Name: Elevate Your Analytics: A spotlight on Spotfire Copilot™ 2.0 and key use cases
Uploaded: 2025-07-01T16:16:09.210Z
Duration: 51 min 40 s
Description: Elevate Your Analytics: A spotlight on Spotfire Copilot™ 2.0 and key use cases

Transcript for "Elevate Your Analytics: A spotlight on Spotfire Copilot™ 2.0 and key use cases": Hello, everyone, and welcome to today's webinar, Elevate Your Analytics, a spotlight on Spotfire Copilot 2.0 and Key Use Cases. We're thrilled to have you with us today. I'm JP Richard-Charman, and I'll be your host for this session. Now before we get started, I just wanted to cover a few housekeeping items to ensure you have the best experience. The webinar will last up to around forty five minutes, and we have a q and a section, that will be held at the very end. If you have any questions during the presentation, please use the q and a panel located on the right side of your screen, and we'll adjust as many questions as we can during the q and a segment. We have also made a few assets linked to today's webinar available in the doc section. You can find that on the right hand side right next to the q and a tab as well. So please feel free to access these. Now right next to the dots tab, we will have a few polls during the session. So you'll be able to answer these polls in the poll section, right next to in between chat and docs, which you don't currently see, but these will appear very shortly. Now after today's session, a recording of today's webinar will be made available on demand, and we'll email you the link shortly after the event. Now with that, let's dive in. I'm extremely excited to introduce our presenters, Craig Hayward, architect at Liberty Energy, and Ahmad Fathahi, senior director of data science here at Spotfire. Now before we dive into, our presentation, just wanted to take you quickly through the agenda of today's session. So we'll be covering the topics of what is Spotfire, and then we'll be diving a bit more into the different use cases of Spotfire and its overall value. And then we'll be looking at, Spotfire at Liberty Energy and how Liberty Energy have really utilized Spotfire. Additionally, we'll be diving much deeper into the different features of Copilot and showcasing what's new in our latest release of Spotfire Spotfire Copilot two point o. Now with that, I'd like to hand it over to Ahmad and, to start off our first poll as well. Alright. Let me see if I can run. Alright. Perfect. So, just a quick reminder to participate in the poll. Please go on the right hand side, click on poll and, answer, you know, what's the status of the generative AI initiatives at your company. We're very curious to see where your organizations are. It's been a very hot topic for a couple years now. It's very interesting to to hear it firsthand from our audience, from our own, community. We'll we're gonna give it another few seconds. And we're gonna have by the way, after we have seen the, contents of today, we're gonna have two other two more quick questions toward the end. Let's give it another maybe ten seconds. Let's get the last votes in. I don't know if everybody who votes already sees the live results. I assume so. It's actually interesting. It's there's a clear winner at this point. Alright. So most of you are already considering it with some already building it. So awesome. And let me see. And alright. So kudos. We're in production. Two votes. Great. So, let's go on back to the presentation. My name is Ahmad Fattahi. Thanks for the warm intro, JP. So let's jump right into it. Alright. Okay. So, so I'm gonna start with a quick intro of what Spotfire Copilot is in case you haven't, seen it. This is a historical view of, our foray into AI in general here at Spotfire. The Copilot is by no means our, first attempt with AI back to, you know, 02/2018. We started to apply some AI in, the the recommendations. Back then, you know, Gen AI wasn't a thing. So, we've made some improvements after 2020 in Spotfire 11 by improving and, you know, adding interactive AI. After Spotfire 12 in late December twenty twenty two, or '23, I should say. We added the Spotfire Copilot one point o. That was the first official release of Spotfire Copilot. And now, you know, fast forward to May of this year, to Spotfire Copilot two point o, we added a bunch of features. It's a major addition to the previous version 1.1, supporting, you know, bringing your own data and docs. You know, we support all the major large language model platforms, in addition to Microsoft, Azure or Azure OpenAI that was in the previous versions. We added support for other clouds as well and a few other features that you're gonna see, in more detail in the presentation today. And in the future, we're gonna see, a glimpse of what we what we're working on in our r and d team and cutting edge to apply, agents and, use them for for the Spotfire platform. So what what is it? You know, what features are in Spotfire Copilot? It is, a extension to Spotfire supporting natural language interaction with the product, you know, at its heart. It does expose a number of Spotfire's native features through natural language interaction. And as we'll see, it also adds a number of, you know, you can argue new features that comes with LMS and the integrated features that we've built custom built for, that integration. It's by no means just a wrap around like, any any like one LMS, put in in Spotfire. It it offers a significant amount of integration and custom logic, leveraging LMs for Spotfire. So it's a nontrivial, addition to Spotfire. It does offer features such as, q and a, general like any other LN, but also specific to Spotfire. You can, use your own documents as context. The user can interrogate data, create charts, based on the natural language commands, and it also can generate code to extend, Spotfire with, the data functions. Without further ado, let me show you a quick demo. It's it's a very short demo. After my first segment here, Craig is gonna show a, exciting demo of, their own, vertical oil and gas use cases, as well. So you're gonna see a plenty of features in in action, and then I'll I'll cover, a little bit more of the the features themselves. So what I'm gonna do here is, you know, all you need to know beforehand is this, fraction or this this parameter called, a processes capability or CPK for short. And all we need to know is basically what it means is consistently producing, the same or meeting the same specs. You know, let's say you're you're producing, a chemical and you you need the pH to be, you know, a number, or within a range. As as as long as you meet that, your process meets that and meets that consistently, your, capability is higher. And the the way it's defined, the higher, the better. So that's all you need to know. So let's dive in. So, on the right hand side, you see the the Copilot panel. Left hand side is the traditional canvas of Spotfire. So the user starts by asking, you know, visualize, you know, starting from the the the basic, statistical process control. Let's let's first look at the runs rates violations, assuming the data is already in Spotfire. So you see the visualization, which is not the the simplest visualization. It's a it's a tree map, but it has a third variable as well. So, the visualization gets created, again, through natural language for the user. And as soon as, as soon as, it's it's it's like, you know, it it can be generated in in in Spotfire or reproduced. And at that point, it becomes a a native visualization. Let me hide this. Okay. The the the next feature is, you know, when the user wants to explain the page, which is which is a great feature for generating reports. Also, we have seen sometimes it creating, insights that might be a little bit hard for the naked eye of of the user. So as you see, you know, it's it's very detailed, not only, you know, showing the the descriptions, but also, you know, ending with overall conclusions. And and sometimes it actually spits out some anomalies or unexpected observations. Next one. So the the user now, asks a question within the context of a document that's privately uploaded to the Copilot. So in this case, you know, back to the the process capability I was talking about, we we start from the definition. What is even CPK? So it taps into that internal proprietary document and provide the definition with the references on the bottom, the citation. So if if you wanna learn more, we can go and check or read more. Next, you know, what is considered a good or bad CPK value? So it says, you know, one three three and above is good. You know, above one is okay, acceptable, but anything below one is is not good. So it means, you know, not meeting the the the specs or not being consistent. Now let's move on. Now that we know what it is, you wanna see, you know, based on the data from our process, but we don't have the CPK value. So let's create a data function to calculate this the CPK. And you note that, you know, in this case, it just knows, you know, what CPK is, what the definitions are. And as it is with, other coding assistance like, GitHub Copilot, for example, it often gets the the the programmer, the coder 80%, 90%, sometimes a 100% of the way there. So in this case, there are a couple adjustments that the user that the programmer needs to apply, but it's nevertheless clean code in a matter of seconds. Saves a lot of time. All the user is left to do is to hook it up in Spotfire and basically say, hey, run this data function on these columns of of my data. And by the way, write the the results, the CPK that you calculate in in this other table. At this point, as you see, it becomes like another native, data function and the data lineage is there. The values are calculated. Now let's, let's see, you know, how we did. You know, ask some exploratory questions. So this is the data interrogation mode. How many parameters have CPK more than one three three, which is very good. Only four out of 30 or so. So it's actually not a super impressive process. 26 are more than one, which is kind of okay. And, now let's visualize, you know, to to to continue the exploratory process. By the end of this process, we'd like to generate the reports and then and share it with our colleagues to, to see how how our process is, is performing. So the user asked to create, a horizontal bar charts, that shows the CPKs. And by the way, you know, do it in, in in three colors based on the value with the limits. It does that, immediately just upon the, the natural language, commands. Now we are adjusting the colors to basically match, you know, the bad ones, showing up as red and good ones, great ones as as, green and so forth. So again, it was, right at the finger point of the user, with a few few sentences. Now let's create a a description. So instead of typing from scratch, now the Copilot creates a description of, this visualization. We are seeing this horizontal bar chart. So it defines what each color means, what are the observations, trends, outliers, interesting finds, unusual observations. So it's not just, you know, describing what it is. It also adds a little bit of, nuance or wisdom or knowledge. And sometimes we have seen, especially with minute visualizations, it generates some some things that are hard for for the for the eye to see as well. So now we have the, you know, most of the the seeds of, a nice report with visualization, and the description, text description in one place. The last mile of generating a report for publication is we wanna publish this as a PDF, for for some of my non Spotfire consumers. I might not know how to do that. So let's ask Copilot consultant itself, how do I export my page to PDF? And here's the description. So this is the Spotfire help mode. It tells you exactly without having to leave, Spotfire or search online or anything like that. And, and that's it. You know, we have the steps and and we can follow to generate the, the PDF. If I can advance to my next slide. So, what we saw was basically this, free natural language extension to software that software Copilot is. I mentioned that we have custom built, a lot of logic and software, including traditional code and system prompts if you're familiar with how, generally, the AI works. So all of these features that you saw happen, in a, in any close collaboration if you like between our custom code and the large language model. And by the way, after, Craig's section, I'll cover the architecture so you'll you'll see, for yourself how things work. It, helps both experienced users and new users of Spotfire to be more productive, more efficient. Typically the new users, you know, we cut the learning curve shorter, significantly shorter for them, time to value is faster. And the, more, seasoned users, we help them pass on some of the more routine parts of their job to the Copilot while they focus on the brainier or more niche and more exciting parts. It is available for download. It becomes your own instance. So it's not a as a service. And, it the architecture from from the beginning, we always architected it in agnostic way. So if you are your organization prefers Azure or GCP or AWS or even we have a small number of customers who prefer everything on premises. It's technically possible, as long as, as long as, you know, you can you have the right resources to to run a LLM efficiently and effectively, in house. That's usually the bottleneck. In architecturally, we are we are a trusted. Let's let's summarize the value points. So, it empowers mainly the efficiency productivity, for for all sorts of Spotfire users. It can as we saw in the demo, it can help enable generate value immediately. You don't need much introduction or learning necessarily on Spotfire. Obviously, that helps, but you don't need that. It it helps creating and modifying, visualizations on the fly through natural language. Quick exploration of datasets by through questions in natural language, that's possible. It significantly expedites and improves generation of reports and narrations for publications. And sometimes even we have seen the, you know, one sole single user themselves. If they're more of a reader person, they use this feature to to to read and understand the analysis better. It, can expedite creation of data functions in Python. And, on top of that, the quality, can be improved because the code is clean and commented. And also it enables, you to, or any organization to, help their their users have their context, the internal context documents, at their disposal right next to the charts without having to leave the, you know, the analysis in Spotfire or without having to ask a colleague or go search somewhere else. They can write their task for, insights. And, I couldn't, I couldn't resist in sharing this with you as well as a result of our close collaboration with Microsoft, and and the value that, they saw we had generated and and the success. And actually kudos to many of you on the call. Great, adoption by our customers and partners. They decided to publish this success story with us. It's published and the link is below. And once we share the the slides afterwards, I encourage you to, check this out. And if you are I know you know the audience, you know, you may come from different, different industries. We have generated a number of different case studies, with full demos and descriptions. What I showed, was intentionally picked to, kinda speak to a, you know, any kind of process. But if you come from manufacturing or energy or life sciences and so forth, we have all of these and all of these are hyperlinks. They're published in our community. So these are full analyses and then processes, with Spotfire and Copilot assisting, the user. Some of them go pretty deep into the industrial and very cold use cases. And with that, I like to pass it over to, Craig and our friends at the Liberty Energy Energy. Thank you, Ahmed. So good afternoon. Good morning, everyone. My name is Craig Hayward, and I'm an architect at Liberty Energy. I've been at Liberty Energy for just over a year now. Before that, I was at Spotify for fourteen years, so I have a good broad knowledge of Spotify technologies. I've architected and employed real time analytical systems for some of the world's largest companies in energy, IT manufacturing, financial services, and sports telemetry. Before I demo how we've used Spotify Copilot, a little bit about Liberty Energy and some of the challenges we have faced and solved using Spotfire. Liberty Energy are an upstream oil and gas company started in 2011 in North Dakota and now operate in 12 states in three countries and employ over 5,000 employees. These are a few of our major innovation milestones. In 2013, Liberty Energy Liberty Energy introduced its first dual fuel diesel CNG fleet. In 2014, Liberty Energy started using containerized sand. 2016 brought the debut of the first generation of Quiet fleet. In 2018, we saw our virtual integration when we acquired SD nine, a fluid end manufacturer. 2020 featured the acquisition of One Stim as well as Freedom Proppant Mines. In 2021, Liberty acquired the containerized sand company, PropEx. In 2023, we deployed our first fully gas powered fleet of Digifleet pumps as well as launched launch Liberty Power Innovations, which provides containerized CNG and the disk logistics. And in 2024, our CEO was tapped to the next secretary under the second Trump admin administration. Liberty has used Spotfire and supporting components extensively, and we've developed a product we call Atlas that aims to address three key challenges associated with the world of fracking. The first major problem is the ability to combine live and static data. The second is to provide a user with functional control of that data. And the third is streamline the data collection and transmission in such a way that it doesn't burden operations. In order to address these three problems, Liberty saw an opportunity to develop a tool that would meet these demands, push the needle, and innovate. The end product of this journey is Atlas. Before we show you how we are integrating Spotfire Copilot, I will explain briefly how we have used Spotfire to address those three challenges. So the first problem to be addressed is combining live and static data. Liberty has historically relied entirely on field personnel to synthesize the live data that they see and record it in a way that it can become soon. With our current process, we are now providing customers the ability to integrate directly into a live stream of information from the field and simultaneously receive static data aggregated from field personnel. In Spotfire, we also tackle the issue by utilizing multiple data sources to combine live and static data from their respective sources to provide internal and external users with the tools to more effectively analyze the metrics from their operations. The second issue lies in user control. In the past, Liberty Energy has used screen share software to provide real time data viewing of on-site activities. This was great for providing visibility of time series data, but without control of the data, decisions that could be made were limited. With Atlas, customers can now interact with their data coming live from the field and view KPIs from their operations in real time. In addition, view lab operations, customers can view or configure historic stage POC blocks and interact with one second time series data. The third issue is employee workflow efficiency. What we mean by this is capturing the whole picture without adding any additional workload to field personnel. Atlas will eventually take away some of the routine reporting tasks like stage emails or data transcription, allowing the field to focus on verifying data quality, ensuring the job is pumped to design and responding to customer requests. At the start of the job, the live stream data is configured to upload one second data to the cloud, where it is received by network of data brokers. Once the live stream is established, ATLAS is then populated with this set and baguette style of data transmission without human intervention. So how does this look architecturally? We have leveraged the major Spotfire products with support from open source technologies, some TIBCO technologies, and some cloud native components. Primarily, this is Spotfire, where we deploy all the components, automation services, WebPlayer, Tare, R, and Python services. All our components are containerized, and we deploy everything, including Spotfire, to Kubernetes. Spotfire Copilot is also deployed to Kubernetes, so we use the same deployment and monitoring capabilities that we use for Atlas. We have integrated Spotfire Copilot with a large language models provided by Amazon, Bedrock, Anthropic, Cloud, Sonic. Having everything deployed in AWS means we can easily create this integration in a secure and performant way. Spotfire Copilot supports integrations to AWS and supports the use of MongoDB for our vector database. Liberty, as a company, likes to be leading edge, and we are a fast adopter of new technologies. We want to embrace AI to help and augment both our internal Spotfire developers and also to aid our field personnel and, eventually, our end customers. With this in mind, we are initially targeting the use of Spotfire Copilot to two personas. For our developers and analysts, we want to use Spotfire Copilot to boost our productivity in creating dashboards and data functions. In addition, we want the ability to expose Copilot to crews and engineers, allowing them to ask complex questions to gain deeper into deeper insights into the dashboards and data that they see. To date at this, we have loaded industry documents and Liberty Energy documentation, which we store in s three and use the data loading capabilities of Copilot to load these documents into our vector database. To allow users to ask industry specific technical questions and receive explanations in the context of decades of research with references to the data and eventually written industry papers. Okay. And so to the demonstration. We have created multiple demonstrations today that show how we leverage Copilot to help the personas that we have chosen. These personas are internal developers and crews and field personnel who potentially have different levels of knowledge, different levels of expertise, so they can use Copilot in many different ways. So the first demonstration, this is a single well view of a live fracking operation. We have tables and plots showcasing the real time data, not unlike what you'd see in a data van live in the field. A treatment plot at the top showing the major pressures and fluid rates and a chemical plot on the bottom showing the chemical concentrations within the fluid. For the user to obtain an overview of how the process functioning, a simple question can be asked that returns a summary of what the page is showing. Spotfire Copilot returns a detailed summary of what the current state of the fracking process is. The output looks at all the visualizations on the data behind them and provides the current values for each key metric. It provides some historical analysis of the time period chosen and any notable observations that need to be brought to the attention of the field personnel monitoring the operation. A simple use of Spotfire Copilot to start with that shows how users can get a similar of the data they are seeing in the context of the what the data and visualization is representing. Our crews in the field are very experienced, but this can be useful if they want a detailed explanation of what the data is telling them. The next demo shows an historic treatment plot, which is using historic data to give a summary of a stage of a well fracturing operation. This is a lot of information on this page, and it may not be fully clear if there were any anomalies or operational issues that users need to be aware of. This is again useful to feed personnel who want a quick summary about this stage operated. Did it run as expected? Were there any unexpected downtime? What the key what were the key treatment and chemical metrics? Again, we can ask for Spotfire Copilot to explain the page. Spotfire Copilot again summarizes all the main pressure trends, provides rate analysis to the flow of fluids, provides us some of all treatments, chemicals, and problems, and summarizes any notable events. You can see that there are a few operational issues during this operation. One of the key metrics that frac designers and engineers look out for is hydraulic efficiency. So what is hydraulic efficiency? Let's ask co pilot. So hydraulic efficiency is defined as the effectiveness of getting to target pump rate or slurry rate, which is the rate of the flow of the fluid containing the chemicals and proppants while not exceeding the designed treating pressure. This can also be represented as a time to max slurry rate and how well it maintains that rate once it is once it is reached. The hydraulic efficiency is the comparison between the idealized rate that is determined at design time and the rate that is actually achieved. It is a key metric in determining crew efficiency. So now we know what hydraulic hydraulic efficiency is and why time to max rate is important in determining crew productivity, we now want to use Spotfire Copilot to help us do the actual calculations. So we focus on the historical treatment plot again and ask Copilot to give us a summary of the historical treatment plot of that stage. This is to drive the conversation with Copilot from that visualization. What is key to concentrate on is the blue slurry rate line on the treatment plot at the top of the dashboard. Okay. So phrasing this question correctly, asking Copilot to use specific metrics, we can ask Copilot to calculate the time to max rate for this stage. We also ask it not to create a data function at this stage. Slutter rate is the blue line and Copilot Copilot will calculate that for us now. So Spotfire Copilot calculate that in about eighteen minutes, which is if you look at the plot, that looks broadly correct. There is a time period between fifteen o eight and fifteen twenty eight when the slurry rate is increased and then sustained at a consistent rate and stable rate, and the treating pressure remains constant at this time. We could use this value to compare with the ID layers rate for this stage and work out what the hydraulic efficiency was for that specific stage. So if we now switch to the developer persona, we have made the decision that the time to max rate should be provided as a key calculated metric for each stage summary and displayed as part of the visualization for that stage summary. So we can ask Copilot to create a data function that we can embed in visualization on this page. We ask the question to Copilot, and Copilot creates a data function for us with the instructions on what to do to enable this. We we must be very clear how we construct the question, the data to ask Copilot to use, and the format we want the output to be. So let's wait for this to happen. There is a little bit of manual work to use to correct the input parameter channel metrics and the time metric values. But as you can see, if we output the data function to a data table, we can see from the data canvas, it calculates the time to match rate by executing the data function. So if you look at the data canvas and we run the calculation, it's the same value that we had previous, around eighteen minutes. We can then create a simple table to display this part of the stage summary. So we can now see that using see how using Spotfire Copilot increases developer and analyst productivity. By constructing the question correctly and using Copilot's ability to write data functions, we now have a reusable data function usable data function that we've embedded in the visualization that calculates the time to max rate for each historical stage. There was a major increase in using the Spotfire Copilot to do this rather than developer or analyst manually unique for themselves. Finish that Another example of this is showing how we can calculate a key metric called flush volume and then using the data function created to calculate the flush volume for each stage when it's selected. Flush volume is the amount of fluid injected at the end of the stage to clean out the wellbore. We don't want to over flush as this can affect the wellbore region of fracture fracture. Flush volume is the difference between the final slew total and the slew total when the concentration of the fluid at the bottom of the wellbore goes to zero. So if you manually configure as we did previously, you can ask additional questions to call pilot how this data function works and what the outputs will be. And as you can see, as you select a new stage, then the flush for you, the difference is displayed based on the data function calculation. Again, another example of how we could use Copilot to increase analyst and developer productivity. Finally, one of the key metrics to calculate in the fracking process is ISIP, instantaneous shutting pressure. It's the pressure measured in the wellbore immediately after the pumps are shut off during the fracturing treatment. After friction pressure in the wellbore, a near wellbore region dissipates. ISIP is a key parameter for understanding fracture behavior and optimizing stimulation design. It is used as an input to calculate the frac gradient. The frac gradient is used to calculate the pressure needed to break the formation and initiate the fracture. But rather than use my explanation or what ISP is, we can ask Copilot. We can ask a broad question, and Copilot will use all information contained within our VectorDB to send the correct context and assist the LLM with generating a better answer. Spotfire Copilot generally always generates a data function unless you ask it not to. You then manually create and edit the data function in the usual way so that this value will then be calculated for each historic stage. We can take a look at the data function and check for any anomalies. We can then also create the visualization like we did before. That means that this is now calculatable for each historic stage. And as we swipe the different stages, then the ISP is recalculated each time. And if you look at the time of the plot, we can see that it's showing that the ISP just calculating is reasonable. Okay. So that is the final demo. Be before I am back to Ahmed, I want to thank the Spotfire team for their continued collaboration. As I said at the start, Libris likes to be forward thinking and use new techno knowledges. So the support we have received both with the use of Copilot and the use of Spotfire on the wider Atlas project from the Spotfire team has been very much appreciated. The partnership has been what's made this project such a success. Thank you. Thank you Craig, that was awesome. It's actually, very interesting how you I I love how you, slice it up by different personas, showing actual use cases, how, different different people in your organization, different types of use cases are enabled, and we are lucky to have, innovative and forward looking customers like like Liberty Energy. So we really appreciate the partnership as well. So let me go into the slideshow mode. Is my screen showing okay? I'm gonna assume so. Okay. So, let's talk about or wrap it up, the the features that we saw. So we saw a number of examples of, question and answer, general or Spotfire specific help questions or questions within the context of, proprietary and privately, uploaded and then processed documents for each organization, like what we saw, in the craze demo, about the installed definitions of of ISIP, for example. Interrogating data, you know, when when the user asks, you know, how long it takes to to to hit that pressure point, you know, the the answer pops out. Charts getting created through natural language, we saw several examples of that. And we saw several examples of data functions, Python code being generated. Most recently in May of twenty twenty five, we went generally available with Spotfire Copilot two point o. In this version, we added support for creation of nine new chart types. So we basically cover creation automatic creation of, all of the native Spotfire chart types. We have improved under the hood how data questions are answered. Basically we refactor the whole, the whole process, in a way that, we we have the the large language model generated query that runs on your data where your data is in Spotfire without, shipping a large amount of data to the LOM. Explain page, is is is a new feature, much asked for. Before this, we only, explain one visual, one chart at a time. Now we can generate reports, off of the whole page page. The the UI and UX has been improved significantly, more consistent with the rest of, Spotfire and smoother. The the one before last, support, web clients that was, that was very wise. We asked for probably, arguably more than any other feature on the screen. Before this, we only supported the the the client, the analyst. Now, we can serve, all the, all the consumers or typically web users. And, we are now officially supporting, more large language models, because we know that our customers in you guys, are not on one, but, basically, the the the three major ones, Microsoft, Azure OpenAI or, AWS Bedrock, you know, the, cloud models. We we typically find better quality, better, better performance if you like for for the LN performance and the GCP Vertex AI with their Gemini model. And this is the, the architecture. Or yeah. So I'm gonna spend a minute or two walking through, following a question. Basically, what happens? Where data goes? That's much asked for. And also tell you exactly what components come with or make Spotfire Copilot and what each company, each organization need to provide. So bottom right is everything we saw to anchor ourselves. That's Spotfire, canvas that the end users, is is lives there, sends the questions there. Let's say let's start from, I don't know, data interrogation question. You know, what's the, what's the, highest CPK value, for example? It first of all, it it's processed through this, package that's added to Spotfire. So that's the first component main component of Spotfire Copilot. That's a CPK or, sorry, SPK file or package file added to your Spotfire server. It's then shipped to the orchestrator, which is the you can think of as a quarterback or the the director sitting in the middle. That's a container, or containerized image where you can, you know, you can run it wherever you run your containers. This is the second main component of Spotfire Copilot, that you get from Spotfire. It then, at that point, tries to infer, what, what is it that the that the user wants to do. In this case, it, uses the LMM itself based on the question when when it was asked, what's the highest CPK value, it infers that it is a data interrogation question. It puts it on that branch of behavior that, hey, this is about answering your question based on existing existing data. At that point, off or continuing that branch of behavior, the orchestrator decides that or follows this pattern that to answer this question, it works with the LMS again to generate a query based on some metadata that's shipped to the LLM. So the LLM knows, the shape of the table, and together with the question that the end user asked and along with a minute detailed system prompt that we provide to the large language model, the query comes back. We have deployed a number of checks and and quality assurance under the hood, tests that runs in real time, in the orchestrator together with Spotfire. And, once all of these checks are passed and it the orchestrator makes sure that the query is good and it executes, it sends it back to Spotfire. It runs on the data. The final answer is is produced and it gets, again, you're packaged up or or padded with human language additions and it's shown to the end user. So, and all of this happens within the four, five, six, eight seconds that the user is basically looking at the screen waiting. So it's as you see, it's a it's a close collaboration integration between Spotfire, the orchestrator, and the large language model. The left hand side of the screen is for the cases where the user wants to ask questions in the context of the documents. For example, the definition of the ISIP that we saw in in Craig's demo or the, you know, what is considered a good CPK based on our own internal documents. It can be your own definitions of chemical, processes or industrial values or process definitions, as as you wish. You can all, process them, upload them here and process them offline. In real time, they become they they they sit there available for the orchestrator to tap into. So if the orchestrator infers that the question is about, a context here, it consults with these, documents first or indexed, resources first. And these are, by the way, two examples, Azure Cognitive Search or, or any vector DB. It's it's again agnostic. The top end results, typically five or 10, are then picked and then we send the orchestrator sends the original question along with those, search results to the LLM and asks it to answer the question within the context of those search results. So basically, the process of generating the answer is augmented, through the, the the retrieved results from these documents. So it's Retrieval Augmented Generation or RAC, if you have heard heard about RAC. So, this is basically the the flow of, your data, the questions, you know, where each component comes in. And again, to summarize, when you deploy Spotfire Copilot, you get a SPK file added to your Spotfire server, a, image of a container for the orchestrator. That's what Spotfire provides. Each customer, each organization provides their own large language model. And, you know, these are a few examples that we that we've seen, much used. We also support, AWS, as well. As well as in any vector database, we we do here or any shared service. And, a quick look at the, you know, the cutting edge, the R and D activities without going deep into the details, this this one to make sure, we're on the same page. This is a future state. So we are, working very hard on, deploying or leveraging the agentic architecture, to make things more efficient. And also as you see in the middle box here where you see Spotfire, TDD, life science energy, you know, the the good you know, student observers may see that, you know, these are today. By today, definitions are apples and oranges. But at the same time, it's the new universe where, you know, many functions or skills are offered and promoted as as a as a skill or agent or a tool. So that's the the future that we're envisioning. We're, we offer, in a scalable and fast fashion, all these skills. And at the same time, we enable each customer or partner, to add extend Spotfire Co, by adding the the skills. It can be some a skill or a agent that's, optimized for predicting ISIP in this case or or for life science or any other, vertical. And, here are a few resources. As a quick reminder, in the, this webinar platform, the right hand side, there's a doc section that you can see actually with these links.