AI-Powered Innovation in Engineering Design — with nTop and Neural Concept

Episode Summary

The episode delves into how artificial intelligence (AI) is transforming engineering design and product development through the lens of nTop and Neural Concept. Thomas von Tschammer from Neural Concept, a pioneer in AI for engineering design, and Brad Rothenberg from nTop, which provides high-fidelity parametric modeling tools, share insights on leveraging AI to accelerate design cycles and enhance product performance. Both companies focus on enterprise-level solutions rather than small and medium-sized businesses.

Key technical insights include the integration of AI as an underlying technology in traditional engineering ecosystems, enabling faster and more robust model parameterization. Brad discusses how nTop's approach has evolved from a simple software sale to a process change that requires close customer collaboration for successful implementation. Thomas emphasizes the importance of running boot camps to educate engineers on the potential of AI workflows and address current tool limitations.

For PLM and engineering professionals, the episode underscores the transformative power of AI in streamlining design processes and unlocking new levels of product performance. Companies adopting these technologies can expect significant reductions in time-to-market and enhanced innovation capabilities, making it essential for them to explore how AI can be integrated into their workflows.

---

Full Transcript

Michael Finocchiaro

And we're live. Welcome once again to the ⁓ AI Across the Product Lifecycle podcast. This is Michael Finnicara, your host. And I'm very, very pleased today to present Thomas van Tschammer of the Swiss Startup Neural Concept and Brad Rothenberg of Intop. Both of them actually are located in New York, but next time they'll have to be in the same office, I think. How are you guys doing today?

Brad

Very good. Good to see you. ⁓ Thanks for having us. Yes, we realized this morning we're both like a couple blocks from each other. if you want to pause it like 20 minutes, we probably could be in the same.

Thomas von Tschammer

Very good.

Michael Finocchiaro

Hahaha Next time, next time. We'll do a follow-up and we'll do that. ⁓ Why don't you guys introduce yourselves? I think that a lot of people have probably heard of nTop and Neural Concept particularly the announcement in December of the 100 million from Goldman Sachs was a big deal. But maybe introduce yourselves and introduce your companies before we dive into the questions.

Thomas von Tschammer

should.

Brad

So Brad Rothenberg, CEO, founder of nTop and we're a new modeling tool for our customers to basically build very high fidelity parametric models very, very quickly. given today's discussion is around AI, I don't think of nTop as like an AI app itself, but I think of nTop as AI infrastructure or underlying modeling technology enabling AI. And so exciting to talk with you guys today and.

Michael Finocchiaro

Exactly.

Brad

Me and Thomas have known each other for years now and finally starting to get some really good work together in the market. So excited to chat about that.

Michael Finocchiaro

That's totally awesome. How about you, Thomas?

Thomas von Tschammer

Yeah, thanks a lot, Michael, for the invitation. So Thomas von Tschammer I'm one of the co-founders and managing director for the US. Of Neural Concept, I've been in Switzerland for the past 10 years, background mechanical engineering in computer vision originally. And about two years ago, I moved here in the US, so neighbor of Brad, to open up the US branch of the company and build the team here. And about Neural Concept, very quickly. So we are pioneering and living in the field of AI for engineering design and product development. We bring that intelligence layer on top of the traditional engineering ecosystem of CAD CA tools. And as Brad was mentioning too, with the goal to speed up design cycles, reduce time to markets and really unlock the next level of product performance for large engineering organizations.

Michael Finocchiaro

That's awesome. Yeah, so you guys are both really focused on the enterprise market and not the SMB market, which is interesting. So back in, if we look back to November, 2022, there was quite the pivot point, right? There was a turning point in terms of suddenly the normal Joe knew something, there was something called artificial intelligence that was gonna change his daily life or her daily life. There was also a lot of skepticism and also bullishness on different people, right? So I was interested in the two of you, like, were you guys more bullish on, you know, the possibilities of how LLMs and this kind of computational AI was going to transform engineering or were you guys a bit skeptical and kind of a wait to see? I was just interested in that. Either of you can pick that up. It's okay.

Brad

was gonna say, I mean, was it really three years ago now? It feels like 10 years or something like that. I think.

Michael Finocchiaro

I know really, right? Well, ever since COVID, everything feels like a decade, right?

Brad

Yeah. So, I mean, I think when the LLMs broke out, I mean, we had been looking into AI since the earlier, like even when I was in school before Entop, we talked about neural networks and AI. And I think we were always, or I specifically was always bullish on the potential of AI. And I think what happened when the LLMs entered the market, it finally became like common knowledge to everybody that AI was here. It's not a fad. It's... It's going to have an impact and it has an impact on just like normal parts of everyday life. It's not just like something that's a tool that people use. so I think at the same time, there was also a lot of these like really silly, like text to CAD demos that were coming out with the LLMs or like code, you know, we have, we have CAD is code and we're going to hook an LLM up.

Michael Finocchiaro

for academics.

Brad

to that and generate geometry from that. And my God, we can make a bracket with four holes in it. Like, holy crap, this is like, know, revolutionize the world. ⁓ At the same time, I was thinking, you know, with nTop, the entire nTop model is backed by a directed acyclic graph of operations that's represented in like a JSON format. And LLMs are really good at reading, editing.

Michael Finocchiaro

Yeah.

Brad

modifying, generating those. And so we were experimenting with that too and found some pretty interesting use cases of having an AI kind of co-pilot or partner with you in terms of helping construct, edit, tweak, modify an end top model. And so I do think at the time, what I was really excited about was actually that like, we're a new piece of engineering software in the market. The legacy tools have 30 to 40 years of features built on top of traditional boundary representation technology. To me, AI is like, okay, this is an accelerator. All of a sudden, AI could accelerate building a complete feature set around an implicit model that nTop is built on so that you can build sheet metal components, you can build injection molded components, you can design airframes that are, you know, with traditional machine bulkheads. And so,

Michael Finocchiaro

Mm.

Brad

I think this, that the transition to AI, you know, is essentially the, the, the new thing. If you look back in the history of engineering software, one design cycle on the drafting boards in the sixties probably took, you know, a couple of weeks with CAD. You can reduce that to a couple of days with AI and end top. seeing, you can now reduce that down to minutes, or you can run, if you have unlimited compute, you can run hundreds of thousands of. different variations on a design to understand exactly if a certain aircraft is going to meet a certain mission requirement in minutes. And so it's really, to me, it's an accelerator.

Michael Finocchiaro

You might need a nuclear power plant to do that next to you, but right for the park.

Brad

Yeah, or our customers making turbines to power it.

Michael Finocchiaro

It's incredible. ⁓ It's amazing how fast this is moving. How about you, Thomas?

Thomas von Tschammer

Yeah, would definitely relate to what was saying. So in our concepts, if you think about our company, we started in 2018. And before that, we were doing research in computer vision, which is a class of AI models. So we are relatively old, if you can call it so, in the industry and for AI in itself. So we were there way before LLMs were even a thing. But because of that, we're always convinced about what this can bring to the industry. But as Brott was saying, way beyond just superficial text to CAD or LLMs that you just plug in interface and hope for the best. Because engineering requires a whole other level of details and has different requirements. So it's not that simple. ⁓ But really, in 2022, what happened is that it shifted the mind of people because they also realized the potential. of these approaches for engineering as well. ⁓ So we really switched from conversations where with executives in the engineering space where, you know, they were asking if AI is a thing, right, to how should I be implementing AI at scale within the organization, right? And we really see that difference in their mindsets. ⁓ And then similar to what we saw is that we, know, as modern companies, you AI native companies, we are extremely well positioned to leverage the capabilities of these LLMs, right? Similar to nTop, you know, our platform is based on Python. LLMs are extremely good to interact with code, right? What we're seeing is that it can accelerate drastically the development of these AI-driven workflows using this LLM agent on top of what we've built, right? And just earlier this year, we've announced, you know, this design co-pilot, which is this LLM layer. sitting down on top of our existing platform where we give this geometric and physics context to these agents.

Michael Finocchiaro

Awesome. We're going to get to the integration of the AI into your products in a moment. But even before that, I was thinking too, it's interesting how two of the biggest revolutions were just having one line in middle of the screen, right? Google, the first thing you saw was just a line where you could type in suddenly search the internet. now with the LLMs, it's kind of interesting how simple things are, art becomes so incredibly powerful. ⁓ How has ⁓ AI changed? the, uh, well, let's say how I have the new things like what cursor was so 2024 now, right. But how is, like, how is the end top and you and neural concept using AI and the development, um, of code? I mean, is every single developer using cloud code or some kind of code pilot? You know, how, are you actually integrating that? I'm sure it's been a total revolution to both of you guys in terms of how developers work on a day to day basis, right?

Thomas von Tschammer

Yeah, definitely. Yeah, definitely it has. ⁓ Now we have embedded best practices using AI at the core of our software development process, right? So every single engineer in the team has access to these agents to support their development, right? All of that is centralized into ⁓ global soft knowledge that everyone can rely on as well. But for the code, you know, this is the lowest-end input and the first application that we've deployed.

Michael Finocchiaro

You can pick that up, Thomas. Go ahead.

Thomas von Tschammer

very rapidly and now we're using it scale. But this goes beyond just code. We use that also for the different teams, marketing, go-to-markets, ⁓ to build different strategies and others using custom agents that have access to all our documentation, our material internally at Nora Concepts. That allows also new people to ramp up much faster. We are in the phase of very fast growth at Nora Concepts. And we're seeing that these agents are incredibly

Michael Finocchiaro

course.

Thomas von Tschammer

incredibly powerful to share this knowledge with new people joining the team, make sure they can get up to speed much, much faster because they have basically someone next to them to whom they can ask these questions. And this agent has the full context and the full story as to what the company has been about.

Brad

Yeah, I mean, think that's kind of the thing where it's like almost most impactful. Like as you're ramping up in a new job, you want to just ask somebody next to you, like, hey, like, how do I do this? Or where do I find this? Or where do I get this? Or if you're learning end top, where do I find, how do I do this? Like, hey, I want to, you know, parameterize up a two panel wing. How do I do that? And having an AI agent that you can just ask and do that doesn't distract somebody else and break their flow of doing things. And so we see a huge... ⁓ benefit there. And same thing with Thomas, like our dev team is kind of unified around clogged code. Some people use Gemini as well with both those tools internally. But what's amazing is just how much the tools continue to improve. Like they're accelerating. Even in the last three months, clogged code is exponentially better than it was a year ago. ⁓

Michael Finocchiaro

it's insane,

Brad

you know, it continues to improve whether it's like for each pull request, having an agent that can review the pull request when there's new bugs that get filed, having an agent that's able to look at that, figure out what the solution is, et cetera. ⁓ It also varies on the team, right? Like, cause we have some very low level, like kernel developers, and these guys were like very hesitant or very resistant to the AI tools up until about three months ago when it flipped and the dev team has been embracing it. ⁓ There's still humans in the loop for all the code that gets submitted into the code base, but it's more like a copilot or a co-assistant that helps with the code.

Michael Finocchiaro

But. So do you think sometimes you're gonna have some agents showing up for scrum meetings? I is it changing fundamentally sort of the, are we going from a waterfall and agile, some new mode that's sort of agent powered and agents are sort of in the middle of all that?

Brad

I mean, I think if you just look at the number of features that we've been able to deliver in the last six months compared to the prior like three years, it's an order of magnitude more per sprint, per month, per. for really.

Thomas von Tschammer

Yeah, to your point, Michael, I don't think this would replace engineers in the engineering room, right? But it will really empower them. That's what we're seeing, right? The same highly skilled engineers able to deliver much, much more being assisted by these disco pilots in the development. And that's true internally at Nura Concept, but that's what we're also seeing with engineers, right? Leveraging AI in engineering design.

Michael Finocchiaro

Okay. ⁓

Brad

Yeah. And the other thing too is like our field, like, you know, our field team that goes on site and works really closely with customers. They're also highly using AI in terms of like, yeah. Well, and also like just to like, if they have to build something new in this, in the tool, being able to ask an agent like, okay, what's the best way to parameterize this or what's the best way to structure this or, know, can you clean up my model? Can you clean up my file?

Michael Finocchiaro

to capture requirements and feedback. yeah. That's amazing. ⁓

Thomas von Tschammer

One of the first use cases, for example, here for us as well to Brad's point was looking at the pipeline developed on the platform and asking, what does it do and where are the errors coming from if there are any that are being raised because that's where agents are extremely good at. And that allows you to debug and iterate to others of magnitude faster, typically.

Michael Finocchiaro

It's also amazing, it seems like there's a lot less hallucination. Have you guys put a lot of, have you had to put a lot of guardrails around it, you know, using like GSD or Ralph loop or whatever, or have you found that it's actually just because the quality of Opus four or five, it's just so many leaps and bounds over all the previous ones. Just.

Brad

I mean, in the AI that's used in production, so we just did a webinar around this where we've been partnering with the guys at Astari and SysGit where you can basically have a plug.

Michael Finocchiaro

that was an awesome podcast. everybody's got to watch that one. Anybody, people listening, you got to see that one. It's really good.

Brad

Yeah, well, the idea is like, how do you go from requirements to flight in order of magnitude faster? And, ⁓ you know, in that workflow, the guard rails are built by a combination of humans and AI in the form of unit tests. And so in using AI in any production application, think the guard rails are really critical because they, these, they still do even the latest cloud model will still like. make mistakes or hallucinate. I am interested now in these like open claw like workflows that are just like consistently running and always running in the backend. Like imagine like a team of, you know, configurator, arrow guys, structures guys, propulsion guys, all working on a program, have an AI agent and they're all doing their own work and calcs and then sending their work to the agent to basically organize, compile into reports, tell them where the bottlenecks are. say, okay, where is L over D right now compared to rev? Like in some way, I think an open claw like workflow is almost like the next generation of PLM. Like it's an active and live PLM. I think that, I think this is gonna be a big change that's going to happen in the next year.

Michael Finocchiaro

Yeah. I talked to Adam Keating, he had sort of the same idea that ⁓ there's gonna be a fundamental change to the data management by AI. I still think we're gonna need a system of record. Sometimes you still have to write it down like in stone because the AI forgets after a little while, but I agree that there's something coming. Go ahead, Thomas.

Thomas von Tschammer

think what's very important here, and especially in engineering workflows, is that you cannot start from scratch. You need a very solid foundation of digital ecosystem if you want to leverage this agent to scale and put it to it. You cannot start from scratch because we need guardrails as Brad was saying. All of that needs to be embedded with physics, with requirements at the core of it. These models need to be able to interact with the different tools in the ecosystem, whether it's N-TOP and others. to be efficient, right? It's not like you use Chagy Pity on your own and you ask it questions and here, know, get hallucinations, you know, you can live with it. In engineering, you cannot afford to have that, right? So you need to have these agents living at the core of your workflow, integrated with your different tools that you're using on a daily basis, so that ⁓ you can do these checks, this can be grounded in physics, this can be grounded with the requirements that you have. ⁓ And this hallucination is not even a topic, right? Because it's not structured.

Brad

Yeah, I mean, the same, the same, you still need the same things. You need traceability. So when you see numbers in somewhere, you need to know where do those numbers come from? Do those numbers meet the requirements? they, like, where are they coming from? Why are they coming from? And you need, you know, associativity and parametrics, right? So you need associativity. So numbers are related to other numbers or data is related to other data and you need parametrics. So like when you make a change somewhere that change is able to flow. through the process, like if I change my engine from a 50 horsepower engine to a 75 horsepower engine, how does that change flow through to the design?

Michael Finocchiaro

There's more empowered digital thread, is really where AI is going to be super important. fact, David Beagle and my friend just commented that elevated spec based development is the future. Thinks MBSC may raise its head in the future to help structure this. Well, I think that's what Cisco is doing, right? Cisco and TraceBase are doing.

Brad

Well, I mean, think this is the thing, like the systems engineers always were kind of separate from the design engineers. And they were like a different layer where like the design engineers would work on the 3D model, which in my opinion, no offense to the systems engineers is the source of truth, right? That's the 3D object. That's where the real engineering, the real engineering happens. The systems engineering definitely represents the architecture of the system. And I think

Michael Finocchiaro

Mm-hmm. Hahaha

Brad

In the past, systems engineering was thought of as just like elevated box diagrams representing an architecture. And I think actually what's going to happen is the systems engineering is going to become directly integrated, connected to, by, and ⁓ interacting with the 3D. And you're going to have this connection between the two. I think ⁓ there's SysGit, I think Parri at Flow is doing really interesting stuff on the, yep, on the systems modeling.

Michael Finocchiaro

trace that space as well.

Brad

And I think there's going to be a real digital connection, like a real interlocking of the system's tools in a way that doesn't really exist between like Cameo and Katia, for example.

Michael Finocchiaro

Right. Well, actually I would direct this at Thomas, but I was thinking too, that one of the transformations that AI is making in engineering is bringing all of these things that were separate disciplines altogether, right? Because it used to be you had, as you just said, Brad, you had the guy that was doing design who had no idea, who didn't even care who the systems guy was. And then you have these simulation guys all the way on the other side of the campus. and even worse, you have the manufacturing guys who we would never talk to in a million years.

Thomas von Tschammer

Yeah.

Michael Finocchiaro

And now all those distance are coming together, right? And manager has to be thinking about design for mission, design for sport and all that stuff. Right, Thomas?

Thomas von Tschammer

Yeah, definitely. We have a very good example to illustrate that. It's typically aerodynamics or car, right, in the automotive industry. The aerodynamics is not purely an engineering problem because you have the aesthetics of the car that come into play, right? And how are companies working together today? Well, you have the design team, the studio team on one side. You have the aerodynamics on the other side, right? The studio team calms the requirements or suggestions. The aerodynamics, have to very quickly

Michael Finocchiaro

worse.

Thomas von Tschammer

get those iterations through and try to give some suggestions back to the design team. And you have this back and forth, right? Which is today a major bottleneck. If you want to have a nice looking car, that's aerodynamic, right? And what we're saying to your point, Michael, but then that goes across many more disciplines is that we're bringing those teams together around the same table because now they're able to iterate together in real time and then move from intuition led iterations to data-informed design decisions. And that's a big, big difference. Now they're able to explore these what-if scenarios live together, but then they can make the best trade-offs between aesthetics and aerodynamics. But as you were saying, that goes across disciplines, ultimately, and in the whole company. Essentially, when we talk about digital thread, up to today, digital thread was a big thing, specifically in aerospace, because you wanted to ensure this distrustability. But beyond that,

Michael Finocchiaro

exactly.

Thomas von Tschammer

there was no purpose to digital thread. And this was not used to accelerate any iterations per se, right? What we're seeing with AI here is that this is giving this purpose of acceleration to the whole digital thread. So now you have a real clear incentive to build that layer because now you can use AI on top of this to accelerate your workflows.

Michael Finocchiaro

Great. I think that we still have a problem between our side and the manufacturing side, but that's all another conversation, right? So let's talk a little bit more deeper about the two platforms, the neural concept platform and the end top platform. How is AI embedded in it? You started talking about that Brad, like, you know, there's the co-pilot angle, right? Where you've got the guy, there's, Assistant sitting next to you and you're doing your design, but there's also maybe using fundamental models that could be ⁓ agents, which Tom's just probably going to talk about in how you do workflow inside an application. how have you embedded AI into the actual application?

Brad

You wanna go ahead, Thomas?

Thomas von Tschammer

Yeah, so I mean, for us, AI is at the core of what we built, right, started in 2019. So again, we started not necessarily using LLMs, right, but we started with models able to learn the laws of physics directly in 3D, right, the models that could ingest a 3D geometry file, typically an N-top file, right? We have a cool demo coming up, going with Brad, about this. know, models that could learn directly from 3D and then predict the laws of physics very quickly to speed up these design cycles. Then as I was saying, our company being AI native, being built on the modern tech stack of today, ⁓ using LLMs on top of what we've already built was straight forward and was just a logical next step for us. We've integrated LLM as the second layer of AI, right? That engineers could interact with the platform, the documentation, could speed up their developments on top of this strong base that we already had. ⁓ It's more than ever at the core of what we're doing at RaConcept.

Michael Finocchiaro

but really the DNA of Neuro. How about with INTOP? I think it's a little different, right?

Brad

Yeah, I mean, I was kind of saying in the beginning, like I think of nTop more as like AI infrastructure. And so what I mean by that is like when we started, when I started nTop, the problem we were solving was how do you make a model that can be, how do how do you build a really robust 3d parametric model? So a model that can both capture breadth of design space. So if you have an aircraft, how do you have like a parametric model that can create a wide range of aircraft, but then also be reliable and robust and really, really fast to update. And so what's been interesting with the wave of AIs that, you know, one of the requirements for AI, whether it's building and training a surrogate model that you can use for AI physics, or whether it's an AI copilot helping you construct, edit, tweak, modify, and the kind of more generative AI use cases. one of the requirements that that puts on the model is a model that's very, very robust and very, fast and easy to evaluate. so making the choice early on to build nTop on top of signed distance fields rather than B-Rep, the goal of that was to produce kind of really rapid lights out operational models. And so what we've seen is that our model works really well when connected to a traditional one of the foundational LLMs in terms of editing, tweaking, modifying the model. We also have a product that's called nTop Automate, which is basically the lights out version of nTop. we've built, that can be deployed via container. You can wrap an agent around that that can act as your nTop co-pilot. So if you, as a human, are making edits to an aircraft model and then immediately want to like run thousands of variations of that. You can send that model up to automate, have automate deployed in a cluster of thousands of machines and have it produce and process and generate infinite variation on that model with the performance checking. ⁓ We also have a really robust lower speed CFD solver via nTop fluids, but we have ⁓ really, really solid workflows for going from nTop models into the different physics that you would need for hypersonics applications like with Vulkan and Fun3D. or other applications like STAR CCM, Fluent, et cetera. so the areas where we, like again, I see Entop as a AI enabler rather than like just an AI app. And so all of the tools that are building AI applications we're partnering with or we wanna partner with.

Michael Finocchiaro

Right.

Brad

whether that's on the physics AI side, whether that's on ⁓ the generative side as well. So it's been really exciting to see this. again, there's like a number of new startups in the space, whether they're building kind of generative AI or whether they're building.

Michael Finocchiaro

What do you think? Turns out...

Brad

know, physics AI models like ME AI, physics X, past or these other companies as well. And, know, super excited about the work we're doing with Thomas and the team. You know, we have our end top summit coming up on April 21st. Thomas will be there. We'll be talking about some of the stuff we're doing together. And, you know, if, if you're interested in joining that, can register on our website.

Michael Finocchiaro

Awesome. was actually going to ask that at the end, but that's a good lead into where we can see you guys next. Of course. No, it's going to be good. I wish I could go. I'm too far away from California. do you think that we're going to see the emergence of... Well, let me go back. I would rephrase and say, when do think we're going to have the open AI moment for engineering?

Brad

Well, we can say it again at the end also.

Michael Finocchiaro

When is that pivot for that or the, what's the word, the, well, the tipping point, where's the tipping point where engineering becomes, we have that transformation that as big as the 2022 moment for LLMs, when does that happen for engineering and design? Are we there? Are we not there? Are we a year away? Are we 10 years away? Just wondering what you guys think. It's sort of the end of this section, you know, in terms of. We're now four years into the revolution. We've gone through MCPs. Now we did the open cloud, whatever that was. When do we have the tipping point? When do we have our open eye moment on the engineering side? Sorry, go ahead.

Thomas von Tschammer

Today, I would say today we have the tools to do it, to be there, right? Definitely. What happens is that we, know, as an engineering company, you've been building cars, planes the same way for the past 30 or 40 years, relying on tools based on, you know, the legacy stack, which makes it hard for you and there is inertia to adopt these new tools at scale, right? But today with what we're seeing, there are all the tools that you can bring together. to have these aha moments in engineering design. We have many applications where the company is using the NC Platform to run designs, large design campaigns, explore way beyond what they could do normally. And then engineers, they see the final design and they have to even try and reverse engineer the design to try and understand why it's so much better than what they have been able to do in the past, because it goes beyond anything they could think of. So I would say we are much closer than we think. But we need to bring all the pieces together and in enduring in highly constrained environments, we need to do that carefully, which is why it also takes time.

Brad

Yeah, I I was, I was on site with a mutual customer between nTop and Neural. And it was amazing to see the, like, it was almost like seeing the neurons cross when I saw a demo of how they were using Neural and then how they were using nTop. And the customer was like, wait a second, these need to be connected together and we need to be running these models in the, you know, from the Neural platform and training models and bringing them back into nTop and. creating new ⁓ parameterizations in nTop and running and training and organizing and generating new variations. And I think I completely agree with Thomas in that like the tools are there. I don't think there's gonna be like a chat GPT moment in engineering. It's just the way the nature of engineering, like we have products that are in production today built on processes that were developed in the 70s and the 80s. And I think just like, if you look at how composites entered the aircraft industry, right? Like you wouldn't think twice about designing a new airframe that's a composite structure today. But like 20 years ago, you know, that was a really hard thing to do. And even though we could build composite structures in the eighties, I think was the DARPA program. ⁓ That was.

Michael Finocchiaro

Right.

Thomas von Tschammer

And to your point, Brad, we are seeing a very big difference between what we call the digital native companies. So, know, companies building hardware, but that started, you know, maybe 10, 15 years ago and the legacy ones, right? The pace at which these digital natives are able to adapt to these new tools is very impressive, right? And I can tell you that from what we're seeing in the industry, working with these companies, they are much, much closer than we think ⁓ to a real native experience.

Brad

Yeah. I

Michael Finocchiaro

What?

Brad

I couldn't, couldn't agree more. mean, if you just look at the way like Blue Origin and you look at the way that Spectre, which are both public about how they've been using nTop, how they're using nTop. It's unbelievable what they've been able to do with fairly small teams where like in the past, it would take a hundred engineers to generate a hundred concepts and then a thousand engineers to do the detailing on one. of those concepts, now you could have one engineer produce thousands of concepts and a small team of a couple engineers do the full detailing of that concept.

Michael Finocchiaro

So sort of the possible design space has just gone bigger exponentially, right?

Brad

Yeah, what's amazing is that this is in the US and I mean, I know Thomas is seeing this because of mutual customers that we work with. Like there's this like revolution happening on the defense side where like, you know, for years it was all about consolidation into like the four primes. And in the last couple of years, you've just seen this explosion. We've seen this massive explosion of like new defense startups, which are these, you know, pretty, ⁓

Michael Finocchiaro

Yeah, and your other guys like that,

Brad

I think that's a

Michael Finocchiaro

So like.

Thomas von Tschammer

Yeah, Brad knows that better than me, even better than me, I guess, but Mike would be amazed by today in the engineering industry how designs are mostly led by intuition. It's today, how do we design and how do we iterate? It's an engineer that's based on years of experience and a very skilled engineer will say, okay, this should be how the next design look like to improve the performances. We are still in traditional industries very far away from these data-informed design decisions. But as soon as you're able to unlock that, you unlock a whole new level of performances. Because you go from one iteration, which is based on a static parametrization and a few design variables, to now a space of design option, which is almost infinite. And we're seeing these shifts.

Brad

Yeah. And I mean, I feel almost like I feel like I've seen been seeing the future over the last two weeks of how engineering gets done. Like I've been sitting in one of the things I love the most is like sitting in design reviews with our customers and just seeing how design reviews get done. And I'll kind of give a couple of stories. So like a couple of years ago, I was sitting in a design review with a customer that was just picking up N-TOP and they were working on some component parts integrated into the wing of an aircraft.

Michael Finocchiaro

Nice.

Brad

and they were all sitting around this kind of this Katia model of the component that they wanted to modify. And you had like a couple of subject matter experts in the room arguing for an hour and a half, whether they, how they should route certain tubes. I don't want to say like too much about it, but how they should route tubes through this structure of the wing. And they were basically just arguing for an hour and a half spinning a model in the last. Two weeks, I've sat in design reviews for like a group three drone, so like a thousand pound class drone, a group five drone, so more like a F16 size drone, and then also a hypersonic vehicle. And in all three of those design reviews, you had teams of like the entire design team. So you had aero people, you had structures people, you had propulsion people, you had the weights people, you had the GNC people.

Michael Finocchiaro

Shit.

Brad

And they all have their own responsibilities, right? Like the aero people want to make sure that it's going to fly. The GNC makes sure it can fly stably. The structure people make sure it doesn't fall apart. The propulsion people make sure it can go fast, right? The weights people make sure it's balanced and the CG's in the right place. in all three of these design reviews, they were all sitting around end top models that were, and they were asking just as many questions, but as soon as they would ask questions, people operating nTop would make those changes and immediately they could see the impact of those questions. And so it was really like a live data-driven process. And they brought back a data set from like the night before. And these teams are running like 20 to 50 new revs a day on a model. And in some cases, looking at that data set in Neural and seeing how that can be done. And I think like, as you integrate what Thomas and the team are doing, you now get that additional level of fidelity in terms of, confidence in the predictions as you're making changes to an end top model, immediately seeing like, oh, well, if I, you know, change the fineness ratio of the fuselage, I'm gonna lose some fuel, but I'm gonna have less drag and I'm willing to take that trade in this case, or maybe I'm not, you know? And so I think that to me has been, I feel like I've seen the future and I wanna get this story. out more on how these amazing teams are working because they're moving really, really fast right now.

Michael Finocchiaro

Well, thanks for sharing that. In fact, I feel like I see like a pretty provocative question is, what's holding us back from that tipping point? Is it the legacy software? Is it all that Katia and all that 3D experience in the index and Creo and not enough use of neural and nTop? Or is it more psychological, more people? Like we're used to doing things a certain way and it's going to take a while.

Brad

I don't think it's the legacy. I mean, look, I started nTop because I am obsessed and love CAD software. And so I don't think Katia or NX are the bottleneck. It's just these tools were developed for highly detailed design. Once you know what you're going to make and adding all the attributes, keeping track of stuff in PLM, like the really, really amazing tools for that. There's no better software on the planet. In my opinion, there's no more impactful software on the planet than traditional CAD software and the simulation tools built on top of it. And so I don't think those tools per se are the bottleneck. I wish I could just be like, yeah, all those tools are the bottleneck and everyone should just like switch to nTop and Neural and like that just will save the day. But I don't think that's the case. think it's understanding when to use the right tools. And I think the reality is most of the products that are out there today that new engineering is being done on are not like brand new. And so they've been built and developed on top of new processes. And so the time it takes to learn and integrate a new process, there's the perception that there's not enough ROI to do that. So people are not going to change their processes. And I think the way we've addressed that is set up our design sprint program where we'll send in our field team that can help. produce the outcomes that customers want or show the outcomes that customers want without them having to like train up and learn the new tools. And at the same time, we have a really strong education program. So Matt Mueller is in the other room right over there. He's our education lead. And when, when nTop is taught within the schools and everybody comes out with just like pure native nTop skills and neural skills and CAD skills, I think that's where you'll see the transformation. I don't think there's one, like I said earlier, I don't think it's gonna be like one moment where all of a sudden, holy crap, like the chat GPT moment. I think it's gonna be more like, whoa, that happened really fast. How is every new system starting and fellow? Yeah, exactly. how did that trans, because it's, engineering is such a large and...

Michael Finocchiaro

That's a tipping point. It'd be more retrospective than, okay.

Brad

slow to change, which makes sense. It should be. I don't think I want like to know that the engineers at Boeing are just like trying every single new tool and moving that into production on the plane that I'm flying to LA today. I want that. I want to know that that plane, I want the most boring possible flight.

Michael Finocchiaro

Thank you.

Thomas von Tschammer

Right, think that's also where we have a mission. think you said it, companies is on the education, right? Because we have, as you were saying, I'd love to be able to share with everyone out there what we're seeing on the field, right? Because this is truly impressive, right? But it's also our mission to educate, to upskill these engineers to explain where we believe each tool fits, right? And so we should have, I should have Matt from our team who is our head of engineering and development. Enablements meet with your your game right because I think they have the same mission also based in US To your point as well here. It's it's really a matter of Identifying which tool fits where right one of the questions we are often asked in the initial meetings is Okay, you guys are gonna you know replace simulation It's because why would we need simulation if we have you know a I models that can predict physics right away But actually what we're seeing

Michael Finocchiaro

you

Brad

100%.

Thomas von Tschammer

It's not true, right? Simulation will always stay when you need these high fidelity results to take the final design decisions and tweaks. But you will use simulation in a much smarter way. When you know that you have a much larger freedom to explore, that's where you want to use these accelerated design loops, right? That will then direct you towards the right design to simulate and then to prototype. It's the same shape that we've seen with simulation and prototypes 40 years ago.

Michael Finocchiaro

Okay, so before we shift into the last section, I wanted to ask just a question because the demographics of my audience, I've got.

Brad

Well, I actually, one, sorry, one, one quick question, just to like a, a thing like, you know, is there a world where we do, like, I agree with the shift from like, you know, the shift to using traditional finite element solvers and the shift to using CFD solvers versus wind tunnels. We still use wind tunnels today. It's not like wind tunnels disappeared. You know I mean? And I think that same shift is going to, we're seeing from, you know, if a wind tunnel test, costs a million dollars and a CFD run costs $10,000 to set up and run an AI run, know, once you have, assuming you have the model trained, it might cost a couple of dollars to run. If it's another order of magnitude speed increase, like is there a world where the traditional CFD tools do go away and everything is just AI based and My answer to that is I don't think that's the case because I don't believe that there's like one super AI model that's just gonna like beat everything, but I'm interested Thomas in like if you think there is a world where that stuff goes. Cause people, I'm sure people ask me that also, just the same, same way people ask you that.

Thomas von Tschammer

Yeah, no, I don't believe so, at least in the short term, right? I believe we will still need to rely on simulation as we still rely on prototypes today and wind tunnel, right? So wind tunnel hasn't gone away because we want to do these final checks ⁓ even though we have simulation. There is a very famous sentence that says, you know, all models are wrong, but some are useful, right? I think it's very much applies to CFD, for example, but now also to AI, right? The question is, where do you want to use them? Where should you be using them?

Michael Finocchiaro

Hmm.

Thomas von Tschammer

in your design workflow. But what we're saying for sure is that while it won't replace fully simulation in the short run, it will drastically reduce the need for it because you will not run blindly simulations anymore. But whenever you run simulations now, it's a very well targeted purpose or final validation for a final tweak that you want to run. So you know why you're running it because these are costly simulations. These take time. So every time you do one, you want to do it a purpose. That's the shift we're saying.

Brad

The other thing I love the idea of is like, if you have a neural model set up, you know, if I send some N-top geometry into the neural model, you could tell me how accurate this, does this model fit within the space that this model represents? And if it's outside of the space, just upgrade that to run high fidelity CFD on that model. And now you can't your surrogate model. And so I think that way of working for me is like the

Thomas von Tschammer

Exactly. Exactly.

Brad

the ideal way of

Thomas von Tschammer

Today these loops are automated, right? So it's exactly what's happening, right? So the models, they have a way to tell whether they are within their boundaries of knowledge, if you will, right? And if they don't, we can automatically trigger a high fidelity simulation, recalibrate the model on the fly. But all of that can and should be fully integrated. So that's why we're talking about this mature digital stack that needs to lie under these tools.

Michael Finocchiaro

So I was starting to say that the demographics of the people that watch this podcast, are quite a lot of younger engineers or people that are still, you know, haven't really started their career. And I think there's probably a lot of anxiety about AI is going to my job. I also think that startups like Entop and Niro's Concept would rather the best engineers work for you guys instead of going to some boring job at Meta or Google, right? You'd rather have the best guys working at yours. What should those guys be working, those women and men be working on and how, you know, what kinds of things are you guys looking for in the engineer of the future that's gonna help transform in top and help transform in neural?

Thomas von Tschammer

I think more than ever for the engineers working with us in large engineering organization, you need to work on your fundamentals of physics, on your domain expertise, right? Because to your point, Michael, I think actually there's never been a more exciting times to build hardware today. It's because what we're seeing is that engineers now, they are spending their time on doing fun stuff, know, taking the right design decisions, making the right trade-offs, right? The time where you would... manually set up your simulation, your machine requirements, wait for the simulation is getting over, right? At the time where you need to apply your domain expertise to understand the big picture, the requirements and fit all of that to take design trade-offs is happening. And that's where you need this very, very strong domain expertise, grounded in physics, grounded in engineering. So if anything for the new wave of engineers, work more than ever on the basics. so that you can be empowered using this AI tool.

Brad

I mean, I think there's, there's nothing more powerful than somebody with really good intuition into engineering, working with an AI tool to accelerate them because they're able to see immediately and call BS when there is stuff that's wrong versus just letting the AI run and produce something that's nonsensical. And so we have this internship program over the summer where we bring in interns and their goal is to basically design, build and fly.

Michael Finocchiaro

Hmm.

Brad

an aircraft to a certain set of mission requirements. Matt organized that. We're filled up for this summer. have six interns coming in, starting in a couple of months. But I think for us, it's a mix of understanding the physics, understanding geometry. Like geometry at nTop is really, really important. So being able to understand how you represent and model and create implicit surfaces is critical to

Michael Finocchiaro

Nice.

Brad

and top working and solving more engineering problems.

Michael Finocchiaro

But I suppose that another prereq is that the engine also knows how to leverage AI, right? You've got to the physics.

Brad

Nobody that I don't know one student who is in school or graduated in the last year who is not an AI native person. I've never met. I've never met them. It's like everybody that I talk to interview work with in our program. They're all used to working with AI used to leveraging AI understanding. Now it's not not Not everyone's an expert like the PhDs that are in Fayez's lab at MIT doing, coming up with new transformer models for physics. Like those guys are doing really unique stuff. But like just in general, I think everybody in school right now is already working with AI, at least in engineering schools that we're talking to. mean, Thomas, have you ever met somebody who's just, I mean, we tend to.

Michael Finocchiaro

Thank you. Incredible quick shift in like three years.

Thomas von Tschammer

That never happened.

Brad

Yeah, we hire interns. don't usually, a lot of people that we hire are not like fresh grads. We usually hire people that have been in the industry, but our internship program and our EDU program is where we work with recent grads and yeah. ⁓

Michael Finocchiaro

So you guys are working with pretty bleeding edge companies. At this point, I usually ask the two people, the, average, if you look at digital maturity from a scale of one to five, one being we're still using Excel for almost everything and we're still using email to, I've got these agentic digital twins that are completely autonomous. ⁓ I suppose you guys, mean, most of the people I talk to, usually their customers are between one and two or zero and I'm just, I'm wondering, I'm thinking from what you're saying, a lot of the companies you're working with are pushing past three towards four already, but, or not. So the first question, because there's two parts, the first question is where do you see the customers today? Where do you see them? Are we one, are your customers between one and two or are they pushing past that?

Brad

So good example of this is we're working with a customer and early on they were like, you have to integrate Entop with our custom weights tool. Had a name and you know, through meeting after meeting, they're like, yeah, you got to integrate with this weights tool. Our team has this, this is the tool that like makes us successful. And when we finally were on site integrating Entop with the weights tool, it was

Michael Finocchiaro

Okay.

Brad

basically the biggest spreadsheet I've ever seen with so many worksheets. looked like, like you couldn't read any of the names. was like thousands of worksheets that had like thousands of parameters that you can feed it to come up with the optimal weight. And so I think that's where most customers still are.

Michael Finocchiaro

Of course. It sells. So still at one, one or two. Wow. Is that what you're saying too, Thomas?

Thomas von Tschammer

I would say so even though we see a big spread between different types of industries and companies. I was mentioning the digital natives, which typically have a more horizontal structure, so meaning that engineering teams can take their own decisions on the workflows and tools they're using. So it's less hierarchical, which also allows them to be much more agile. So within these companies, at the team level,

Michael Finocchiaro

Right. Mm-hmm.

Thomas von Tschammer

we are seeing some pretty impressive stuff. It hasn't reached yet a full corporation. It's happening. But ⁓ in those teams, these companies that are typically younger, as we're seeing five, 10 years old, building hardware, we're seeing at the team level, maybe to your scale, Michael, maybe closer to three at the moment.

Michael Finocchiaro

So my thesis is that when, rather than using the big three, if customers decide to go with an end top or a neural instead of ANSYS or whatever, does that move the needle faster? is using your stuff going to really push that? Is there an epiphany? People are using your tools that are AI powered and there's an epiphany like, oh my God, if I was doing data governance correctly, if I broke the, if I break the silos between all these departments, my God, we're going to go so much faster. Is it a catalyst for digital transformation using your tools? Have you seen that?

Thomas von Tschammer

Yeah, so I mean, the first comment is, I would say it's not using NC or to Brad's point and top instead of the traditional tools, right? It's with in conjunction. I think that's very important because you need this mature stack to be there and you will need it, right? However, what we're seeing and that where it becomes very exciting is that we work with companies that are building their digital ecosystem today. Now in this era, they are building it with a purpose, which is this AI purpose, right? We are at the center of discussions with companies where they have to decide on their traditional stack, but because they doing it with AI and NCI at the core of it, they can make it with a purpose and from the beginning being grounded with this AI-driven workflows, which makes it much, much more powerful from the beginning. It's much easier to do it from the beginning, adopting AI at the core of it, than trying to catch up afterwards without thinking about it upfront.

Brad

Yeah, I I kind of second that it's like, it's not about replacing and choosing one stack versus the other stack. It's about choosing the best tools for the best parts of the process. I will say like where I see customers adopting neural, where I see customers adopting end top, they're building much more automated and lights out and kind of AI native workflows because they're able to see the benefits of doing that compared to like if they were to try and deploy like a BRat based CAD tool for rapid design iteration, right? And so I think a lot of people love to like talk about how, you know, the big CAD tools, they were built in the seventies and eighties and it's eighties technology and it doesn't work. Well, you know what? There's like really amazing technology from the eighties that people use every day to solve the hardest engineering problems out there. And it works and it's awesome. Like, Vulkan CFD solver in the U S that's written in Fortran and it's amazing. Right. And so just because something was like developed a long time ago, it doesn't mean it doesn't work and it doesn't mean it's bad per se. Like we're, I'm talking to you on windows and windows has been out since I was a kid. Right. Like, and I love it still. So like, I think it's just knowing where and how to deploy.

Michael Finocchiaro

Right. You

Brad

software like neural, software like nTop, along with the tools that are amazing that we use every day also. And that's why, you know, for us, we have this field team that works really closely with customers to help them parameterize models and nTop because it's a different way of working than you would parameterize a model in a CAD system. The benefit is it's more robust and faster, right? And so, and I know the... this way of working with customers, think, at least since we've, you know, we started with nTop and we were like, okay, we're going to sell you some software. We'll sell you some customer support hours and good luck and go. And people were able to make like little heat exchangers and stuff like that pretty successfully. But now when you're really deploying nTop in an aircraft development program, it's really about how you evolve the process. It's more of a process change. so it... That works really well when we're working really closely with our customers.

Michael Finocchiaro

Awesome.

Thomas von Tschammer

thing that, for example, worked very well for us, last point here is we run this series of boot camps in person. We run, I think, seven or eight of them last year across all the regions. Next was in Detroit on April 30th, where we bring these engineers together in the same room around the NC platform and work together in building these AI workflows for two full days. And that's on this education, showing them the art of the possible. understanding what's the limitation of today's tool, right? But also what these unlocks it. And then we're seeing super interesting conversations following these bootcamps because people realize the potential of it. And then follows up with, you know, deep dives copying session. Okay. How can I now apply this for my next generation car, airplane, engine, you know, whatever.

Michael Finocchiaro

Awesome. Well, thanks, man. That was a really great conversation. I certainly learned a lot. It's a lot of fun. ⁓ We were talking, you mentioned earlier where you guys are going to be, but why don't we go through that again? Like there's the Threaded Live, of course, and Warwick on the 25th of March and Miami on 13th of April, but I believe ⁓ CDFAM and we'll go ahead. Where are you guys going to, where can people see you and meet the Brad and the Thomas over the next couple of months?

Brad

So the next big event for us is coming up April 21st in LA. It's the End Top Summit. ⁓ Thomas and the team will be there also so you can see both of us in one place. So definitely if you're in LA or if you're close to LA or you want to get to LA, you can register on our website, ⁓ which is the spot to be. Then CDDC is the next ⁓ big event, I would say, and that's coming up in July as well.

Michael Finocchiaro

the end. Awesome. And I'll see you at PNM components also in Cambridge, I believe.

Brad

yeah, we'll be at the PLM components and George Allen from our team will also be there.

Michael Finocchiaro

That'd be really cool. Thomas go.

Thomas von Tschammer

Here on our end, we'll be at, I'll be with the team at GTC, NVIDIA GTC in a couple of weeks. So we're exhibiting there and we have a few guys around, if anyone's planning to attend, you should check out the presentation with JLR, Jaguar Land Rover. Pretty cool what those guys have been able to do with us. It's gonna be exciting. Then we also will be at SAE Conference WCX in Detroit, where we've been invited for a keynote mid-April. And then of course,

Michael Finocchiaro

nice.

Thomas von Tschammer

very exciting and top summits. Looking forward to it.

Michael Finocchiaro

That's awesome. can't wait. Well, I hope you guys have a fantastic 2026. Sounds like it's off to an amazing start. Thanks to everybody for listening. had a lot of good participation and we'll look forward to seeing you on the next podcast. Thanks everybody.

Thomas von Tschammer

Thanks everyone.

Brad

Thanks. Yep, thanks everybody. Good to see you Thomas.