So welcome from our side as well. So time to value always matters. Who knows how that feels? Yes. So we know also how that feels. So we have many projects where this is exactly the point of being faster. And that's why we thought we gonna present some insights on our AI rapid prototyping and how we did it together with Union Investment to improve their time-to-value. Okay, so everybody has been introduced. We have Sebastian from the group. He's the group leader of experience design and robotics. And we have Martin, one of our managing product designers. So what makes us talk about time to value? So with our foreign employees, Prodynar basically designs and operates AI based systems for our clients throughout Europe. And we have all these service areas where we basically have the same goal of enabling our clients to make their ideas of AI enabled platforms come true. And to put it a little bit simpler, we basically take the requirements and we want to put out the value stream. And for this, we have basically three pillars or three services that we offer. So it's design, implementation, and of course, managed services. And today, we want to talk about design and implementation, more about design, a little bit less about implementation. And that's why we want to get you familiar with our context and the client, and then also talk about the challenge and the solution. And that's where I can hand over to Sebastian. Thank you, Stefan. As Stefan said, I want to give you some, briefly, some context to the proof of concept we did. Already said, I'm working with Union Investment. We are a big asset manager in Germany. We have about six million customers. We have several products regarding, with respect to financial products, a lot of funds, asset management services, and retirement provision products. Financial products themselves are difficult, and they need a lot of explanation. And with our broad range of products and the high amount of customers, we have a high demand for good content. And this content needs to be delivered fast. So at the moment, the asset management market is under pressure. A lot of new players are coming in. Margins are shrinking. And that's why we need to accelerate our time to market. And this is actually what we are trying to do with this case here. Yes. And as a first step to increase our time to market, we changed our organizational structure. We were coming from a classical project management structure. So the project management and IT delivered. And today, we are having a scaled, agile delivery organization, which basically means we have so-called domain journey teams. Those teams are responsible for shaping the requirements. They are responsible for specific user journeys. And then we have subdomain teams who are responsible for the development and the product and software delivery to the value stream. But this was not enough. We still have long durations from ideation to coding or product delivery. Because you all know that ideation takes time. Designing takes time. We need to technical validate all that stuff. And so we decided to start this cooperation with PRODYNA to especially accelerate the prototyping process, which Martin will explain to you from now. And the context in which we did this prototype is a retail customer website. It contains a lot of information regarding our products. It has several calculators and specific fund information and searches. And that's the context with Martin will go on. Thank you, Sebastian. So actually, before going forward, I'm going to do a step back and really focus on the impact that AI rapid prototyping has reintroduced it at Prodina as a part of a workflow. Really had when we introduced it mid of last year, I think it was. And we immediately saw that it has a huge efficiency impact on especially those three first pillars. Meaning you come to a system that is capable of generating really an interactive experience with an idea or a requirement. And then you're really moving forward in a fast mention. So you're really tackling many, many parts of your X design, which took weeks to create before that approach. And you even can do parts of the technical validation also ahead. Which led us to basically this approach where it says prompt to validate a prototype. And what was obvious, like right from the beginning, is that it's not really a two or three line is a prompt. So that prompt needs to be validated and curated upfront with a lot of expertise. To be able to really create an experience or a prototype that furthermore then also can be validated with stakeholders, but also with real users in the end. And this really compressed to a very short amount of time. So I think after the first learnings, being able to create such a thing after a couple of days, we really went down into one or two days of a compressed workshop format. Where we're able to really gather the requirements, gather the idea, gather the priorities and just pick a few design decisions, right? Up front in the beginning, enrich that prompt and generate something that can be immediately tested and iterated on each device. So there were a lot of learnings to be taken from this one. And in the end, you need also to really consider the technical implications, also the framework. So if you're planning to go to Magnolia or to going to have calculator functionalities, for example, or some certain other pages of functionalities that are specific to your brand, you need to think about those up front. Two more things, we figured out that creating like a validated prototype for greenfield projects was quite easy, because it did not need to put that much experience into it. However, when it came to brownfield projects, like an existing well-grown brand, a lot of expertise that is already there, but not yet curated for agentic use, it becomes more effort like in the beginning of really preparing those assets. And out of this flow, I would like to explain from the right-hand side where we said, we want to create a low-threshold opportunity for product owners in the beginning to really come to a system where they can put in a prompt together with a workshop team. So that after it's been built is really a wow effect. So wow effect not only from a visual standpoint, but also from its contents, from really the expertise, knowledge about products, but also like the structure of the contents. It follows like the brand guidance, it follows like the brand guidance, accessibility, it knows about products already, and it has the structure really to close on interests from potential users that are going to visit this experience. Now, how should we get to this point? We came up with a three-pillar approach, basically to get there, and this is unique. Basically, it's a proprietary chatbot from Union Investment that they developed on themselves. However, it uses like commercial models. It's based on GPT 5.5 in the latest iteration. And in the very first step, what we did was to have an interview agent that is really asking questions, production design questions to a group of workshop members, being the product owner, topic matter experts, UX designers, and so on. So to this group of workshop collaborators, it put questions in the very beginning, but it was also capable of extracting information from already prepared knowledge. So everything that exists, basically, we prepared to this agent already. So product knowledge, context knowledge about the company itself, knowledge about the target groups, accessibility, also information about the framework. So would it be like their sort of design system potentially in the end that it will be displayed in and so on. So within this experience, which took about 40 to 90 minutes, depending on the complexity of the use case, we created a raw answer from this first agent, which was then handed over to the filler agent. The filler agent's responsibility was to really put this raw answer into a structured format, following a section-like master prompt, also covering what is the context, what are the user groups, the user groups, how to really speak to them, what is the tonality, how many user flows are there, and so on. And there were questions also being asked if something could not be answered in the very first round. So it could be enriched again by the group of partitioners in that workflow or in that workshop. And then also the result was handed over to the very last agent in the setup, which is the validation check agent. This agent really started with a clear context. It just became the answer from the previous one and checked it against the structure again. So in terms, the first one got, or the second one got confused and hallucinated. Maybe this one was the one to take it really against the format and do the corrections. With that, we had a readiness score for the prompt that was being put out. So we could be really sure, have we overcome like the 90% readiness for the prompt to proceed or do we need to go one step back to enrich the information still? And with that, we went to FigmaMake. So potentially you ask why FigmaMake in the end. And it's just a tool and it can really be replaced with any other Gen AI tool who is capable of using the latest models and creating the front end. It just was widely available to the union investment product owner, so it was chosen at this point. And it's really a large prompt at the very end, which was put in and within five to six minutes, it was really doing its work and creating really a landing page in the Magnolia style that could be validated really with the stakeholders, with the product owners and actually just test and iterate it even more. So there's an obvious increase in efficiency in this workflow, but I think we learned one more key aspect out of the whole workflow. It's a bit of a preview to tomorrow's session, which we will use to go into much more detail within this flow. And the thing is, it's not only a benefit to efficiency, but also to feedback culture of the team itself. So what we discovered is that once the work is being handed over to like a neutral collaborator of the workshop, being the AI agent, you're really in the position of doing Q&A work now. So without the team, no one is taking like real ownership about his own personal work, but all of them are really interested in increasing the quality and for the outcome in the end, challenging assumptions and being about the quality in the end of the prompt and the output of the system. With that and the next steps, I would like to hand over back to Stefan and hope you will join tomorrow's session as well. Thank you. Yeah. Thank you, Martin. So what we have here is now a real code repository, a code repository that links to the solid design system repository where all the front-end components are stored. And this design system has been also created by Union Investment to serve all their websites. And these two repositories are basically interconnected. And the code that has been generated by Figma Make is basically already applying all the design principles. So the only thing that you need to do is find out how you use that, for example, in Magnolia. So when you follow an agentic software development lifecycle approach and you have your agents ready to talk to you and do the work, you can basically feed in prompts and ask the agents to tear down the page that has been created and generate all the artifacts that you need in your Magnolia system. So being at dialogue structures or components or content apps. So basically the AI is able to take the output, break it down and help you implement the Magnolia part of it. We learned that if you feed them real projects and how you have done it in the past, it's much better, the outcome. So it's again this topic of investing into having the agents really ready to focus on your particular problem in your particular domain. So what are the takeaways? In general, and this is not only connected to this subject that we presented, the key takeaway is number one is never underestimate the design phase, especially when it comes to customer experience systems or platforms, which we all know of. So it's really important to get the design right to know what the customer really wants and not what you think the customer might want. And if you start at the right spot, really at the beginning, you will have a huge benefit increase and value increase in the end after the implementation. So what Martin also showed you and talked about is that the right tooling can accelerate you. So we have been talking of reducing the ideation phase from multiple weeks to just a few days. And with the agentic software development lifecycle approach, it really requires an initial invest. Martin talked about it. We will see more tomorrow. And I guess you will be really amazed how much work and how much context you need to feed in to get really reproducible results in the end for your domain and for your products. And now the good message for all the developers in the room. So for now, what we see is that you still need developers and people who really know their craftsmanship to get the code ready into the final product. So this is still needed. And I think it's quite a few years ahead that the developers are mainly replaced and just the AI is doing the thing. So that's a good message for all of us in the room, I guess. And what we have seen and Martin talked about it, we started with all this AI rapid prototyping approach last year around this time. Maybe it was August. And what we have seen an increase of features and possibilities with all these toolings and interconnecting repositories, writing agents and so on. We see that there will be more disruptions in the future and you need to constantly evolve your setup and your agents and your whole setup to really stay focused and have the capabilities that are state of the art at that time. And as you said, please visit us tomorrow right here on this stage and you will get more input and more insights into the use case together with Martin and Sebastian. And for now, thank you very much for listening. Enjoy the time. Stay cool and see you around.