Awesome. So, yeah, my name is Sebastian. I'm a strategy consultant for EMEA, currently based in Aachen, next to the border to the Netherlands and Belgium in Germany. I have been with the company for a few years, come from the analytics space, and love everything around enablement training, upskilling and webinars, which is one of the reasons why I'm here today with you. But with me, I have the one person that I always go to for any data question I still have, Chaitanya. Thanks, Sebastian. So, yeah, my name is Chaitanya. I am a data architect based in Frankfurt, and I work, or I love to design and erect data systems. I'm with Interworks a little over six years now. Awesome. And then, actually, Vicky, it's you again. So before we go into our actual content, we need a little bit of participation again. Wonderful. Yes, we have another poll. I promise I won't keep on throwing these at you the entire session, but as I said it really helps us to tailor the content towards you moving forward. So the question, where do most of your quick data questions get answered today? So we've got an array of options for you. So a BI dashboard, ad hoc in Excel, an AI chat bot, a colleague, or they go unanswered, which is really something that I'm hoping we won't see too many click. But please do participate and it really supports us in making sure that we get information that's tailored to you. So yes, thank you very much for those who are participating. I'm going to close that poll and I'm going to share the results just to give you all some visibility. So hopefully the answers we were looking for. Yeah. I'm already loving the fact that there are many who are getting their answers from an AI chatbot. That's amazing. Alrighty. Then let's go ahead through our story for today. So for those of you who have joined the last two webinars of that series here, we started in May. There was a lot talk about AI already. So we started in May with our journey to BI three point zero with our beautiful Chachi Petty images with this one green protagonist here that guided us through the different layers starting with the old time before Excel even was the thing. We went through Cognizant Mind strategy and so on. Then, of course, Tableau, Looker, Qlik Sense, or QlikView at the time. We talked about AI, but then we had our second session almost a month ago, when I remember correctly, about AI analytics. Oh, yeah. Before I forget that, we also had this beautiful slide about Excel that is still there. Although most of the analytics tools out there are predicted it would just die out at some point, it did not. It is still around and used more than ever, actually. Not just Excel, but everything around spreadsheets, of course. Then last time we went into the AI analytics space and we started with this big gap between our users on the right, usually our business users, people who make decisions, need insights, and people on the left, which is not really people, which is but more the data side of things. So we had this big gap in between that we tried to close from the analytics side. We also went into a little bit about the ROI if that does not happen. And we talked about specific analytics apps, in particular, Sigma at the time that enables analysts and business users to actually cross that gap here from the right to the left. Today, we are talking about AI first and not AI only. What do we mean with that? So before we start talking about anything else, let's first quickly look at a very, very small chart because over the time, and with that I mean since twenty twenty two onwards, there is a kind of regular influx of AI apps. So I imagine this old curve here, something like that. There's a new innovation, a new AI solution out there. Usually, it's just one or two or something like that, and then everything blows up and the market is flooded with new solutions out of that space. After some time, it's consolidated again and something new emerges. For example, at some point, someone said, hey. What about agents? What is that? Suddenly, were a few agents. A month after that, there were billions of different agents out there and so on and so on. And I think we could actually let this graph grow if we wanted to. Why am I showing that? Because quite often, I hear and we hear that AI can do everything right now. Maybe not everything at once, but for each and everything out there, there might be an AI solution or an AI app out there. And that is certainly not completely wrong, although it's still arguable. The thing is there's still some manual and human elements involved in the whole thing. Some things that we need to think about, that we need to build out beforehand so AI can actually do things. And this is the main message that we have in here that everything is driving on today. AI first is something we should go for. We should think about how to leverage AI, especially to get to the answers that we need. But it's not AI only. We can't just say, hey. Here is some AI. Please do everything for me. I need to know that. This won't work, at least not yet. Maybe we get there at some point, but it's twenty twenty six. We are not there yet. So last time, we looked onto the right side here, our business users, and how they can cross the gap. Today, we are taking a slightly different perspective and start from the left side. So we are looking more at the data side and how we actually can make analytics insights available from the left side so people on the right side can actually use them quite simply, easily. For that, we have prepared a really cool demo. But for that demo to be shown, and this is one of the big reasons I have Chetanya with me today, we need to build that up a little bit. So I'm going to show you a few really cool new images. I hope you are used to that already in all these webinars. We have really cool blueish ChatGPT images. And before we go into any demo showing anything in the AI space, we have to set two things up. Number one, we have to talk about different kinds of insights and then also the current data setup that leads to insights in the end. If you have any questions in between, as mentioned before, just use the chat ask way. We will usually get to you right away. So let's start with the insights. We defined three different kinds of insights when we talk about AI out there. We have quick compound and conformed insights. Now, if you're hearing that for the very first time, then you might be as confused as I have been at the time because those words don't come that easily. So what do we mean with that? First, let's take two seconds to appreciate that visual here. It took quite a few iterations to get to that point, but what are we actually seeing here? So first, of course, our protagonist here. Our protagonist is our business user or our analyst, whoever we are thinking about, who is trying to get answers. As I love to say, whenever we are dealing with data, whenever we're dealing with the data bases, warehouses, with the dashboards, with insights, it's always about getting answers to something. I have a question, the data should be able to answer that, and I'm trying to get there somehow. This is the bucket that our protagonist here has in his left hand. He's collecting insights. This cool droplet here symbolizes our quick insights. Now what are quick insights? This is what we usually don't need that much of data modeling for. It's the quick questions that someone has that does not have a very long, let's call it shelf life or half time. Quick insights are usually pretty ad hoc. I need to know something right now or something or some decision I need to make right now. So usually quick insights need to be there within the next seconds, minutes, maybe an hour, but definitely not longer. And usually, after we use that insight, it's gone again. We don't need it anymore after that. So it's a small droplet here out of the pond at the very beginning. Yeah. Then we have our yeah. An example for that might be I don't know. How many clients do I have in a very specific business area? Some and so this is a question that I might need for some decision I need to make right now. And maybe there's not a full blown out dashboard just for that one number available. Then we have the next layer here, compound insights. The different streams that we are seeing here are symbolizing different data sources. Different sources where the data comes from that need to be combined. So there's a little bit of data modelling involved here already. And these are things that vary a lot in terms of shelf life. So could be one attack request that I have that still needs data from different data sources. It could be some regular question that comes along basically every week. For example, I think we put the example here that what were the numbers on each and every weekday in the last week? And if you are like me as an analyst, this is a question I tend to ask every week. So at some point, I'd love to have some solution that gives me that on a regular basis. Also, compound insights is everything that you need for a meeting, for a presentation on the next day, something that needs to be trusted, of course, but we get to trust also in a few seconds. Then last but not least, we have conformed insights, which is this great I don't know what it is, building castle, water. I don't know what it is exactly. This is here our confirmed dam. Dam, maybe. Yeah. But, I mean, it's very royal, very royal dam. So here, we definitely need a lot of data modeling, but for a good reason. Now here for conformed insights, these are the ones that everyone needs to agree on. So for example, if we have some metrics in our organization that everyone needs to know and also that needs to be defined the same way for each and every person. This is where we talk about conformed insights. Usually, there's a lot of calculation happening there, a lot of logic happening there. Also, there we find things like an annual financial statement. So something that is even legally binding and needs to be legally trusted. So a lot of work goes into conformed insights in the end. Now if you're still struggling with these three words here, I took it upon myself to explain it to myself a little bit easier with the example of a kitchen. But before I get there, this is basically what I just said, although in slightly different words. If you want to take a quick screenshot of those three elements, feel free. I don't want to read ourselves through that. Three, two, one. Alright. So a different way to approach that is a kitchen. And I'm very proud of that image here. Again, this took a lot of iterations, especially since I wanted to show the previous image here in that window on the upper side. But okay. What do we have here? So let's imagine we are in a kitchen and we are hungry. Quick insights is this apple here. Now what constitutes that as a quick insight? So if there is an apple and I am craving an apple, I'll just take that apple and eat that apple. If I'm craving an apple, but the apple is not there, I will probably not invest time to go to a supermarket to buy one apple. The invest is just too much for that. And this is actually one big symptom we are seeing with Quick Insights nowadays. Quite often when they don't exist yet, it's pretty they are pretty hard to come by. So someone needs to build that metric, needs to build that insight if it does not exist yet. And that invest to build that is usually way more expensive than the insight itself. So quite often, those just don't get built. So quick insights quite often are missing. Then we have this sandwich here. It's a very round sandwich for some reason. But, yeah, let's pretend this is a very ordinary sandwich. The sandwich is something that I can build myself. So I go to the fridge here on the left side, take one, two, three, four ingredients, and I put that sandwich together. There's no recipe involved or something like that. I've learned how to build a sandwich when I was a kid. I learned that while growing up. And also, doesn't really matter if the next sandwich that I make is exactly the same one or slightly different. It always depends on what I want exactly. And then we have those cakes, let's say for a birthday party or something like that. This is our conformed insights. Cakes are plant. There are recipes for cakes. We care about allergens. We care about how long the baking time actually is. We want the outcome always to be exactly the same. We want them to be reproducible in the end so we can trust the process. Also, I'm pretty sure that my guests in the end will judge my cakes. They might not judge my sandwich here, definitely not my apple, but they might judge my cakes here in the end. When it comes to data modeling, and I mentioned that quite a few times already, how much data modeling is necessary for those steps, we discovered that we can't really get rid of data modeling with AI. That would be great. And maybe at some point, we are getting there. But for now in twenty twenty six or mid twenty twenty six, there's still some data modeling involved here. But the amount of data modeling differs quite a lot. So for Conformed Insights, there's a lot we need to do. Compound Insights, and Quick Insights, a lot less. Alright. So this was Insights. Then one more thing before we go into our demo, and this is the data setup that we usually go for and that we need to go for when we talk about AI. So best practice until now and also used a lot for confirmed insights is this usual setup that hopefully most of us here in the call have seen or heard about before. So we have some data sources, some databases somewhere living. We might hopefully have a data warehouse or a data lake house somewhere where those tables can be ingested. Sometimes this is natively possible. Sometimes we need to go some way around here and extract and load those different databases with some tool like c data sync or Fivetran to ingest it from there into our data warehouse. Then, of course, we have a BI layer there with dashboards or tables or whatever we need to find our insights at the end. This is cool, but, of course, there is a new player around. And I say player here because our dashboards have been the interface for our data consumption. Now we have AI there and AI chats specifically are another interface. This is something we talk to to get our insights. How does that now look especially for quick and compound insights? Well, for that one here, we are concentrating on AI only. So we're getting rid of the dashboarding part here. This might be a different topic, might also be possible, but not for what our purposes are for today. Also, we are introducing another concept for getting data into our space here, which is called data virtualization. Now we still have our data warehouse here, but there's something new that that we try to leverage. Data data virtualization itself is not a new concept. This has been around for decades, but there are a few new tools out there that makes life a lot easier nowadays, specifically with AI. For everyone who has never heard of data virtualization, not a problem. What is that? Imagine you have a few different data buckets. This well, whatever that is here on the right. I'm not sure. This lake here on the upper right. We have a river. We have some glacier lake. We have some pond here. Different databases that might speak different data languages. Now we here as the data engineer, as the analyst, the business user, we speak one data language usually. Mostly SQL in the analytics space. So SQL structured query language, something that basically every relational database and specifically BI tools speak. Not every database speaks that or is set up like that. Data virtualization is a kind of universal translator for all those tools. And that means we don't have to load all that data into a warehouse all the time. We can actually directly query that data without setting up specific credentials each and every time that we do that because we have our data virtualization tool that translates everything for us. And such a tool we are using today. So scenario time. For that, we are using exactly three different tools to make life simple. We have a data warehouse, we have an AI, and we have a with data virtualization. The tools are Snowflake. This is our data warehouse for today. We're big fans of Snowflake for years, basically. We have Claunch that we are using as AI chats and AI agents. The way, everything that we are showing here would also work with Gemini, with CheGPT, with whatever else you want to use. I've heard even directly within Snowflake, if you want to keep in the tool, you could still leverage the Snowflake AI, specifically Cortex there, to make everything happen that we are trying to to make happen here. But more on that if we have time later on. And then for the data virtualization tool, there's some new tool around called Connect AI from SeaData. This is a kind of fresh tool on the market, and we discovered or rather our data architects discovered that. And usually when they are hyped about something that then everyone on our side is hyped about something. So for data virtualization, we are using Connect AI. The data side is also pretty, pretty simple here. We pretend that we have some tables already in Snowflake. There's an orders table with some business metrics or orders data. Imagine every company, every vertical, every branch you want to be in for that one. It doesn't really matter. In that table, we have some sales data or revenue data for specific countries. And for that, we also have that dimension table here with a little bit of nation mapping. So each country code gets mapped to the correct nation name. Also, CJ took it upon himself to CJ? Schaeffer? Yeah. Took it about himself to build a semantic layer for all of that. Semantic layer meaning there's some kind of translator document that makes AI understand what we are talking about in terms of that data. Sometimes revenue is called sales, sometimes called something completely different, sometimes metrics need specific definitions that everyone agrees on. So the AI also knows what it needs to do. And all of that is written down in a semantic layer. If it was just like that, it would be easy. We could just connect those tables to end the semantic layer to our AI, and we would be done. Actually, if we just have one data source or just one of those tables here for quick insights, that would already be enough. But in our case, we have external data sources. There's one foreign exchange rates data source from the Frankfort Stock Exchange and also we have some other table here and I have forgotten where we took that from, but consumer price index and inflation data. We are currently in the cosmos of comparing trade deficits between countries. Very political topic currently. So for that one, we have our own data here and we have created data. This is what we have power over. And then we have this foreign data. So data that we do not curate and that that we do not want to curate from external sources. Now with all of that, I would say I'm done. I am going to head over Chaitanya, who will talk about what is actually happening there and also how we can now use that directly in cloud and also with our MCP servers. Chaitanya, up to you. I'm going to stop sharing and the stage is yours. Thanks a lot Sebastian. So Sebastian has made my task really easy by explaining almost all the concepts we saw there, what are type what types of insights and what kind of data is there. I just wanted to show you before we start the demo that actually that data exists. And you can see that I'm using Snowflake as my data platform, and here is the bilateral trade data. And here, how it has been modeled. Snowflake, meanwhile, shows really beautiful data lineage diagrams. You can see that I have a very medallion like architecture going on in Snowflake where orders raw, I have the orders table, and then different views, customer, line item, nation, supplier, etcetera, etcetera. Those then get transformed into my insights schema, where I have very small star schema made out of it, which is fact line item, dim nation, and date. And based on these two important ones, I have this bilateral trade semantic view defined. Okay. This data has been taken from which is synthetic data, but still it is a very lifelike data. It shows us, okay, what trades are happening, what trades, what goods are being shipped from one country to another and so on and so forth. Very much believable data. What does this data doesn't have? And that's exactly why we have, those data sources. And this is, a scenario which happens in many organizations, that not all data is probably that relevant. By the way, you can also see that I had to go through a lot of steps to arrive at this. Right? So for my confirmed data, that means I had to I had to make sure that I'm taking the right columns of the source data. I had to make sure that, okay, the logic or the joints that I'm making are correct ones, and then only I can tell that, okay, in this bilateral trade semantic view, these are the metrics, these are the facts, these are the dimensions, and so on and so forth. But then there is some data that we probably do not require every day or that does not come along with our trade data like Sebastian said. So Frankfurt or API that gives us the exchange rate data because if we can imagine, if we are trying to analyze the trade between USA and Germany, they do have different currencies and we need somewhere down the line some kind of exchange information. And similarly, there is a World Bank data which provides us with two important economic indices or indexes, which is consumer price index and the inflation. What I've done is that although you can see the four different data sources here, for this demo we only needed to bring two of them together. So I have actually, I can also show you by going through that, okay, These are the actual, API endpoints that I have taken. It's the same endpoint with different parameters, which translates or which Gets instated into two tables and similar thing is also here. I have used two different endpoints, one for the inflation as dictated by the API and one for the consumer price index. And all of that I have bound together in a nice workspace. Workspace is nothing but a virtual, you can say, organizational structure. Anyone who is going to connect to Connect AI is going to see this structure as if it's a database schema and table. So economic data will show up as database. Then underneath that, there will be economic indicator schema and exchange rate schema. And underneath that, it it will be the same tables, which I just showed you, that are being exposed. So now let me go to the AI agent where I have set up one project. I gave, instructions to the AI that okay, you are supposed to look at these and these places and, you should only look at these places to answer these questions and so on. So some minimal configuration I have done. By the way, you can also notice that the, like Sebastian said, in my data platform, I did have to take definitely more effort than what I have taken it on the Connect AI side. On the Connect AI side, I have simply connected to those data sources, and, rest of it, will be done by by the AI agent itself. So when I'm setting up the chat, am so you can see that these two connectors are now available here to Claude, And I will be mostly instructing, or I will be using only those two connectors and Claude should looking at the information only from those two connectors. So, let's start. So I have written down the questions so that I don't have to type in front of you and make some spelling mistakes. Let's start from very simple questions. Okay? By the way, I'm not going to write any. SQL. I'm far from that. Normally I love to, but for that I could have done that in Snowflake as well. And today that's the whole point that I want to show you that I don't have to anymore. Right. So let's go through a very factual, very simple What is the date range available of the trade data? And now, I'm not telling it and it found out or it knew that it had access to these two tools and it is asking me whether it should look into this data to find out the answer about that. So I'm just going to allow it. By the way, it shows the correct tool in two tools without me telling it. That is because, multiple reasons. First of all, the MCP servers are providing that context to claw. And at the same time, I have also provided it with instructions in my project setup that, okay, if it is this data, then you should look at this connector. If it is this data, then you can, you should look at this connector for anyone. So, by the way, first of all, you can see that it has made two tool calls until now. First is the bilateral trade semantic view, and the second is SQL execution tool. And I'll explain what they are when it's coming back. So the first one, so in the Snowflake MCP. Okay. Yeah, so now it is giving us, okay, this is the month range, monthly granularity. We have that and it has used these two tools to, it is also giving us a query ID by the way. I did not ask it to give us any query ID or anything. So we can go back and look at the steps. So each MCP server exposes some tools. Snowflake so MCP, by the way, for those who don't know it, MCP is model context protocol. That is nothing but a protocol that allows the AI tools to communicate to each other. So each MCP It was by Anthropic, right? It was invented by Anthropic and they immediately made it open source and that's why it thrived. And now MCP has become this global standard, how the AI tools communicate between each other. So each MCP server exposes or publishes multiple tools. So here we can see that Snowflake is doing this bilateral trade semantic view, the one semantic view which I showed you here in Snowflake. And then Claude is sending it a message, and this semantic view basically is nothing but it is being shared or used by the Cortex analyst, which returns us a nice SQL query. And then Claude takes this query and asks it to execute it to Snowflake itself. Claude is not trying to come up with any kind of results by itself. So let's go and maybe ask it a few more questions. So again, this, you can see that these questions are based on just one table, by the way, time quickly read out the question. Sure, so the second question that I'm asking is what all market segments are represented in the trade data. Okay. So this time I'm going to say always allow, because we now know that it is. It actually goes to snowflake to answer our questions. And if you look at the question, it's also a simpler question. We do not need any other combination of data sources for it. All it has is it's it's simply it can answer that from just one table. Okay. What are the market segments? If we think in terms of SQL, it's basically just a select distinct query on a certain column. Right? That's all that all we know. And it tells us, yeah. Okay. This is machinery automotive and so on. Now let's try This is basically my insight right now. Right? I just ask a question and immediately get the insight that I wanted. This this is your insight. Exactly. Let's take it a notch further. Okay? Let's see an example of a compound insight where I need to combine the data from at least couple of sources. And what I'm doing is I just told you that, I can also show you, by the way. So here, if you look at it, at the columns of this, there is there are only, values in this table, but you will notice that there is no currency associated with that. So the assumption is that, every nation, the the customer nation's currency holds. And so what I'm trying to know here is what was the value of imports for Germany for month of Jan twenty twenty six? So I'm looking at now historical value expressed in USD. So Germany's value will be assumed in euro, obviously, and therefore, Claude will have to go somewhere else to find out. Sorry about those popping up on my screen. Yeah. We had some poll running in between. Vicky, would you care while Claud is working to quickly share what we got. Absolutely. I have the results in front of us now. Yeah. So meanwhile, you can see that, it went to Snowflake, brought the value of imports in Jan twenty twenty six, and now it needs to convert that Euro value into USD value. So now it is looking to the economic data, MCP server that is that is there. Right? So this time as well, you can see that it will ask for permission every single time. I'm going to, allow it. So you're basically giving it permission for both tools to talk to them? Not more no. I'm giving permission to Claude to talk to them so that Claude cannot make any unsolicited calls without my knowledge. But now I have given it always allow, so it it will be just so I'm just saving few clicks of mine. It's thrusting it. Yes. EMPP servers, it will it will do that. So now you can see that it is saying that, okay. January twenty twenty six, it is one point three billion dollars of imports valued in USD. And it is also giving me, without me asking, it is giving me the explanations as well. Right? So trade value, euro, importance currency, and then it has to convert that into the so look at the steps that it had to take. It first of all had to get the trade value in euro. Right? That is the only information it was going to get from, from the trade data. Then had to figure out the average exchange rate for euro against USD for the month of January twenty twenty six and then perform that conversion. You can now see that there were multiple steps involved in this. And that's why makes that as a compound insert. It was still not something too complex a query or something that that was or that that will that will have to be disputed because we were just, pulling one metric. Right? We were just pulling the import data for one country for one month. Relatively simple. And then there was just one conversion involved. Now let's I think it would be Apple, right, from before? This is more like a sandwich. Yeah. Exactly. It had to go to at least, two sources. It doesn't matter about the sources. It's just about, how much effort went into there to just arrive at that insight. Right? There were multiple steps and so on. Now let's take it a notch further. So this time I'm going to ask it. I'm really going to so by the way, these questions I have written half an hour before our webinar. So these are not rehearsed questions. Claude has not seen these questions before. This is truly a live demo. Okay? Right. Yeah. Now I want to know what was the dominant category for imports for France in year twenty twenty four. Okay? If you now, by now, figure it out, this is still a single source question, but the number of steps that we have to take into this are quite a few. Right? And because we first need to find out the total amount of imports for the year twenty twenty four for country France. And then basically we have to find out or the AI has to find out, okay, what was the dominant category in this? So it now says that, household, I think it was, if I'm not wrong, it's two point seven billion, highest of all the categories. And it also provides us with all the other data. Just because it is AI and it likes to show off, it is also telling us that the segments are very close, you know, within just four percent. So dominant is a narrow lead. Of course, they are going to be synthetic data. Who knows what the actual values were? But still, had these been actual values, the insight still holds. Right? Everyone just following Shetanya's voice, you just read out what AI mentioned at the bottom. Right? Yeah. Exactly. So this was not your human analysis, but AI helped. Yeah. This was an AI help. So something that I did not even ask for, it is telling me that. Now I'm taking, like, really. So look at that. How much goods did Germany import in twenty twenty one? Okay? And what will be inflation adjusted value of these imports in year twenty twenty five? I'm really taking it. And now this is a complex question, isn't it? I'm looking at the historical data, and then I am trying to find out, okay, whatever the value was, I want the current or the present value or the last year's value of those imports. AI is going to have to do a lot of things. Nothing against the dashboarding tools, nothing against the BI tool, or mainly about the dashboarding, but think about how much Active modeling or, you know, perceived modeling will have to go into. A dashboard if we are trying to, or if you user is trying to ask questions like these, right? And if you think about this in this particular scenario, or from an analyst's perspective, these are all very real questions. So, now it's also giving us the answers in stages and it's first telling us that, okay, imports of twenty twenty one thirteen Billion. Now, German CPI, it is going to find that out. Right inflation adjusted in twenty twenty five terms. Fifteen Billion euros, So I somehow do not believe this. Okay? And I'm like This would be my next question, actually. Yeah. Exactly. Great. How the heck do we trust that that number is correct? Exactly. How is how is it how can it be that the AI is not making up the values and so on? So maybe let me ask that. Show me all the SQL queries you used to alright. I mean, this is one of the biggest issues we still see, which is within general in the BI space. Right? Can I trust that number? Yeah. Do I know where it comes from? Correct. And as if it, you know, as if it anticipated my question, it was very much ready with that question. And they said, look. This is what I used for getting the imports of twenty twenty one. I'm not going too deep in it, but I can see that, okay, it's going into the line items, which are the facts and selecting only these columns and selecting only customer nations. Then it has definitely filtered that on the year and customer nation as nation key, and then it has sum it up. Right? So So for everyone here in the code who has never seen a SQL query, this is one. Yeah. And then what it is doing is, it it has it has part so that that is just first part of the query. The second thing it is shooting to ease for the consumer price index, right? And it it's showing, okay, it it fired another query this time to the economic data and to the economic indicator schema. Remember I had shown you that in that economic data workspace, there are two schemas, economic, sorry, exchange rates and economic indicators. And out of the economic indicators, it went to the consumer price index because I asked it a question about inflation. This is the AI magic, by the way, that it infers. Right? I didn't have to tell it that to get the inflation data, you have to look into consumer price index. No. You just tell it as if it's your assistant, and it will figure out. It will research. It will figure out where to ask, where to look for the information. And then it found out for between the years, these four years. So then these were the results, twenty twenty one, twenty twenty five, and so on. And then it was Python, of course, because it did some internal calculation on its own, which AI can do, of course. And this is the answer that we got. But again, if I am skeptical even now that I do not trust that. Okay. Sure. AI is telling me that this query was there. Was this really the query that got executed? And for that, what I can do is that I can actually go and try to look into the query history. And indeed, this is the query. I think I I opened the wrong one. Wait a sec. Yep. Yeah. This is the this is the query that was fired on the Snowflake side. We can see the duration. We can also see the query profile if we are interested in it. And on this side, I want to look at the World Bank. So if I'm not wrong yeah. Yeah. This is the consumer price index query. So you can see that even though we are dealing with AI, these systems are such that the tools that we are using, Snowflake and Connect AI, they are making it possible that everything is traceable and everything is, you know, we can make sure that the results can be traced back to its source. We have the lineage. You know, we only have access to the right data, by the way. So that is also something I can show it to you. So currently I'm logged in with a user which has higher privileges than the user I have used in the MCP. But if I simply change my role there, Actually Oh, yeah. That role I have kept separate, so it's not even here. That role Super secure. Yeah. Yeah, exactly. So that role, I went one step ahead not even to attach that into the hierarchy. So that role can only see these two tables and this semantic view and nothing else. Okay. And similarly, also on the connect AI side, I have created a separate tool. You can see that my user has access to all the data sources as well as the workspace, but I have an exactly I didn't want that because I didn't want AI to get confused or to have it excessive access. And that's why what I did is that I went ahead and created a custom tool, which only exposes the economic data. And in that, it exposes these five tools. So you can also see that I have I have disabled the insert and update tools because we we only simply want to read the data. So Can I quickly ask, Shaitanya, for everyone who has never used Connect dot ai before, how long did did that take you? Only in ConnectAI. Only in ConnectAI. The only effort that so like you explained, right, that, anything that goes in the data platform takes more effort because we are doing more testing. Are making sure that there is more conformance and so that, all these things in connect AI all I had to do so sure in connect as well. So there is no. Data modeling will still be there, right? But it is a very light touch data modeling and to connect to these two sources and, to create a workspace out of it. All I had to do, I think I spent around. Three hours or so. Complicated API would take slightly longer, but I think just imagine that if I had to bring these, these APIs into snowflake. I would have been doing even with help of AI coding agents. I would be doing. The week. Yeah. At least two, three days. Because I'll be testing the APIs and so on and so forth. So, yeah, this is power of this is, you can see that it's first or assistant analysis. This is not. Only analysis, because we are making sure that the data beneath it has been reviewed. Data beneath it has been modeled. There are right controls in place and that's exactly what we want to achieve. Right? AI first or AI assisted, not AI only data analytics. Back to you, Sebastian. Yeah. Awesome. Also, I just answered a quick question in the chat. If you want to look at that as well, it's about token usage of that process. Feel free to answer that. Also, see one raised hand there. I'm not sure I can give anyone the mic, so we need to put that into a q and a or the chat, actually. Would be in a Q and A section or in the chat section would be great. Alright. So after we have seen this whole setup at work, I think we are going to wrap our webinar, but of course, not before asking everyone to ask questions. This is our Q and A section. Also, there is one last poll for today. Vicky promised there weren't there wouldn't be that many, and she was right. So one last one for today. If you could decide, in part of this, what would help you most in the night next ninety days when it is about AI and how to take the next step in there? Absolutely. I did limit myself to a maximum of four polls. I think it just keeps everybody on their toes, doesn't it? I know it's lunch time and a lot of people may or may not have eaten, so they can get a little bit fatigued. While we leave that poll open, we've just got a question that's come in. So what business problem does C data solve that we cannot solve with Snowflake, Cortex, Semantic Views or our existing ingestion pipelines? Chetania, you actually very recently did an article, and I'll share the post with everybody on this call, But if you could walk us through that, that would be really helpful. Absolutely. So to put very quickly, first of all, CDATA has two tools. They have a tool called Sync, which is a proper extract and load tool that does not come with the AI MCP capabilities. And the Connect AI comes with AI MCP capabilities. Okay? So why I liked or preferred using Connect AI for this is that Connect AI will help you. It's not that you won't be able to do the same extract and load activity using Snowflake. You can very much you very much can. You can use Python for that, or you can use the tool of your choice and then do everything go through that data modeling round again. But if you looked at what I showed you in the C data Connect AI, what I did about is that the value I find of that tool is that if you want to experiment with some data or if you want to play with the data, which is going to be there, that is not going to be influenced by your organization. For example, in this scenario as well, the orders data is probably something that my organization has generated. These are my orders, but the foreign exchange data is not something that I can influence. Right? And for that, if I want to bring that data in my data warehouse or in my data platform, usually there is a lot of effort involved. Connect AI can only help you reduce that effort because it is making you that data available directly for querying with minimal modeling. And again, if you want to, it can also get that same data into your data warehouse so that you can perform further modeling on it, amalgamate it, and so on and so forth. So that's the value that Connect AI is providing. By the way, Connect is not only providing an MCP interface, it can also talk to or provide the same data to a lot of BI tools as well. All the leading BI tools are there in the mix. Tableau is there, Power BI is there, and so on. Wonderful, thank you. I'm conscious of time. So with that, we'll close today's session. Massive thanks to Chetanya, DJ as we like to call him here, and Sebastian, and to you guys for probably giving up your lunch breaks. We will be getting a copy of the recording out to everybody who registered, so please watch your inbox over the next couple of days. And as always, if you have any queries or questions please by all means reach out to us interworks dot com we're more than happy to support you with your exploration. Thanks all.