Today, we are launching an interactive Physics AI demo so anyone can experience the power and ease of predicting product behavior, built on AI inference. You need to experience it firsthand to understand the transformative impact of fast and accurate results on engineering design.
Snapshot of the AI prediction demo
To this end, we’ve provided a minimal user interface that exposes a variety of our pre-trained SHIFT models, spanning different industry applications including aerospace, automotive, and industrial machinery. Our first aerospace demo uses the SHIFT-Wing model, a transonic aerodynamic model trained on the largest simulation dataset for aircraft wing design. Over time, we plan to release more models across new applications. While this demo is separate from the Luminary Platform, it showcases the result of our platform’s end-to-end capabilities: generating training data, combining it with external data, training a model, and deploying it.
Inference is like a chef’s intuition
Let’s use a simple analogy to explain the difference between simulation and physics AI predictions. Think of running an engineering simulation like meticulously following a complicated cooking recipe word-for-word, measuring every gram and every second in the oven. Now, imagine running AI inference like a world-class chef who has the experience and intuition to taste the ingredients, glance at the stove and just know how to get the dish right.
Inference is like a chef cooking quickly with intuition, whereas simulation is like a chef cooking slowly by following a recipe.
The master chef (Physics AI model) has internalized thousands of past dishes, or “training data,” allowing them to jump straight to a great result that’s often better tailored to the specific constraints of the moment. In this context, the model is superior for tasks requiring speed, flexibility, and repeated experimentation. Conversely, the recipe (simulation) is fully explicit and reproducible, yet it is slow and fragile—a slight change in conditions can require rewriting the entire process. The recipe remains superior for teaching fundamentals and validating the chef’s instincts.
Why use inference?
Speed matters when it comes to winning in competitive markets. It can mean the difference between winning or losing a contract, ensuring a program is delivered on time and within budget, or solidifying a market advantage. Physics AI prediction dramatically accelerates the time to test and validate ideas, reducing the risk of investing too much into poor designs. It delivers high quality results in seconds without the high cost of traditional simulation: CAD expertise and manipulation, mesh generation for different geometry and flow conditions, numerical methods and solver expertise to ensure the solution is stable, and even post-processing of the results.
While this demo proves the impressive speed of inference, it also shows the flexibility of how a Physics AI model can be deployed. A bespoke interface can be built for different teams and users, including designers, simulation analysts, and program managers. You can even integrate the predictions in design tools like Onshape or Blender. When Physics AI prediction becomes easy and accessible, more people can collaborate and make faster decisions.
Try it yourself
This demo is intuitive to use. There are just three simple steps:
- Select a case of interest by industry
- Specify the geometry and flow conditions using the sliders
- Click “Predict” to generate the physics prediction on the surface of the geometry
Try the demo for yourself and share it with your colleagues and friends via email or LinkedIn and X. Let us know what you think!
