0:00 Awesome. Thank you, Duann. Thanks everyone for being here today. I hope you weren’t expecting to escape the flood of AI news and content. Cuz my presentation today will touch on how AI is coming into engineering in a a different way. We’ll be talking about how we can accelerate the RFQ process through the use of AI agents when we apply them to FEA oriented workflows. My name is Andrew Sartorelli.

0:27 I’m the head of product management and software partnerships at Synera. I spent my whole professional career working in the field of engineering software. I think I see a few companies I’ve worked with here today in the audience. I see customers from a few different companies here as well. So, it’s a a pleasure to be here in front of you all and talking a bit about what we’re doing at Synera.

0:50 For those of you that haven’t heard about before, we’ve been at CDFAM quite a few times talking about our ideas about process automation for engineering. We’re a German-based company. We’re based in the north of Germany in Bremen. You can see some statistics. We’re a venture funded company here and some of my my colleagues. But actually the the most interesting part for me is on the the lower lefthand corner here.

1:11 The companies that are using Sen today to automate their engineering processes. We’re talking about the major automotive OEMs in Germany, Volkswagen, BMW, MAN making trucks and buses. We have also aerospace companies like Airbus, consumer products companies like MA and maybe a big name most people in the audience know, NASA also using Synera, Ryan Mlullen at CDFAM in New York talking about his vision for Texas to spaceship and it’s trying to achieve that through the use of Senara.

1:45 Now, for the engineers, designers in the audience, right, there’s a lot of challenges we face in day-to-day work. The idea that we can add new people to the workforce, right? It’s a it’s a challenge to hire people. In Germany, you know, we have a expectation that there’s going to be a shortfell of over a million engineers that need to feed our companies in the next decade. And this creates a real problem for companies to achieve the goals that they need to achieve.

2:12 If we think about the productivity pressures, you know, a German automotive OEM, it typically takes them somewhere between five and seven years to design a new automobile. If you look at China, they do it in two to three years. And if Europe European companies want to be competitive, they have to find a way to bridge that gap. And of course, as we get more and more complex products, the expertise that people need to design and develop these products becomes more sophisticated.

2:38 And often times this expertise becomes a bottleneck in a process. So if we think about a typical product team and and what they’re doing, you know, maybe these vacation blocks here are a little bit too small for for those of us here in Europe. Usually, you know, August is the time when everyone’s taking a full month off, but you have a lot of different stakeholders in a project and each one of those individuals need to coordinate with other folks and oftentimes trying to get the time to meet with everyone becomes the biggest challenge to a project moving forward.

3:09 And so this really starts to hinder and delay projects and and drag out timelines that really shouldn’t be that case. If we think back to that situation of experience bottleneck, right? So the ideal scenario is maybe a designer is linked directly with a a simulation analyst. This one:one pairing, but often times it’s not a onetoone pairing. It’s you know we have nine or 10 designers to every one simulation engineer and that one simulation engineer has a huge stack of work that they have to get through.

3:36 But often times it’s even worse than that. You know, there’s probably 30 or 50 designers that are all feeding this one simulation engineer with designs that he has to analyze and get back to them with feedback on. And that creates weeks or months of bottlenecks and delays. And often times design is no longer synced with the the broader engineering organization. So you know, I’m working for a software company.

4:00 What can we do to kind of solve these problems? So at you know you’ve heard us talk before about this idea of you know we’ve created a product to allow engineers designers and folks related to create automated workflows. So we have like Grasshopper we’re very much inspired there to create a visual programming environment for engineers. So we’re allowing them to create these automations and also access the engineering data that they need to create these automated processes.

4:31 And of course, we’re not in this alone. So, we said that we don’t want to reinvent the wheel. As a product manager, I think I’ve been asked four times at three different companies to make orientation workflows for additive manufacturing. It’s a job I never want to do again. And so, we’ve partnered with over 20 different software companies. I see Rushik in the back there from CDS, one of our partners.

4:51 We’re partnered with EOS. We’re partnered with Simscale. David, I don’t know if you’re still here. We’re partnered with 3MF to bring a 3MF connection to our marketplace. The idea here is that we connect to all of the tools that you’re looking for, whether it’s CAD tools from the the big companies like Autodesk, Seammen’s, PTC, to PLM products, from companies like Contact. All of these tools and connections can be accessible inside our marketplace.

5:18 And this allows engineers, allows designers to create automated workflows with the tools that they’re already using today. And for those of you following us, you’ve heard us talking now the last few months about this idea of bringing an agentic layer, AI layer to what we’ve been talking about in the past with automated workflows. And so now what we’ve we’ve done is we are allowing you to use AI agents with access to your company’s knowledge.

5:45 So your enterprise data systems, if you’re talking about a PLM system, if you’re talking about a SPDM system or a requirement management system, you can tie that now in with an AI agent. You can link it with your favorite large language model from OpenAI, from Anthropic, from whatever other organization you’re you’re interested in working with. And we’re also partners with the major hyperscalers like Amazon Web Services.

6:09 So you can deploy agents in your enterprise environment. But what is an AI agent other than the thing that’s flooding all of our LinkedIn feeds? So an AI agent kind of combines two really nice superpowers. So what you have is you start out with a large language model and also you give it memory, the ability to communicate and reason, right? And this means we can start to pass requests to it and we get responses back in natural language which is which is great, right?

6:37 Engineers are great communicators. So of course we’re going to use text to communicate with them. But the other aspect that we have is also access to engineering workflows to engineering knowledge and engineering applications. So when we start to combine this ability to reason and communicate with the ability to interact with a PTC creo autodesk inventor seammens and x excel because of course it’s 2025 and every engineer still uses Excel.

7:06 And now we combine that with a large language model. I think we all know what a large language model is at this point in time. I don’t need to explain that in detail. But why do agents need workflows? What what is this aspect or component that we’re talking about here? Well, if we think about agents, if anyone’s used N8 other products today, you know that large language model, it can work very well with text.

7:29 It can write code for you. It can write Python. It can write C# and all these kinds of things. So if you go off and you ask a large language model, hey, can you go through my inbox and categorize the most important emails for me, it can do that with a pretty decent level of accuracy. But now if you go in and you say, hey, can you go work on this FEA model for me?

7:50 Can you update my CAD geometry? Larger language models have no context or understanding of engineering data of engineering knowledge of engineering knowhow because they don’t have any context of NX or Alter Hyperworks or any of these other products and so what we need to do is empower AI agents with workflows and that’s that’s what we’ve been talking about what we’ve been working on at Synera for a long time now is creating automated workflows and now we can start to allow these agents to understand and have engineering context and interact with engineering data in a much more useful way.

8:26 So what are the types of problems that you can start to apply agents for in engineering? If we think about the you know traditional rule-based approach to auto automation, what we have is the if we have a legend here of task fuzziness and autonomy, you know, rule-based automations that we’re we’re used to. They’re very rigid. They’re very brittle. You can define it for a very specific task.

8:49 That gets slightly different inputs, the whole automation breaks. We can start to add in AI to those those workflows and the ability to handle slightly different inputs, slightly different maybe output requests, it starts to get a little bit better. But now when we start to bring in AI agents to the the capability map here, its ability to work with different inputs to handle situations it hasn’t seen before is much greater.

9:17 And its ability to reason on its own is even its own benefit that brings us to a new level of capability. But at scenario, we don’t talk about just single agents. That’s, you know, last month’s news. We talk about multi- aent systems. This ability that we can start to create a series of independent agents that work together and collaborate together like an engineering organization. And so if we think back to some of the problems I talked about a few minutes ago, this this idea that, you know, my team has vacations and people need to take time off and it really draws out my my project schedule.

9:54 If I think about if we bring agents into the loop here now, as soon as the first agent finishes a task, the next agent can start to immediately work on the next aspect of work that they have to do. And so this can really start to condense down those timelines. When you think about responding to an RFQ, no longer does it take weeks or months to organize a meeting because my AI agents can do it for me while I’m on holiday in the south of France.

10:20 If we think back to the the situation here with the analyst being overwhelmed by all of these designers feeding them results, well, if I’m an analyst and I start to create my own agents, well, I have one assistant, but there’s nothing stopping me from scaling up my organization to have two assistants or six assistants or 10 assistants that allow me to respond to my team in real time with the knowledge of my best analyst while my designers are fed immediate results.

10:48 In the end. So I’ve I’ve talked about a lot of slides you know I’ve given some things but how does this actually look like in real life? What is what is do we actually have a product here? Right. And so I’ll show a a video now of a demonstration of let’s see if we can get it to play how we’re going to apply this for an injection molding use case.

11:11 You can see this is that you may already know. We’ve got some new nodes here. These are agent nodes. So we’ve got a manager or supervisor here. We’ve got, you know, if you’re familiar with FBA, we got to do some defeaturing. So we have a defeaturing agent here. We can see it’s linked to my large language models. We’ve got a midsurfacing agent, so I can generate the appropriate mesh for my injection molding analysis.

11:32 And I’ve got a simulation agent. All of these agents are empowered with rules and and definitions. So we can see I’ve given it some responsibilities. I’ve given it some goals. I’ve given it some rules. And it’s empowered with tools. These tools are the workflows we’ve already created inside scenario. And we can tell the agent how we want to use these tools. And so once we’ve created our agentic system, we can go off and publish that on the web.

12:01 And we can see here the the the interface. As I said, engineers, great communicators. So we we can use text and of course get the right results. But now we have an understanding of of okay, here’s my agentic system. I’m going to describe for you what I can do and I can give it a text prompt. I’m going to provide it with some geometry and say you know what can you do for me with this geometry?

12:21 It’s a plastic part. Of course, it would be injection molded. And so now it tells you, okay, I’m going to have the defing agent start to look at this and see how we can simplify the geometry. I’m going to have the midsurfacing agent also work. We can see the agent goes off and running. We’ll see a nice little audit trail here of of how these agents are working together.

12:47 So, I’m not like David. This is pre-recorded not a live demo. But now we can see we’ve got a response back from our agent, right? And we can go through that as the engineer. Of course, we have to understand what our agents did. What were the inputs and outputs that they started to use? And we can fully understand that with this audit trail here. Now, we’re engineers, so we want to see we want to see outcomes, right?

13:09 We don’t want to see just text. And now we’re going to be able to go down and we’re going to actually select the outputs that we got here. So, we got some step files out. And now we’re going to say, okay, let’s go off and start an injection molding simulation. I’m a fan of mold flow. So, we’re going to go off and start a mold flow analysis.

13:26 We can say actually here we’re going to use a single injection point. The system’s going to know, okay, one injection point for this part. The next agent goes off and runs. And of course, we can download our MOFFlow results. Open it up inside MFlow and we can actually see the results we did here. For those of you that are injection molding experts, maybe not the best injection point that we selected here.

13:52 And we could probably prompt the agent to pick a better one. But just to see, we can see the fiber orientations. We can see the pressure in the mold here. We’re going to go back and say actually try a few different injection points here because this isn’t the ideal situation. Agent goes off and runs again. Gives us a little bit of feedback and we can go off and now we can see new mobile flow results, more injection points.

14:15 Again, probably not the best injection points and we need to ask our agent to do a little bit of a better job there. But now we’re going to do a drop test. We’re going to connect to LSDina where we’re going to actually see if this part can withstand from being dropped off the stage. And so in the end, LSD goes off and runs in the background with her workflow and we get a nice little animation of our automotive component being dropped.

14:43 We could prompt the agent to go off and change a few other parameters, change the orientation of the drop, these sort of thing. But hopefully you get an idea of how an agentic system can start to approach different engineering problems and replicate the engineering organization. Now it doesn’t just have to replicate the engineering organization. We can also replicate other organizations you know so we mentioned at the start RFQ RFQ is kind of a related adjacent topic to engineering and so here again we see another agentic system that we’ve created I’ve sped up the video a little bit here but we involved a commodities analyst we involved a costing engineer we involve a CAD designer in this process and again we get a nice little prompt of what my system can do what my agent authentic system can do because we’ve defined all of those ahead of time.

15:35 And so the nice thing here is you’ve got a bunch of engineers that have created your your system, defined the constraints of the agent, and now your salesperson that’s in the field in real time while they’re sitting with a customer can go off and talk with this agentic system and get feedback about how much it’s going to cost to manufacture a filter assembly. So we’ll say I want a thousand parts.

15:58 Here’s my file. And now we have the agents going off and running and and doing their own thing without any without any other user involvement here. And so it starts to understand what it needs to do to generate a bill of materials and how it’s going to assemble the components together. And in the end, it’s going to come back to me and it’s going to give me a nice little report and it’s going to tell me it cost this much to manufacture my component.

16:29 I’m going to estimate you can manufacture it in these different locations. We’ll give it a minute here to fully respond and we can see the report now in the the chat window. Now, at this point in time, you know, the the benefit of an agentic system is I can now prompt it to change things slightly here. So, I could ask it, are there any lowerc cost alternatives?

16:56 It’s going to go and it’s only going to run the aspects of the the process that it needs to run unlike a traditional automated workflow. It’s not going to run the full process. It’s not going to have to go through a 5h hour workflow here to get one change in the system. We can prompt the system and it’s only going to update certain aspects of it. We can prompt it to change the amount of components.

17:18 And it’s also going to only adjust slightly there. So, short and sweet. Hopefully not too much AI for everyone on, a Thursday morning, but, hopefully you all take away a little bit of an understanding of how AI is coming to engineering different ways. Not just the traditional approach of reduced order modeling or speeding up simulation times, but actually how we can speed up the overall engineering process for different organizations. Thanks a lot. To learn more about the CDFAM computational design symposium series, to see the archives of previous presentations, and to learn about future events, visit CDFAM.com.