies on computational fluid dynamics (CFD) simulations to design advanced weapons systems, particularly high-speed vehicles operating across transonic, supersonic, and hypersonic regimes. MSBAI’s GURU Generation 2 autonomously drives complex CFD workflows on High Performance Computing Modernization Program (HPCMP) supercomputers, significantly enhancing productivity and reliability for all HPCMP users. Building on the prior operational GURU Generation 1 system installed at AFRL DSRC, this next-generation AI platform will drive meshing tools (Salome, GMSH, NASA Refine), solvers (CREATE CFD suite, SU2, OpenFOAM), and visualization (ParaView) to streamline simulation setup.

Targeting deployment on HPCMP systems (Raider, Carpenter, Nautilus, Blueback), GURU Generation 2 uses a hierarchical AI architecture. It fuses neuro-symbolic reasoning, blending symbolic logic with neural networks to ensure transparency, explainability, and adaptability. Scalable multi-agent reinforcement learning (DD-PPO) orchestrates workflows by training autonomous agents to collaborate across distributed tasks. Cross-domain knowledge transfer enables the system to adapt skills across diverse tools and domains, minimizing retraining. Techniques like Joint Embedding Predictive Architecture (JEPA) learn shared representations from multimodal data for tasks like anomaly detection, while AutoML (DeepHyper) automates hyperparameter tuning and model selection, optimizing performance. Scalability is achieved through distributed training and parallel processing, leveraging HPCMP’s unique hardware and schedulers for efficient resource use.

We have already achieved an 80% reduction in CFD setup time, a drop in failure rates from 88% to <2%, and 84% boundary layer capture propagation for high-quality meshes. These advancements accelerate DoD development cycles, delivering faster and more reliable analyses essential for national security.

MSBAI began by automating laborious digital engineering tasks, such as the set up of CFD, finite element analysis, orbital mechanics, and trajectory simulations. Now, GURU has expanded into autonomous operations use cases, where complex time-series data must be continuously interpreted, to ensure system reliability, detect anomalies in real time, and sustain performance across energy, manufacturing, space traffic control, and other high-stakes environments.
IMPACT
Accomplishment: Developed and deployed GURU Generation 2, an AI-driven system that autonomously configures and executes high-fidelity CFD workflows across leadership-class DoD supercomputers; Result: Reduced simulation failure rates from 88% to 2%, cut geometry setup time by 80%, and increased boundary-layer capture accuracy from 8% to 98% – Enabled faster, more accurate design of hypersonic vehicles and streamlined digital engineering across the DoD
PRESENTER
Allan, Grosvenor
allan.grosvenor@microsurgeonbot.com
425-829-9123
MSBAI
CATEGORY
AI/ML for HPC
SECONDARY CATEGORY
Comp Fluid Dynamics
SYSTEM(S) USED
Raider, Carpenter, Nautilus, and Blueback