Insights

Physics AI Engineering Notes

Technical writing about Physics ML, NVIDIA PhysicsNeMo, surrogate modeling, scientific software, GPU infrastructure, engineering AI, and production implementation. Nothing is published yet — the topics below are what TecunTecs intends to write about.

Planned topic

Can PhysicsNeMo accelerate your CFD workflow?

Where PhysicsNeMo workflows fit relative to conventional CFD, and how to judge whether acceleration is realistic for a given class of analyses.

Planned topic

DoMINO vs. MeshGraphNet vs. Transolver

How geometry-aware, mesh-graph and transformer-based approaches differ, and which physical problems suit each.

Planned topic

What simulation data do you need for an AI surrogate?

Design-space coverage, solver consistency, held-out cases, and the practical data readiness questions that precede training.

Planned topic

From CFD data to a PhysicsNeMo training pipeline

Turning accumulated solver output into reproducible datasets, preprocessing meshes and geometry, and structuring training runs.

Planned topic

When Physics ML is the wrong answer

Signals that a workflow should be optimized, parallelized or re-meshed rather than approximated with a learned model.

Planned topic

How to validate engineering AI surrogates

Error distributions, out-of-distribution behaviour, and acceptance criteria defined in engineering rather than ML terms.

Planned topic

Deploying PhysicsNeMo inference in production

Containers, CUDA environments, GPU scheduling, batching, and the operational concerns of serving Physics AI models.

Planned topic

Building interactive applications around CFD surrogate models

APIs, scientific visualization and interface design that let engineers actually use a fast predictive model.

Planned topic

Using Claude and MCP around validated engineering tools

Orchestrating deterministic solvers and validated models through language-model interfaces while keeping results traceable.

If one of these questions is live for your team right now, it is usually faster to discuss it directly than to wait for the write-up.

In the meantime

Talk about your engineering workflow

Bring the simulation bottleneck, the dataset, or the technical question you are trying to answer.