Name: Virtual User Group
Uploaded: 2026-01-28T11:30:03.103Z
Duration: 1 h 30 min
Description: Virtual User Group

Transcript for "Virtual User Group": Hello, everyone, and welcome. Thanks for joining today's SnapLogic virtual user group. We'll get started in a few minutes. And as you're joining, feel free to introduce yourself in the chat. I'll just give it a couple of more minutes to allow people to join in. Jim. Hello, guy. Welcome. Perfect. Give it one more minute before we start. Everyone great. I think that we can get going. So, hi, everyone, and welcome. My name is Maria Hercules. I am the director of customer success in EMEA here at SnapLogic. I'm based in London, and I'll be your moderator today. Yeah. So this is the first user group for the year. These sessions are designed specifically for customers, and the goal is always the same, practical insights, real examples, and things that you can actually take away and use. So we'll start looking at the agenda. We will start with an overview of the SnapLogic as the MCP fabric, then we'll move into a practical MCP demo, and then it'll be followed by an agent creates a success story from one of our lovely customers, Rabantage, and then we're gonna make some time towards the end for q and a. We've got just under an hour together. We've got about forty to forty five minutes worth of content and time towards the end for questions. So as we go through, please feel free to drop questions in the chat. There you You'll find your chat or the q and a towards the top right of your screen. And so with that, I'm gonna hand it over to Dominic. Thanks, Maria. Hi, everyone. Welcome to the user group. So my name is Dominic Wellington. I'm the director of product marketing for data and AI at SnapLogic. Despite the name, I'm actually based in Italy. So this is probably European group that we have today from SnapLogic's side. So today, I wanted to talk about what SnapLogic is doing regarding one of the exciting new topics in the AI space, which is MCP or the model context protocol. If you haven't heard that acronym, then I'll be quite surprised that it seems like every single article I read on LinkedIn is about MCP these days. But let's get into why that matters. So towards the end of last year, we at SnapLogic ran a survey asking people about how they were using AI, and, the results were interesting. Very many people many very many of the respondents of survey are already using AI in their role. 50% are even already using AI agents. So the pool may be slightly biased by the fact that these are SnapLogic users or at least people aware of SnapLogic, so perhaps more oriented towards data work. But still, those are pretty high numbers. Employees are interacting with AI almost every day of the week, and they report that they're already saving more than three hours on work related tasks. So all well and good. But despite this strong adoption, 94% of the respondents reported barriers to effective AI usage in their organization. And lots of the problems boil down to issues with data privacy and security concerns. So people are using AI more in their personal work than companies adopting it in a structured way. And so, that's the aspect that we think is leaving a lot of value on the table. What we see is that, as Gaurav Dhillon, our founder says, AI projects die of data dehydration. And this is backed up by any number of reports, you know, pick whichever analyst firm, university, etcetera, you care to name. But the results tend to converge on the fact that there's a lot of money being spent on AI, but there isn't yet the return that companies have been expecting. And our thesis is that the reason for that is it's these are very easy projects to to get started to show a cool demo in the lab in an isolated environment, but it's very hard to get it deeply integrated with all of the data, the apps, the APIs, the services, the things that you use to run your business today. And luckily, SnapLogic is in the business of doing precisely that. So that's why we think that we can fix this gap between the the promise and the investment that's being made in AI and the concrete returns and results that we expect from it by putting in place that connected fabric using SnapLogic and specifically adopting the the model context protocol via SnapLogic. So the reason for that situation with AI dying of dehydration and IT not trusting it is that there's just myriad of isolated systems that individual teams have adopted over time to do one specific task or one specific part of the overall work. These might be running on premises. They might be SaaS tools. They might be all sorts of homegrown things. They use internal data, they use third party data, they use by all sorts of different teams in different ways. And when there's a task, which honestly is most tasks, that needs to span multiple tools, multiple teams. One team needs to talk to another. A team needs data from different systems. People have built point to point integrations over time using all sorts of different techniques, different tools, different standards. And so it's very, very hard to get that overall view if you're starting from that point of what's sometimes called the spaghetti integration and going in every direction. If we use SnapLogic on the other hand, and we as SnapLogic users all know that we have the ability to connect across all of these systems to connect anything to everything, to rationalize all of these different connections while satisfying the needs of all of those different teams, all of those different use cases, all of those user bases. And AI agents are only the latest take on that type of integration. So they have the same needs in terms of universal access, in terms of security, in terms of compliance that any other use case had. So through the SnapLogic platform, as SnapLogic users and professionals, we are well placed to help our colleagues and the the various other teams in the business to achieve their goals. And the reason for that is because we have this agentic integration platform that is capable of talking to all of these different systems through one single platform. So maybe you put SnapLogic in place years ago. I've seen some of the names in the charts, some longtime customers for one particular use case. But then as more use cases and more use cases emerge, you've been able to to broaden the adoption to extend the coverage to more and more systems, more and more use cases from classic data integration through application and API management, and now into AI agents, but always thanks that same solid platform with all of the pre built snaps that can connect to anything, the low code, no code development environment, the built in SnapGPT integration, Copilot, and so on and so forth. And that's what enables the agents to to do their work. AI agents have the same needs that any other use case has. They need to connect and orchestrate across different enterprise resources, data sources, apps, services, etcetera, in order to do their job safely and securely. Agents are slightly different from many of these because they're not as deterministic. If you write a script or something like that, it will always do the same thing. If you have sets of agents doing their job, they work in a rather different way. So they're given some goals, some objectives that they need to achieve, and then there's a question of orchestration across large language models, across the various tools that are available, the memory the agents use to keep context within a specific interaction and across multiple agents. But then underneath, they're interacting with the same resources that we mentioned before, the same data, the same apps, same APIs, and potentially the other agents throughout the environment. And so that's what enables us to do that work, to provide the data that prevents the AI agents from dying of dehydration that enables them to deliver that value. Whereas we have all of the snaps, all of the connectors, that gives access to both structured data, the classic, you know, data in a database rows and columns in a table, but also unstructured data. So I have an s three bucket full of PDFs and I need to do something with it. So through SnapLogic Agent Creator, we have support for all of these different types of data that we can import, we can ingest, we can digitize and cleanse the unstructured data. Often the problem there is that it's not by definition well structured as immediately suitable for use. We can use vector databases to do things like RAG, and we can do retrieval augmented generation. So we're providing a ground source of truth to the larger language models, the AIs to reduce or eliminate the hallucinations and focus their responses on the data that we provided. And so we can do all of these things just within SnapLogic agent creator. We can build all of these experiences using the tools and techniques and the common platform that we have available to us. But not all agents in the environment in the enterprise IT environment are built with SnapLogic agent greater. People are building agents, as we found in that survey, using all sorts of tools, using a variety of LLNs and so on. And this is a problem for for the industry because it's very hard to get an overall view. There is no one platform that everyone is building their agents in. Agents are springing up everywhere. And so towards the middle of last year, originally, it was OpenAI that came up with the model context protocol. It's now been open source. It's been handed over to the Open Source Foundation. So this is a way that agents can interact with tools, with resources that are hosted elsewhere so they can communicate with existing systems in the environment. So this is very important because it means that the AI agent is no longer isolated running in a box. It is talking to the wider world. It has access to all of that data. So in a SnapLogic terms, what that looks like in practice is that SnapLogic can act as what is called an MCP client. So if you build an agent in or if you any pipeline, in fact, any SnapLogic pipeline has the ability to call out to MCP servers running elsewhere outside the SnapLogic platform. The third party MCP marketplace, mcp.so, lists over 16,000 public MCP servers, and that's probably an undercount because many of these servers run inside the corporate firewall, so they're not visible in third party marketplaces like that. So there's any number of systems that's already MCP enabled that are offering MCP tools and resources through a hosted MCP server. And now all of that is instantly accessible to you as SnapLogic users or SnapLogic developers with the new MCP SNAP PAC that just recently became generally available. So if you haven't taken a look at that yet, then it's it's well worth taking a quick look because it instantly brings enormous capabilities within arm's reach. But what's coming is perhaps even more interesting when SnapLogic gains the ability to act as an MCP server. So this is a capability that's currently in preview. It's is expected to become generally available within the next short while. I'm not going to make a commitment on a public webinar, but it's imminent. We already have tons of users working with this in their own environments, so we've not found any showstoppers. We expect this to go live shortly. And where this becomes interesting is it means that people working in all of the the various AI agentic development environments, if you're working in cloud desktop, if you're working in RUE code, if you're using any of these tools, these have the ability to act as MCP clients, and then you can set up your SnapLogic pipeline as an MCP server. And that is interesting because that means that your AI agent now has access via SnapLogic to all of these systems that SnapLogic can talk to, which is pretty much everything within the enterprise IT environment. I put some logos here on the side of, common run the business tools, but it can be literally anything that your SnapLogic is talking to today now comes in the scope of your AI initiatives even beyond the SnapLogic platform. So that becomes potentially very, very interesting to extend the scope and make AI real in that way and start delivering some real results. So to recap, the reason why SnapLogic, which after all was born as a data integration pipeline long before the current wave of interest in Gen AI, should be key to your AI strategy is because it has these three characteristics as the universal connectivity, a thousand plus pre built snaps, over a 100 that come with the agent creator bundle that is specific to AI use cases. It makes it easy to build AI experiences directly within SnapLogic, but also via MCP to offer up that data fabric to AI agents built elsewhere wherever they're running. With the data transformation and preparation, we can take the data, we can format it, we can prepare it, we can do whatever we need to make it accessible to the AI model. If you have some really legacy business tool that's doing its job and it's not been changed in a long while because it just works, you don't need to go and change that to make it accessible to AI. SnapLogic is designed to do just that, to give it the access and transform the data and format it so that even the crustiest part of your business tool is available instantly to AI agents. And finally, with the agentic orchestration, you have the ability to use a whole set of agent creator capabilities to be able to not just build agents within SnapLogic, but also observe and manage them. And there will be more to come on that front as well. We're working very hard on that front. You'll receive invitation soon to our springtime agent fest event, where we'll be talking more about this orchestration capability, especially once it starts to span beyond the the SnapLogic platform on its own. So with that, please do ask questions in the chat. I'll be around all day, but I will hand over to my colleague, Constantin, to show you what this looks like in practice in a real world environment. Great. Thank you, Dominic, and, welcome all also from my side. I will walk you through our demo scenario we set up today, and we will jump into the shoes of Phil. Phil is a really kind of frustrated shop manager for a retail company. He needs to take care of all the staff planning for the various shops in his region, and he has several challenges he needs to face. And and one of those is that all the data is stored in a legacy mainframe application, And all he has in his daily work is some exports for in CSV. He needs to work with and and try to get everything together. The shops he has to manage are also quite quite big, each having more than 20 employees to assign. They have different kind of roles. They work different hours. And also, the shop needs to be staffed differently depending on the time of the day. So in the morning, typically, less people are in the shop than in the afternoon hours, and all this needs to be taken care of. Also, there's short term challenges, like somebody gets sick or needs special, time off for a certain period, and this, makes it even more challenging for him. Also, he needs to take care that every everybody is kind of working his hours. It's the work is evenly distributed, and that nobody is doing too many overtime shifts as they are also quite costly for the business. And, yeah, let's see how we can help him and actually connect to his data and make it available for MCP so we can even make an legacy system AI ready. So with that, let me stop the stop the sharing of the actual slides and get over to my screen. If you have any questions in the meantime, please put them in the chat. And we will start in the fantastic tool of ChatGPT. In here, we will start with the first end user experience. So I'm Phil. I will jump into his role. And before we go too deep, I will quickly give you an indication how this is set up. So in ChatGPT, we have added our MCP server. It's called retail shop planning. And you see down here, it's connected to a SnapLogic address. So it's a MCP server we share from our control plane, and it has various different tools available. We can get available staff for the week in the store. We can get the employee data, the availability. The opening requirements of the shop, staff planning gaps, we can identify, then do the requirements analysis, get the store data, get the time off, and so on. So all these kind of details are required for his daily work. And so let's jump in. Now we have Phil, and we want to manage our shops. And we can ask after we have activated our MCP server for the chat, what shops can I manage? And based on the request, JetGPT is basically defining and selecting the best tool for the job based on the MCP description we have provided. So if we click here on call tools, we see currently it's getting the get stores. And we have the tool description coming from SnapLogic. We have no tool input in this case, and we get some results back. And and is making some sense out of it and lists me all the stores I can use and manage. So in this case, I want to double down on the store in Hamburg with the ID 1003. So I have a question. Let's focus on Store 1003. Give me an overview of the opening hours and the available staff for the store grouped by their role. Now I send this off, and you will notice in a second that, basically, once I want to call any tool or the tool call is required, which involves some personal info or could involve some personal information, I need to confirm it here in in chat GPT that the actual tool calling is should be done. And this is, for example, the case for the employee list for shop or store 1,003 as it includes some first names, last names, hourly wages, and so on. So I need to approve that, and then it will go on. And, also, this is just one part of the whole planning as over time, it can get even a different kind of tools, calls. Our agent here in ChatGPT will make sense and and reason by itself and just use the data we made available on demand based on the requests. So each time we want to get some stores, employee data, we will call SnapLogic and execute the actual pipelines underneath to retrieve the data. For example, here, our opening hours, so Monday to Friday from nine to to 9PM, basically. Saturday, same hours. And then we have our staff available. We have seven cashiers. We have 11 sales associates. So you see also the different times they work. We have our stock associates who are helping in the in the logistics part, and we have our supervisors working in the store. Now I know, okay, who's working there, how many hours do they work, and now I can get down to the business and say, hey. Please analyze any staffing gaps for next week. So I see if there is some demand to talk to the to the staff, to some some people, asking them for overtime and and or to switch their shifts, for example, extend working hours, and so on. And this is quite an extensive task, and it would, in a normal scenario, take probably some hours to fix that. Right? And here, we can now take a look, basically. Okay. We have a dedicated tool which will help to identify some challenging periods, right, which is basically getting the data in the right format out. And we see here now, okay, we have several missing shifts for the sales associates. We have stock associates shifts missing. And also for the cashiers, there are two missing shifts, and supervisors are good. And then we have the same group by each day. And we see already that there are some, yeah, concentrations in the evening hours. Nobody wants to work in the evenings. Right? So that's where we have our problems. And now, basically, our agent is also providing us some recommendation what to do, how to fix this, and has consistently identified a sales associate shortage for all the late hours. Right? So what can we do? Right? I will nobody wants to work, so we need to find, like, a fair schedule or fair distribution. What I want to do now is I want to create a shareable shift plan, ideally in Excel as the people are used to that. And it should fix the gaps with overtime, but should take a look that it's, like, evenly distributed. Nobody is getting too long or too many overtime hours as they're also quite expensive for the business. So we will need to basically pay a premium for each overtime hour, and this will then add some additional burden and reduce our margin for the shop. So we take care of that. And this can look like that here. So we have our shift plan. If we take a look, we see now, okay, these are the okay. Actually, it has just done the the evening shifts, extended test. So that's not that what I wanted, actually. So let's close that down and say, please not only focus on the evening hours, but create a full shift plan I can share with the team. And this should be able to take the full picture into account. Might take a little bit longer as we need to ask more details from our system, but should be able to finish in time. While this is working, as it can can take some time, maybe oh, it's fast today. Let's see. Here is then the full shift plan. Let's check this out. Download it. Okay. So that's not not perfect looking. I have another example might want to share with you here as this could be a little bit look a little bit better. Done a test run as the demo might fail. Right? So let's see maybe this example here. And here we have the, basically, the daily coverage going through day by day from morning hours to evening hours, list all the employees, list their roles. The same goes for the shift assignments. Now here, each person sees or can filter basically by by them here. So if we download it, we we will be able to filter by the name and they have their full assignment when they are need to work. And here we have also an overview of all the the people, their roles, how many hours they have contractually, and also how many hours they have actually been scheduled for the week. And we could even run an analysis on top of the times they spent, basically, how many overtime hours they have and so on. And here we see, for example, actually, we have more people having less hours than required than more hours. So we need to find a good good balance here. And here, could work together with the agent, understand basically who is working or what we should talk to the staff to understand. Maybe they can extend some availability times. Maybe we add some rotation to the shift schedule so everybody needs to take care of some opening hours, so late opening hours, and this is distributed fairly. Now we could drive this even further. I would like to show you briefly how this is done in StepLogic. So this is, for example, our our pipeline, which gets the available stores out of the system. And this is reusing one of our patterns we have done created for for this. It's basically a tool based pattern. So each pipeline with this kind of shape will be can be done used as a tool. And all I've added is more or less the central part here where we actually get the data back from the the source system. In this case, we will use a generic JDBC connector as this is a mainframe application. Could also use copybook snaps for this. And based on your system, this would, of course, change. So in this case, very simple. And then we get this in. We get the result back. And I will also briefly show you all the other tools I've created for this demo. So we have one for opening hours, one to get the employees. And you see it's kind of similar shape. Some are more complex than others. Some are a little bit less complex. So you see the basically, how it looks like. Right? So but always similar doing similar shape of those tools. What we need to define in SnapLogic that we can actually use those tools is done in the pipeline properties. So in here, we will switch off or we will need to provide any input parameters expected from a from a client, basically calling our MCP server. So it's on the store ID or the week start date for this get staffing gaps pipeline. Then we will need to define in the open API section what kind of type of variable is this, what it actually does. Right? The the description is required or not. And in the info path, we need to define a purpose for the actual tool, which looks like that. Right? So we have some some text. I already see I have a typo in here, so we can use that text. And this will actually pass over to the AI agent so he knows what the tool can be used for. Now once we have configured our tools like that, we can also, based on a pattern, create an MCP server. All we need to do is we need to adjust this small little multipipeline function generator. So here, we will go on, select our pipelines we want to use, and and that's it, more or less. From there on, we will say, hey. I want to trigger a task. And here we have done a checkbox, registers MCP server. We will need to select a a Snaplex environment for that. We will need to select a note, and then we can create or set up our MCP server for usage. And with that, we can make any system MCP server ready. We can turn your old legacy systems into an AI already platform so you can build your AI agents and drive value out of this. And with that, I would end my part here and pass over to Maria. Thank you. Thank you so much, Constantine. That was really useful. Great demo. As a reminder to everyone, we have a couple of questions. Don't be shy. Drop your questions into the q and a section towards the top right hand side of the screen if you have any, and we'll try and address those towards the end. And so with that, we're gonna hand over to Jocelyn, who's gonna introduce the agent creative boot camp success story and our customer speakers. Thank you, Maria. Good morning, everyone. Today, we're going to introduce participants from Revente. So I have Seher and Nicolas that will join me. And we're going to speak about their participation in the agent career boot camp. Effectively, how do you learn to do what Constantine just showed you with the development of AI use case within Snapdragon, whether it is for agentic orchestration or MCP server readiness. Before we dive in, Seher, Nicolas, in this order, can I ask you to introduce yourself? Yes. Thanks for inviting us, first of all. I work we work in office in technology department, engineering and architecture team. I work as data architect. Hi, everyone. I'm Nicolas Maison. I work in the same team as Seher, and I'm a principal solution engineer there. And we both participated through the agent created boot camp, which was seen a few well, a couple of weeks ago or something. Yeah, thank you. So to give a bit more context on the agent creator boot camp for those that haven't heard of it before, it's an intense training course. We've been running this training for about a year now. The first iteration of it was in January, actually. And back then it took about a week and a half. Building the agent was much more complicated. The technology has matured a lot, and I think our participants are getting smarter and smarter as well about how to leverage AI, so we managed to commence it to only three days. And in those three days, the first day is a set of instructor led exercises that walk you through the core concepts: front end generating, tool coding, agent orchestration, MCP server creation. And on day two and three, you bring your own use case for a project phase where we help you so that at the end of those three days, you walk away with the knowledge that you need and the working agents that speak to yourself and to your business. And that's what Sejal and Nicolas will be able to present today. It's their feedback and their experience on the training itself, but also their achievements and the use case that they built in that product phase. So, Sigve, I will hand it over to you for the presentation, and then I will let you present with Nicolas as well. All right, thank you. Just moving on to our first slide. First, I want to give you some information about our company. Revantik is a Blackstone company. Blackstone, maybe you heard, it's a globally operating asset management sorry, alternative asset management company. One of those alternative assets is the real estate, And Revantage is providing corporate services to Blackstone and to its partner companies, portfolio companies globally, including Europe, North America, and Asia. And in this use case example, during the boot camp, we focus to our in house developed application called Newton. This is where we store all legal entity information and used by legal team and compliance departments, as well as finance. It's widespread of usage company wide. It has a web based user interface and also REST API access. And this REST API structure is include includes many endpoints and documented on Swagger platform. And in our use case, we focus to connect into Newton API using this agent creator capability of SnapLogic. By the way, we've been using SnapLogic more than a year now. So so far, we had a positive experience, and we want to explore the AI functionality of the platform further. That was one of the reasons we joined the boot camp. And the current state of usage of Newton application is Newton data is loaded to our data warehouse. It's a traditional overnight load, and we have several business views developed in Snowflake platform. And users can connect to this data using Power BI or Sigma for analytical reporting. The purpose of this scenario use case that we developed was allowing a text based querying of data without waiting data warehouse load, accessing to data random basis for ad hoc reporting, maybe as a complementary of the starting reporting functionality of the application. So we thought if if we can provide access this way, it will be much easier for you users to query data because they wouldn't need exactly to understand the table structure, data model of the Newton application to join the tables or locate the right data they need to use. And, so we are planning to build on these use cases, actually. So, the result was really, well appreciated across the company. And we are planning to create sub agents and some predefined prompts. And, also, we are planning to extend the usage of, again, this agent creative functionality and maybe connecting to other data sources within the company. With this, I'm gonna hand over to Nico for the demo part. Just going back to the first slide. Rice. Yeah. So okay. Let me share my screen. Can you guys see my screen? Yes. We can. Yes? Okay. So as I was saying, we had the idea to create basically a simple prompt and let user try to fetch data from this Newton application using its first API. And the idea why we thought that this could be easily a successful essay, let's say, is that because we have at our disposal this open API documentation for this system, which basically it's an industry standard way of documenting REST APIs. And the good thing is that it's made of natural language, which we thought would be a good and easy, source for the LLM to use. And what we were curious about is by just handing that open API documentation to to the LLM, would it be able to query multiple endpoints on its own based on the prompt that we were giving it. And oh, unfortunately, I have to reconnect second time. I was spitting my snap projects for that. It didn't do that to me during the demo, but, yeah, you know, it's the demo effect. So I will try to validate. It takes quite quite some time to validate this pipeline, so I'm not sure I'll be able to show you the out the the output. So what we came up with, and thanks to the agent creator boot camp and Joseline's co collaboration, we didn't build an MCP server, but more a tool for the LLM to be able to call the rest API the way it wanted. So very similarly to what Constantin showed during his his demo, we built a very small and simple tool that lets DLLM fetch data from the rest API of the Newton system. And the heart of that tool is here. It is just a standard HTTP client snap for which there is a simple parameter that is here that lets the LLM query, whatever endpoint it wants, and give even additional filtering parameters, whatever. We can see it here as Constantine was showing that we also simply define in the open API tab of the tool what the parameters are. We we named it parameter, and we gave it a description that would let the LLM know how to use it. And in the info tab, we also describe what this tool purpose was. Then in the main pipeline, we just we just let me expand this. We just connected the tool so that the could use it. We also loaded the open API documentation, which is a JSON which is also available in JSON format, so easily parsable, things like that, and really easy to to manipulate from an MLM point of view. We use the Amazon Bectrobe Bedrock multimodal content in order to fetch the documentation as a parameter for the LLM. The LLM is using Amazon Bedrock. The LLM was oh, the scene able here, I think. Well, it was closed on it behind the scene. I don't know where this is. I don't see a word anymore. And we just give it a very simple prompt. Let me why doesn't this work? Just last one. If I reload, my validation won't be able well, the validation is not very important. So I will reload to show you the prompt. And so we basically asked the LLM to fetch one entity. And in the main response of the entity, we can see that, well, there is shareholder information included. But so it gives the number of shares, the type of shareholder, and an identifier for the for that shareholder. And depending on the type it is, the shareholder information can be retrieved by fetching data from different endpoints based on the type of the shareholder. And this is really what I wanted to see if the if the LLM could figure out by its own. Just by by looking at the open API documentation, if a shareholder was a a general partner, would it go by itself to the general partner endpoint with the correct identifier and retrieve data from that general partner and show it in a in a in a report. So, basically, in the prompt, I asked to I asked the item to create a report, fetching all the shareholders, showing me a graph pie chart with the distribution of the shareholder. And depending on the shareholder type, I asked you to retrieve more information for each of the different shareholder types and to output some good looking report, basically. And because I had to reload my page, I cannot show you one interesting point would have been that in this, we could have visualized the agent flow. And in the agent flow, we would have seen that it, all the steps that the LLM had decided to follow in order to fetch all the data that I was asking for in the prompt. And, basically, it it worked, of course. And by giving it only the initial ID of one of our legal entities, First, the LLM went to the correct endpoint, fetched that legal entity data, retrieved shareholder information in the response, and depending on the type, made a call to three different endpoints and and fetched the additional information of each of those shareholders and basically put that all together in one in one report that I will show you, of course, a sophisticated data because that's quite sensitive. But you can see here that it retrieve information over the the main entity with ID 2201. And here is the shareholder type that I was talking about. So general partner is one endpoint. Fund entities are one separate another separate endpoint. Legal entities is another separate endpoint. And I asked also in the prompt to give information about the funds that were linked to all those shareholders, and it also fetched that from the corrected point. So, basically, it was a a huge success. We showed it to our to our superiors, and, this, gives us, now ongoing project to demonstrate. Of course, it was a first draft, let's say. We did this in a couple of hours, I would say, creating the tool, making it open enough so that the the LLM could use it to fetch anything. And, yeah, that's pretty much it from my side. So if you have questions, I would stop sharing now, but I'm open for questions. I don't know, Jocelyn, if you have things to say. I I might have forgotten to highlight something. I think, first of all, again, this is very impressive work. We didn't necessarily see the visualization in action for your agents, but with a very simple setup, which you thought of and you designed, you were able to generate reports connecting to a variety of inputs. In an application of your own, this isn't like a standard system, which is very well known. And that. ability to create autonomous agents is very powerful. And I think maybe the one question I have is to get your feedback on the experience and if it helps you build confidence in doing something that perhaps you didn't know how to do before. Well, I never had the infrastructure needed or at my disposal to to feed. So just a quick word about our landscape. As Seher said, we have we are using Staplogic for about about a year now, but we're using it for integration pipelines. So we never had well, we are starting with having our own private privately deployed LLMs. And during the the creator boot camp, Joslam gave us one instance of CluedSonet four dot five at our disposal. We use that LLM for other stuff in the company, specifically for coding, but we hadn't the opportunity to to to try to throw in data to throw data at it. So, yeah, that's what we did. I don't remember what your exact question was, Joslyn. I'm sorry. I'm Maybe I can also add something on it. So we were actually four people in the team from Revantech. So it seemed the learning curve wasn't that steep at all because of the easy to use visual interface. But additional to that, I think major benefit for us was we already have those integrations. Right? We already have connections to the source systems. So the nice thing is this feature is built in as part of the SnapLogic platform. So we can already use those existing connections to the source systems. We call them as producer pipelines, for instance. We can embed our LLM prompting or LLM pipelines created with agent creator into those existing structures. So that that makes us make it makes it much easier to integrate to the existing pipelines. Perfect. Thank you for the feedback. Super helpful. Then should we, Maria, go into the general Q and A? Yes. Thank you so much, Nicholas, Eeyan, Jocelyn. That was really great. I think with all this amazing content, we've actually run out of time. So what we're gonna do is we have some really great questions, and we're gonna take them offline and share the answers with you following this session. So really appreciate that. And with that, we're gonna wrap up. Just a quick plug. Like, if you want to keep the conversation going, I really encourage you to join the integration nation as a SnapLogic community if you're not already there for customers and partners to connect, share best practices, and get faster answers for peers and experts. Yeah. And, again, like, really great session. Thank you so much for joining us today. I really appreciate your time. Have a great rest of your day, and we hope to see you in the next session. Thank you, everyone.