VortexAI: an AI platform for meshing

The meshing bottleneck

Mesh generation is one of the most time-consuming steps in any CFD workflow. NASA's CFD Vision 2030 study flagged it as a major bottleneck, and a decade later, the process remains largely manual: translate physical requirements into backend-specific parameters, run the mesher, inspect the result, tweak, repeat.

This process can take several days of focused work and demands simultaneous expertise in fluid mechanics, numerical methods, and the specific meshing tool. For non-specialists, the learning curve is steep. For experienced engineers, the iteration eats into design cycles.

Introducing VortexAI

VortexAI automates this entire workflow. Give it a plain-English description of your simulation and a geometry file (STL or STEP), and it produces a quality-evaluated computational mesh ready for CFD analysis.

Under the hood, several specialized LLM agents work in sequence: they inspect the geometry, reason about the flow physics, compute boundary-layer parameters, set discretization requirements, plan the meshing strategy, and generate the final configuration files. Each agent calls deterministic computational tools to ground its decisions in quantitative analysis rather than guessing.

Three things set VortexAI apart:

• Not retrieval-dependent. Every meshing decision comes from first-principles physics reasoning and tool outputs. The system does not need existing OpenFOAM cases or tutorial databases.
• It learns from its mistakes. A reflection cycle analyzes each attempt, identifies what went wrong, and updates a set of learned rules (playbooks) that guide the next try. The system improves across iterations.
• Three backends, one interface. The same natural-language input works with snappyHexMesh (3D unstructured), cfMesh (2D unstructured), or blockMesh (structured). Backend-specific logic is handled internally.