AI tackles SUV aerodynamics

Luminary Cloud released an AI foundation model with help from Honda and Nvidia. Plus: have you heard of Hestus?

Welcome to Engineering Paper. Today’s batch of design and simulation software news starts with—what else?—AI.

Luminary Cloud announced SHIFT-SUV, a new physics-based AI foundation model for automotive aerodynamics. SHIFT-SUV was trained on a dataset of more than 1,000 simulations of AeroSUV, an open-access SUV reference model that Luminary Cloud parametrically modified with input from automaker Honda.

Geometry variants of AeroSUV. (Image: Luminary Cloud.)

The AI foundation model, which was trained with Nvidia’s PhysicsNeMo framework, aims to give automotive designers a prediction of aerodynamic performance without the need for time-consuming CFD simulations.


“With SHIFT-SUV, designers can explore geometry variations and instantly assess aerodynamic performance, without preprocessing, mesh generation, or solver configuration,” according to Luminary Cloud’s blog post introducing SHIFT-SUV as the first of a planned suite of AI foundation models. The AI model was trained with Nvidia PhysicsNeMo.

(Image: Luminary Cloud.)

Luminary Cloud is releasing both the SHIFT-SUV foundation model and the training dataset as open source, with a call for the community to contribute to its improvement. Luminary Cloud says it plans to continually add to the dataset and retrain the model with 25,000 SUV variations by the end of the year. The company has also developed a demo integration that provides SHIFT-SUV inference directly in Blender (as seen in this video).

SHIFT-SUV fits neatly with Luminary Cloud’s recent projects, including a virtual wind tunnel demo with Nvidia late last year and a collaboration with nTop and Nvidia on a pipeline for geometry variation to simulation to AI training, which the company announced last month.

Hestus’ hackathon and quest for a CAD copilot

If you live in San Francisco and use Autodesk Fusion, you could win a pair of Ray-Ban Meta AI sunglasses.

Hestus, a developer of CAD automation software, is hosting a hackathon on April 24, 2025 that will give out three pairs of the smart shades to the fastest Fusion modelers.

First things first: What is Hestus? I recently spoke with the startup’s founder and CEO, Sohrab Haghighat, to find out. He told me that Hestus wants to use AI to make mechanical design as agile as software design.

“The cost and time of fixing mistakes in the hardware world is way more than what it is in software,” Haghighat said.

Haghighat believes that CAD software suffers from a lack of understanding. CAD doesn’t understand manufacturability. CAD doesn’t understand testing requirements. CAD doesn’t understand the interdependence of parts in an assembly. When a mechanical engineer has to make a change, even a seemingly simple one, it’s up to them to account for all the downstream effects.

“Mechanical engineers are not slow. They actually are as quick as software developers in coming up with the first concept,” Haghighat said. “The moment you go into documenting your design in CAD, the moment you go into making changes in CAD, that’s where the slowdown happens.”

Hestus’ first attempt to fix that slowdown is Sketch Helper, an add-on for Autodesk Fusion that helps users apply constraints to their sketches. It offers on-the-fly suggestions for constraints such as tangents, equalities, and horizontal or vertical lines. Haghighat says features for dimensioning and symmetry constrains are coming soon.

Sketch Helper is similar to Autodesk’s own Sketch AutoConstrain tool, but Haghighat believes Sketch Helper’s approach is better because it offers live recommendations constraint by constraint. In contrast, AutoContstrain generates a bunch of options for constraining the sketch as a whole.

Hestus Sketch Helper (top) and Sketch AutoConstrain (bottom) take different approaches to help designers constrain and dimension their sketches. (Images: Hestus / Autodesk.)

I spoke to one Sketch Helper user, and she praised its usefulness for Fusion modeling.

“It just makes me a lot faster,” said Olivia Li, founder of Spyro Labs, a company that designs and builds stratospheric weather balloons. She’s been using Sketch Helper for about eight months and says she would gladly pay for it, though it’s currently available for free. (Li hasn’t yet tried Sketch AutoConstrain and so couldn’t compare the two tools.)

Haghighat told me Sketch Helper is built with a mix of hard-coded rules and machine learning, adding that as Hestus rolls out more complicated features it will lean more heavily on AI. The company’s grand vision is to develop a CAD copilot that can help engineers in myriad ways, such as offering manufacturability suggestions or automatically sourcing components.

“We want to make mechanical design faster by interpreting what is the next step for the designer,” Haghighat said.

Hestus’ April 24 hackathon will test how well Sketch Helper can do just that. The event will include three rounds of speed modeling, where contestants will race to create 3D models from 2D drawings (in the vein of TooTallToby’s online speed modeling competitions). While competitors aren’t required to use Sketch Helper, Haghighat strongly believes those that do will have an edge.

Intrigued? You can sign up for the event here.

Onshape AI Advisor is now live

Onshape has released its AI Advisor, a chatbot trained to answer questions about the cloud CAD platform. We’ve known it was coming for several months—Onshape founder Jon Hirschtick told me all about it in February—and now it’s finally live.

Onshape AI Advisor is currently available as a beta in the Onshape learning center, the company’s catalog of training resources. The chatbot will soon be integrated directly into the Onshape platform, according to Onshape’s blog post announcing the release.

I asked AI Advisor to tell me about itself (see its response in the screenshot below). It gave an adequate description, but I wasn’t impressed with the citations. It looks like there are three of them, but they all point to the same Onshape forum post about the April 1 update that included AI Advisor.

Citations were poorly handled for most questions I asked, with jumpy numbering and repeat links. Otherwise, Onshape AI Advisor seems to work pretty well for the “How do I do X in Onshape” questions it was built to answer. It gives reasonable responses and the links it does provide are usually appropriate.

I managed to test AI Advisor for one session, but since then I’ve gotten an error message every time I ask it something. AI Advisor is free for all Onshape users—for now, anyways—so give it whirl and let me know how well it works for you at malba@wtwhmedia.com.

Update April 16, 2025: Proving once more that the quickest way to get tech support is to complain publicly, a PTC representative reached out to me with an embarrassingly simple fix for my Advisor error: sign into Onshape. (In my defense, I thought I was, but apparently one can be signed into the Onshape learning center and not Onshape proper.)

Just the gist

  • Gstarsoft, the China-based developer of the GstarCAD software suite, has acquired Hungarian BIM developer CadLine.
  • Siemens Digital Industries Software announced it’s completed its acquisition of DownStream Technologies, a developer of PCB manufacturing software, for an undisclosed sum. (Siemens’ ONE Tech Company growth plan was not mentioned in the press release.)
  • Materialise announced the 2025 version of its Magics software for 3D printing. The company says Magics 2025 integrates nTop implicit geometries and includes next generation build processors that will “make the unprintable printable.”

One last link

Engineering.com contributor Matt Lombard with What does this CAD User want from AI?

Got news, tips, comments, or complaints? Send them my way: malba@wtwhmedia.com.

Written by

Michael Alba

Michael is a senior editor at engineering.com. He covers computer hardware, design software, electronics, and more. Michael holds a degree in Engineering Physics from the University of Alberta.