0:00 Who is it? That’s all right. Just test. So, good morning everyone and welcome to Barcelona. Barcelona is my city. And we will see it now how we are going to to transform the definition of the most of the processes in the body white development. I work in SEAT. Maybe some one of you So, this is the quick introduction of what is SEAT for those who who doesn’t know what is SEAT.

0:37 We are SEAT company who produce SEAT and Cupra R and Cupra models as this one Cupra R that is going to present by the way tomorrow as a world premiere in Barcelona and many other cities in the world. To produce, to develop, and to design car like this, we have proceed big data, we have proceed thousand of vehicle advance, we have keep on the aerodynamics, on the aeroacoustics, on the stiffness, on the crash.

1:04 And this is something that is has been a challenge. We know that we have to face important challenge in the future or maybe the present. I have divided two group of challenges. The first group of challenges is faster and better. We know that in the past, especially in the body white development, my area, we have worked on always sequentially from the beginning, the design or the start, the car, the simulation, at the end, the test, and the production.

1:36 So, this is going not to be more useful in the future. The future will have to change the things to do. And this is why we have already performed with the Cupra R, we have reduced the development time, and we have keep up with the industrialization time. But we know that in the future, if you want to stay competitive in this automotive mobility industry, we have to do the things even faster in the development time and maybe also in the digitalization.

1:43 This is why we have to face this important challenge. But also at the same time it’s not only to run, it’s also to do the things better. Better in terms of the weight, better in terms of the performance in general of the car. I will not meant 100 of variables, of course. This is the first set of challenge. Set of challenge is we have to we want to create concept cars like this in the future.

2:33 This is what we call organizational challenge. In the big companies we are always maybe also in some of the yours we are always connecting people with teams, with chats, with mails, and so on. So, we have to somehow break silos. This is an important challenge, also. And of course at the same time there is no job protection. We know that every every time we have changes position, job rotations, people that maybe change in the companies, new challenge, and so on.

3:07 So, this is something that we have to keep also because of the demography of our companies. For all of these challenge, group one and group two, is where we have a journey together with Cinera. Cinera is a lot of things, but before to continue maybe Tilmann you want to introduce what is Cinera, please. Yeah, thanks. I can do maybe you we can go to the next slide the second.

3:33 Yeah, you can keep it. So, I will deep dive on that later also to explain a little further what it is about Cinera. Maybe just to set the stage right here, we are an AI engineering platform specifically meant for engineering. Therefore, we have basically three different layers. Bottom and foundation, so to say, is the connection to many different engineering-related tools. They’re coming from from the CAD side, from the CAE side, for simulation, also for accessing data, accessing many different other tools coming from PLM and and and.

4:10 Middle layer, basically our own yeah, low-code language to automate a specific engineering task. So, we have built our own programming language with a visual foundation behind that and a lot of predefined tools, predefined templates to build these kind of automations. And on top, you can also see a layer which we call agent-centric framework specifically for engineering. This specific engineering framework is built to use this huge foundation underneath with all the connection to the common tools and the connection to all these automations which we can build with our engineering low-code language.

4:56 I will deep dive on that later. Back to Juan at this point for Yeah, see what’s what’s he is doing on this. Thank you for this introduction. Then, regarding the flows, connecting people, connecting processes. We have this journey started last year with the Sineta. And not agentic. This is not agentic. We created a few use cases where where we reduce the development time from 2 hours to 5 minutes.

5:20 Another one Another second use case on 20 42 hours to 1 hour. And many others I will not repeat. So, this is from the 2025. It’s important, again, the reduction of the time, but also the precision and also the better function, the quality in general. So, many of these use cases are already in production. We are breaking the silos we have commented before. And we are not yet implementing this into the genetic model.

5:53 This is This This is going to be the second part of the presentation. So then I will pass I think now again to to Tilmann. Thanks a lot Juan de N. Station. So coming from these awesome products and awesome figures also. Let me jump to There’s one slide from from Europe but okay it’s good. So starting with a statement on this point. So as you have seen these kind of products like vehicles and engineering vehicles, building vehicles is huge and a complex process to do all these kind of bits and pieces, bring everything together until everything works in this in the way we want to have it.

6:37 Problems which we are facing on the market at the moment is that we have to significantly significantly reduce the time of the market here. And not just in using time, we also have to take care on using less resources of all the human resources, less less resources on the product itself, and saving money, saving saving budget, saving all the these kind of things which are important to be competitive on the market in the end.

7:07 So what we see at our customers all over the engineering domain is that at least 50% of the engineering time will vanish over the next few years. So this is basically an important point which I want to highlight at this at this at the moment that we facing huge changes on the engineering market on the engineering domain and we have to find ways, find processes to be faster and to be more efficient in our development.

7:41 Maybe starting with this with bit of a question. What if a digital coworker can handle also engineering tasks? And I just want to raise a few questions here and hoping for a little bit of a hand sign if this somehow fits fits to you. So thinking about agentic and agents in our daily business, I think basically all of you will use things like yeah, an agent who is taking care of meeting minutes or answering emails or these all these kind of things.

8:16 What is just interesting for me, who is using these kind of agent in his daily daily business? Just be maybe raise your hand if this is somehow common for you already. Nice, thanks you. Already a lot of people and I would expect if I would have asked this question a year ago, the answers would have looked much different. And the same thing I recognize in the moment with all the agentic topics coming to engineering.

8:46 And so I want to ask another question, who is using agents when it comes to topics like adjustment of a pet file or running a fee analysis or all these kind of things. So maybe another hand sign or that. Okay, but basically this is what I would expect on this question. But I also see that if we go one year into the future, this again will look much different.

9:13 And this is exactly what I want to talk about in the next few slides. What if we would be able to build a multi-agent system who is taking care which is taking care on engineering specific topics like if we want to build an awesome product like Juan was was showing and we have many different teams, many different tasks which we have to tackle all over the development of process.

9:39 We have so many meetings which we have as you would do with all our colleagues who have so many many things to discuss, so many things to decide. If we have proper co-workers to with a lot of knowledge behind us to tackle these kind of questions which are blocking us on being fast and being efficient, wouldn’t that be an extreme boost for efficiency all over the European and yeah, let’s say international market when comes to engineering and development processes.

10:13 I just want to explain how this is built. I will come to a real world world example in the end, but just to make clear what I mean when I talk about agents. In general, it is just starting always with the brain. So, our agents have a brain, have to have access to an LLM. So, the LLM is so so to say the the core of the overall agent that it has to be that it has the capability about thinking, so to say.

10:42 Next thing is if we want to tackle problems related to our company, related to our domain, we have to add also context and let’s say company knowledge, so basically a guideline around what’s the agent, how the agent has to has to act in the end. And last point and I would say the importance for my presentation in a moment, we have to add engineering knowledge. And engineering automations.

11:09 And this is what our In the beginning it was the middle layer, what our programming programming language comes comes in that we can basically build automations for engineering specific tasks. And this is what I want to a little bit explain in the next video. So, this is just a little bit of an animation how a workflow in our case is looking like. So, we have we can build all our we have many predefined templates, many predefined building blocks to create automation.

11:41 So, for example, we can upload CAD files, we can read CAD files, we can adjust CAD files, we can do some meshing, and the important thing is we can also access to many different to many different tools on the market. So, if we just want to mesh our component with Ansys or we want to mesh it with some other other mesh product or whatever. We have a lot of connectors to exactly do what your company is supposed to do with the software solution which is available in your company.

12:15 If I then jump basically back to what is an agent, this is exactly what we are tackling inside Synapse. We want to build agents for specific tasks in the engineering domain. Like for example, we want to build an agent specifically for CAD topics. Then we can basically build workflows behind ex- accessing the CAD files, accessing CAD software. Same for for the simulation world. If we want to build something for automating an FEA analysis, we can connect to all these specific tools there also.

12:46 And same for PLM and data access. Coming to let’s say a little bit to We had a lot of fancy pictures now. Coming to real world example. What we built together with SEAT that I’m really proud of. We built a manufacturing analysis agent system which is capable on analyzing components. In this In this case, it’s about sheet metal components. And the overall task is to be fast in the initial validation.

13:20 So, we want to early in our development process know if our our component is really performing in the way that we want to have it just on the on the mechanical side but also from manufacturing side it is important to know is my component in the end manufacturable or not. And this is exactly what’s what the system here overall is is is meant for and I just want to focus on one specification.

13:47 This one is meant for for simulation. This we see the exact same like on the fancy picture in the beginning. We have to add we have to add us an LLM so we can choose what is available in the company like it could be a GPT or maybe an anthropic model or whatever is available on this on the specific server. We have to add guide rates so a specific prompting how the LLM should be used.

14:13 A little bit the same way like it would be on a on a meeting minutes agent so you have to add how it should interact with all the data he receives in the end. And we have to also add us workflow so the the animation I was showing in the beginning with the few blocks you can build and then put together. This is basically what we have on the screen here.

14:39 We have for example one workflow which is taking care on the forming simulation and another another one which is taking care on material library for example. After we have built all these kind of things we just have to hit publish button and the workflow or the overall multi-agent system will be published in the in the in the in the company and everybody with access is able to use And how would this look like then in the end?

15:08 It’s a typical chat window so to say like like on GPT like on on a co-pilot whatever. You have a chat window where you can basically just upload a component so to say so the CAD designer has just created something and the system will always guide you through. So, just upload a component ask a specific question. In this case, this is about about do me do me a manufacturing analysis with a specific material, whatever.

15:33 Just prompt this, hit hit run and then the system will in the background all its different agents. So, the agents which we have built in the beginning will be automatically autonomously called by just the overall multi-agent system. So, there’s a supervisor which is calling and orchestrating all the agents underneath. And so, for example, the first one was something related to geometry. So, we have extracted data like wall thicknesses, like plating size and whatever is interesting for forming simulation.

16:07 And the agent which is running in the moment is just about simulation. So, this is this is the performing a full stack forming simulation in the background with using Altair tools and using proper meshing, using proper software solutions which are common common on the market. After an amount of time, I will still receive a report and recognize, for example, there’s somehow an issue in the component. So, for example, not manufacturable.

16:36 I will also receive pictures where is the problem in my component. Then I can jump back to Catia and say, “Okay, there’s an issue. I have to solve it.” Maybe in the end, we will also have automations for automatically solving these problems. In the moment, this is a manual step. So, we jump to Catia, we change the geometry, reload it again to the system and maybe also ask a specific another question in terms of maybe also do me a costing analysis or do me, for example, a CO2 investigation for the specific component.

17:08 And then something interesting happens. The system recognizes that there’s data missing. So, when it comes to when it comes to costing or when it comes to CO2, there’s a lot of data needed like material like sourcing country, how many components has it been produced and then end. And you have to basically upload this kind of data. You have to tell the system what he should should do in the end.

17:32 So, there is no basically we don’t have a fear of of hallucinating at this point cuz there’s always this rule-based automation in the background which is taking care on all the bits and pieces and being exact on the output. Means in the end you will receive an feedback again from our multi-agent system. Component is manufacturable so there is for example something like final check then component will be produced in somewhere and then and all the other informations will be produced and exported from the from the from the supervisor of the multi-agent system automatically.

18:15 Just a little bit of of references cuz I’m so as I said really proud that we did this together with with SEAT and build a lot of things here together. We’re tackling something like four weeks in two 30 minutes. So so this is this is a huge time saving and a huge potential to to bring down the time to market of developing components, developing products. Also a lot of other companies are we are working together with which is also always interesting to see and yeah maybe coming coming to an end.

18:53 Thanks a lot to everyone that we are allowed to be here. Yeah, we will be still on the event so if there are questions just yeah feel free. Cafe or beer or Cafe or beer. And also my colleagues are somewhere on the on the on the event. If you see someone with this logo, just reach out to him. So, thanks a lot. To learn more about the CDFAM computational design symposium, access the archive of previous presentations, interviews with speakers, and information about future events around the world, visit CDFAM.com.