Name: What’s New: advancing visual industrial analytics with Spotfire® AI
Uploaded: 2026-05-05T16:16:21.139Z
Duration: 55 min 49 s
Description: What’s New: advancing visual industrial analytics with Spotfire® AI

Transcript for "What’s New: advancing visual industrial analytics with Spotfire® AI": Hello, everyone, and welcome to today's what's new webinar, Spotfire AI, the agentic evolution. We're thrilled to have you with us. I'm Jean-Philippe Richard-Charman, and I'll be your host for today's session. Now before we get started, I just wanted to cover a few housekeeping items just to ensure that you have the best experience with us today. Now the webinar will last for around forty five minutes with ten to fifteen minutes at the end for our question and answer segment. If you have any questions during today's presentation, please do not hesitate to use the q and a panel, which is located on the right side of your screen, and we'll address as many questions as we can during the q and a segment at the end. We've also made a few assets linked to today's webinar available in the doc section of our webinar platform, so please feel free to access these. And that's located on the right hand side right next to the q and a tab. 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. With that, let's start dive right in. And I'm excited to introduce our presenters today. Michael O'Connell, our chief analytics officer, Niklas Amberntsson, our director of product of product management, and Vaibhav Gedigeri, our principal technical product manager here at Spotfire. Now with that, I'll hand it over to Michael to dive right in. This is a platform that includes, domain specific models for physical assets, with high compute engineering data, Spotfire and AI and analytics for data discovery, what if scenarios, and confident decisions, and basically gives you a platform to understand what's happening and why so that engineers and scientists can act sooner. It allows you to derisk large investments, monitor business, and operate, optimize operations. The video running in the background here is our is our website, and I invite you to take a look at that. Okay. So, Spotfire and GenAI. This is a set of tools and applications, that act as AI assistants for and and engineers, analysts, and energy and manufacturing with, ambient chat through the Spotfire Copilot, interactive click, with Spotfire Insight agents, and an agentic foundation for creating, managing, and extending the agents. So these include industry use cases, operations workflows, industry frameworks such as ADK and Strand, Microsoft Asset Framework, integration via a two a and MCP. All of our agents are a two a compliant, and we will integrate with other external a two a compliant agents. And it features a chat with data visuals and workflows, around domain rich visualizations in Spotfire, code and visual generation through Spotfire, explanations of visualizations and analysis, and along with the, LM problem context. So this chat with data and visuals and workflows, I'm gonna get into a little bit in the context of the screenshot on the top there, an industry workflow, around well recompletions. And the bottom screenshot is a from our one of our inside agents, which is an interactive click agent within the Spotfire authoring experience. Okay. So the agents, that we have been working on, are bucketed into three groups. They're insight agents for visual exploratory analysis, analysis summaries, data science interactive things like outlier detection, root cause analysis. Second category is adviser agents. So these execute workflows, make recommendations. They're physics informed. For example, production anomalies, with rules based or or digital twin based, tolerance limits. Examples we're gonna show today, well recompletions and asset valuation in the energy sector. And, and then the ecosystem agents, the third bucket, this allows you to, through MCP, typically, access things like Databricks, data virtualization, OSDU data models, even the Spotfire library and the Spotfire license manager, which have MCP servers to them, and tools like Tabley that allow you to access broader communities. So this is very much an agentic multimodal, rag scenario. So you see the, in the top there, the the user in the chat interface makes a user query. The ELM, analyzes the query. Typically, it determines the the intent, and then figures out, do I need additional answer for to additional data to answer this question? The answer is no in a lot of cases. You jump jump down here. You go to the LM with the appropriate tokens, and then push the the answer back to the to the end user. However, if, you do need additional data for the answer, then we'll spawn an agent, go and hit some MCP server like Databricks, or go and hit some other foundation model, and bring back the results and then go back through that loop. So Google and others are referring to this as an agentic multimodal rag system, and that's that's what we do, in the, in the chat interface. And now we very much believe in expert in the loop. We aren't that big on autonomous agents. We have great trust in our customer engineers and scientists. Rather, we are the go to for transparent and immersive evaluation of agent flows. With the immersive experience in Spotfire, you can take a look at what the agent's recommending, and and you can go and see for yourself. Drill up, drill down, drill sideways, spot the fire in the data, that sort of thing. Okay. So one example that Vaibhav gonna show is this recompletions advisor, and this is for identifying wells for potential recompletions using available internal and public data. He's gonna show you some data from, the Williston Basin and give you a demonstration about how that works. So that's an advisor agent, that he's gonna show you. Now Niklas is gonna show you a couple of insight agents that, analyze relationships between subsurface and production or help, construct geoscience analyses on wells and regions or look at decline curves. So in Niklas is gonna give you a demonstration of a couple of insight agents. And then the ecosystem agents, I think Vaibhav is gonna show that, as needed, we'll call out to additional databases for supplemental data from additional systems. And so that's how those three types of agents fit together, in the building and orchestration of, of these analyses, in this case, for, recompletions. Okay. So just introducing Spotfire Copilot, and and we're kinda launching this today. This is, version two dot three is coming out. It's, our gentrified, solution. But the Copilot is a conversational chat interface within the Spotfire UX. It it it leverages an intent chain architecture. You can where you type a question, we figure out what the intent of the question is, and then we get the appropriate tokens, and, or industry corpus data, bring that in, use any LM to to get an answer back from that and have a a threaded conversation. It's a pretty efficient visual first approach, multimodal q and a, allows users to create, explain, and query visualizations, create data science functions, Python, and data with context aware spaces. So it's definitely a subject, expert, AI collaboration environment with a lot of domain expertise, lots of privacy and trust, so secure and governed access to the data, and then scalable agility. You can run, on premises or you can run through cloud provider. Now over the past couple years, we've been working on the Spotfire Copilot now for a couple years, ever since we got access to to to OpenAI through through Microsoft Azure, actually. But, basically, we've been doing things like Spotfire product q and a to flatten the learning curve of using Spotify, generate code to simplify developer workflows, generate visualizations. So basically productivity using Spotfire. That was our initial focus. Now with this release, we're we're moving more away from, just that sort of productivity into actual workflows and advisers. So our agentic advisors accelerate decision making. You'll see an example of that. Connecting to corpuses and databases to provide industry expertise, often through our ecosystem agents, and then getting conversational insights to quickly quickly, interrogate industrial data and get back relevant answers. You know, the LLMs know pretty much a lot about everything, and so getting that having that context into the analysis is is very useful. Now on the inside agent side of the house, again, the tools and calculations using empirical stats and distributions, immersive immersive visual discovery, incorporation of some physics based models, digital twin type things, but, all of that generating insights in context of your visual analysis. But we're also moving the inside agents into a more of a use case centric approach. So, you'll see Niklas talk a bit about, our work with yield analysis and spatial patterns of zones in wafer maps, work with asset valuation, combining subsurface well logs, into with production and to assess so they're very similar to the use case on, on recompletions. And then production optimization, physics based anomalies such as pressure, vibration, and flow, and looking at outliers safe there in decline curves to understand the the root cause. So let me start off with a a demonstration of the previous version of Spotfire Copilot, and this is the foundation. You know, Vipav and Nicholas will give demos on on top of that with the the work we've been doing over the last year. So this is a manufacturing yield analysis. The top graph here, you've got yield on the y axis. You've got the lot ID on the x axis. The top right graph, you've got the wafer yield, as a more of a distribution by lot with a, advanced box plot type of approach. And now we're gonna interactively explore that with the help of the Spotfire Copilot. So let me highlight that area with low yield at the top left, and I can do the wafer maps underneath that. And you can see that those wafers have some red zones around the edge, so a bit of an edgering pattern. And then I've added, an individual wafer bin map to do the electrical testing, you know, at that, you know, at the dye level. So let's see. I'm gonna have to start a conversation here, make me a bar chart showing average yield by x coordinate. So looking across, the wafer, and I can see quickly get back a visualization generated by Spotfire Copilot. And you can see that there's, at the edges, there's a lower yield than across the rest of the, across the rest of the the the data or the the the the slice across the wafer. So now I'm gonna ask a question, you know, create a stacked bar chart showing the count of bin names per lot. So these are the bin tests, the electrical tests that are done, on the wafer, and I can see that, some of those lots don't do so well. There's some lots there that have pretty low results, little low stack bar charts. But each of the colors is a particular category of electrical test, and you can see that Spotfire has given you a nice summary there, of all of the the performance, of all of the lot IDs against this bevy of, like, 10 or 12, electrical tests. So it's a nice nice summary of that that you can get just very quickly, from the chat interface. So, now we've got, you know, a little bit of knowledge about it. Now we want to create a data science function in Python, to further our knowledge here about what is the wafer pattern as a function of distance from the center, in this case, just a general pattern. We wanna find out what are the patterns of of, of low yield across the across the wafer. So, basically, I've asked it to create a to create a data function that calculates distance from the center, of the wafer to every die. So, the Spotfire Copilot has created that data function, Python code. It's put it right into our data function interface. It's allowing you now to map the input and output parameters to the data that you have in your analysis, map that Python code. So I'm just putting the x column and the y column, as as in input parameters, to the data function. And now I'm gonna output, you know, an extra column, for the the calculation. And you can see that that column has has been set up. Now this is a, you know, workflow here that's basically a transparent lineage of what has happened here, reading in the data, you know, producing the calculations, adding it to adding the column, to the existing data table. So that's all handled just directly through the the Spotfire Copilot. You don't have to create your own data function or anything like that. So, you know, then now we're gonna create a scatterplot that shows the average yield by distance from the center. Because now we have that distance from the center for every die location, we can start to use it in understanding patterns across the wafer. So the Spotfire Copilot comes back with that visual that I've requested and then just click it right into the analysis. Now something interesting happens here. We knew that there was, the the the on the far right hand side of this scatter plot, that there was low yield, low yield on the y axis, low value there, and but we didn't know the trend. So it's sort of trending towards that. And then we also didn't know that there was really low yield right in the middle of the of the wafer as well. So that now you've got, you know, quite a bit of context and and pattern about the the the yield there. Niklas is gonna pick this up a little bit later, but, oftentimes, the the the the plates gets hot in the center and the edge, and that can you know, various conditions on the equipment can lead to this type of a pattern. Okay. So let's move on, and talk a little bit more about Spotfire Copilot two three two dot three. As I as I mentioned, we just just released, literally this week. And it's a it's an approach that allows you to bring in industrial agents and advisers and create them yourself, but also integrate with other systems using a two a. It allows you to augment data from external sources, you know, via, you know, ecosystem agents, has a lot of industry context, does chat history persistence, intelligent caching, has an admin console with enhanced observability for Spotfire So just, getting a little bit deeper into, the release for two dot three as you saw above. But two dot three, our focus has really been moving more towards, agents and advisor type agents, but also focusing on a few other things, things like data augmentation and how you could create and add data from, external sources. In the past, you were able to query the data, but with two dot three, you're now able to go one step further of being able to then get, that data into Spotfire as structured tables for further downstream analysis. We've, we heard the feedback that we got from our early adopters and, one of the most, top what it wants was the persistence of chat history across tabs. And with 2.3, you will be able to persist that chat history across tabs and across sessions. Finally, we're also introducing, an admin console for enhancing observability for the Spotfire admins where you can, view all historical, conversations across sessions. You can register and invoke your a to a agents and even manage, OAuth, credential all in one place. So going deeper into each of these. Right? So, for our industry adviser agents, as I said, you can now develop, register, and invoke industry and use case specific a to a agent. With 2.3, it it comes with a well recompletions, advisor, which I'm just gonna, demo in a second. You could also register your own a to a agents, via the admin console, with, agents in Spotfire. Now, we will provide extensible UX workflows for agents, meaning you could invoke these agents via markings, text, or data tables. Finally, just emphasizing on what Michael said, we, you will have the ability to evaluate the agent recommendations as you can see here on on my right. The well with completions advisors returns, certain wells that that would be, candidates for reentry. And at the same time, you could using Spotfire's immersive UX, you I could just plot that word for a ratio by production to just verify, the, the recommendation, of the agent itself. Data augmentation. As we all know, enterprise data, a lot of enterprise data is, unstructured and is stored in external sources like vector databases, graph databases, and other relational databases. With 2.3, you can now ingest unstructured data, from vendor PDFs, field reports, logs, and even data center sources like Databricks. And what you could do with that is you could then return them as structured tables where you can query syndicated public data, or private company, corpuses. Finally, the admin console. So with admin console, you now have, the enhanced observability. Kind of replaces that, previous manual log analysis, with with deep user chat learnings, and and you have more actionable behavioral insight across, all sessions. You could also, as as we as you look on the right here, there's that central, unified agent registry to oversee Copilot conversation, even look at your RAG indexes, and even monitor your user's best specified agents. With 2.3, we also, provide auto summarization, and this is a fairly important, feature behind the scenes where we are using the concept of context compaction, which will, feature asynchronous automatic conversation summaries. Alright. So I'm gonna go ahead and switch my screen share here and over to my Spotfire screen real quick. And, okay. Alright. Okay. So so here, I've created a a recompletions, DXP here. This is, the data that we're looking at is the Wilson Basin. It's a large sedimentary basin, that's spanning Western North Dakota, Eastern Montana, and and even the Saskatchewan region. It contains multiple stacked oil bearing formations. So, going from, top left, you could see that we've plotted these, wells on a map, which I'm gonna use, while invoking our well completions agent. Right below it, we're visualizing the interpolated depth, of all these wells, through our interactive map layers. And then, we have the, well log, which will be, used to, evaluate, per se the recommendations that we get from, our agent and, the, the visualization to, the top top right, which is a water to oil ratio by production, we're gonna have Spotfire Copilot actually create that for us, as well. So let's get into Spotfire Copilot. With two dot three, one of the other changes that we've made is, to our, UI and and UX. As you can see when Michael was, sharing his demo, we've moved away from our mode selection, to more of an explicit, intent. So you can now invoke all the intents by just click, hitting on your backslash, and you can see that, we have a bunch of agents here. And as you scroll down, you can, scroll down to specific intents, be it creating a data function, visual, explaining the structure of your data, explaining your visuals, querying your knowledge bases, and and asking questions of your data. So in this case, I'm interested in the WellWork Completions Advisor. So I'm gonna go ahead and, invoke that, press enter, and we're gonna get into that mode. You could see that the WellWork Completions Advisor is a multi turn agent, so it's gonna wait for my answer, or wait for my inputs, as, as it responds. So we're gonna invoke this agent by actually selecting certain wells that we're interested in, evaluating. So I'm gonna go ahead here on our top right, and I'm gonna pick a few wells. Let's just maybe pick from this region. And there you go. I've marked a few wells, and then I'm gonna come back in here. And then I can choose to send the mark data into the adviser agent, by actually just hanging on the pound sign. And you can see that with the well index, I have about 61 rows about 61 wells. I'm gonna go ahead and actually send that. And what it's doing at this point is it's the agent is evaluating the wells, based on, certain criterias. It's taking into account, things like original productivity, water to oil ratio, formation targets, I believe, total depth, proximity to proven reentries, field membership, and more. And and what it what it'll do is it'll finally provide a rating, and this would be either, you know, the wells are a good candidate for reentry or a strong candidate for reentry, and the action here would be really to just proceed to detailed engineering evaluation. And then the others would be to evaluate. As you can see, that's by the amber sign. This would require further investigation and further investigation could just mean here acquiring more data. This could be, you know, the seismic data or it's just performing more integrity test. And the third one, as you can see here, is you just wanna leave them as plugged in abandoned. There is no action to be taken here. These are not a good candidate. It also returns, the table as you can see here. I get a table here with, all the columns. And towards the end, you can kinda see that, the agent has returned a rating, with a justification as well, on why it made those ratings. You also get the geological context, just kind of telling you, these are targeting that, Madison formation. And then there are deeper targets such as the back end formation, which would be, viable for reentry as well. So now that I have this table, what I'm gonna go ahead and do is I'm gonna go ahead and add this to our data canvas, and this is gonna be important as we visualize. And you can see here on the right, it just created this new table, and I can then use this table to visualize further. But what I'm gonna do here is first have Copilot actually create, a table relation, which will, help me with cross filtering, across visualizations as well. So I can actually have Spotfire suggest what the table relations are. And what it does is it's gonna look at this newly generated, output that we just created via Copilot. And it's gonna look at all the other tables in our analysis, and it's gonna generate certain relationships. And you can pick any relationship that that suits, the best. And there you go. It returns with a few here. You can see here, looking into this in detail, you can see here I can I can create table relations for well indexes? I could do this for the tops data. I could do this for the well sticks data and others. I'm probably gonna go ahead and do it, with the well index, and you can see they here that it's created that table relation for me. Now what I'm gonna do is I'm gonna have it, basically visualize these results for me and, say, now visualize these results for me. What's important to note here is that, you know, I'm not really telling it, you know, what what and how it this should be visualized. It it it kind of understands that, you know, I'm looking at, you know, you I could do a scatter plot, of cumulative oil, you know, versus a water to oil ratio. So I'm gonna just say, the first one, see how this works. Yep. Just give me the steps. Create the create the scatter plot for me. Alright. Now I can see that what's interesting is it sees that there are certain visuals on this page, on the page. It also notices that there is a current, scatter plot which is on the page. So it's actually asking me, do you want me to just modify that or do you want me to create a new one? So what I'm gonna do is I'm gonna have it create let's see if we can just, get rid of that and have it create a new one for us. And there you go. So you can see here that it created that. I'm gonna go ahead and bring that in. Let me just change this right here. And right away, you could see that looking at this, it's visualizing the water to oil ratio by, production. And you could see here that the green ones are the ones that are a good potential candidate for reentry. And it kinda makes sense because those are the ones with, lower water to oil ratio and and higher production. And you can see that there are some that that that may need to be evaluated a little bit further. And the ones that have very high word to oil ratio, these are just dry and you wanna leave them plugged in abandoned. Now as a sec as as a next step, what what you could definitely do is you could then evaluate this further by then just creating a a wet log, mod. With Industry Pro now, you can now use, actions, to create these well log mods. I've already created one here, but I can quickly show you how you could do this. I pinned a few of our, actions and you could you could see here I could create a triple combo, quad combo, custom view. With the Wilson data, we only have the gamma rate here. So, we we don't really have data for, neutron density or resistivity. So I'm not gonna choose these two. I can just create a custom view here. I can go in here and just say run action, and now you get this really nice, UX, workflow here where I could just go in here, select, the data table. This is a well log. I'm gonna go ahead and select the column. See here we have the gamma ray. I'm just gonna add it in here, and that's that's there. And then I'm gonna go and pick the depth. Just gonna look at the depth. I'm gonna see the well log and we have the d e p t column. I'm just gonna go ahead and run this action. And you can see here, it just created, that that well log mod for me. And from here, you can pretty much analyze the gamma rays and you could look at other measurements to reconfirm, the recommendations that were made by the agent. Now quickly, pivoting over, I'm gonna look at, how we can augment, data. Now we're in this well, we're completions advisor, so I'm gonna come out of this mode and I'm gonna try and look at another agent that we have, which is our drilling report agent. And I'm gonna while this is data from a different different set of wells, I'm just gonna ask a question in interest of time. We're just gonna copy this here. I'm gonna try and just create a table for for this particular well where I wanna look at the drill depth per day. Now for context, what does this data look like behind the scene? If you look behind the scene, this is an example of what a drilling report looks like. You could see how unstructured this is. You you have you have these drilling reports created per day, and you can see you have a bunch of fields and you have some unstructured fields in here in terms of the summary of the activities. And you can then have Copilot just create this table. And then you could then further add that table within your Spotfire analysis so that you can then analyze it further. And this could be you could create more visualization, and then and then use Copilot to even further explain those visuals as well. You can see here that it's just invoking the DDR agent, and it will return well, I hope that it returns a table shortly. And meanwhile, that's happening. I'm gonna go ahead and actually invoke or show us our admin console. Yeah. I'm just gonna log in to our admin console here real quick. I'm gonna go back into a Spotfire. Oops. Fifteen nineteen a. Well, it's probably let's just try that again. I'm gonna go in here. Say how many reports reports are there. Alright. Alright. So you see that we do have 165 reports, that we've indexed. So now I'm gonna just go back in here. Let me try and ask the question again. Probably see a small typo in here. See if I can fix that. And there you go. That was just a simple space. But you can see here that it returns a table where you can look at the report ID, the period, the rig name, and you can see that I get this distance drilled. So I'm gonna go ahead and add this as a new table. Just quickly show you how easy that is, and you can now see that I have a whole new dataset that that is sitting as as a PDF file, and now we have this as a structured table within Spotfire. So I'm now gonna move over, to showing our admin console and shifting gears to our admin persona. With two dot three, Spotfire Copilot will have its dedicated admin console. And moving from left to right, you can see here right away, you get a very high level summary of the total conversations across all the session. You can look at the system health. I can also try and, drill into any particular conversation of interest. Right? You can go in here. You can see the full thread of the conversation. You could also see what context was sent to the LLM. And also, one of the good parts about the Spotfire admin console, something that you would not see in any of the external observability tool is you can actually view what is exactly displayed in the client. And this could mean that if there are certain visualizations that you're, you know, that, that may be displayed, that would all show up, in here, all all within this client panel window. As I said, you could, you can now register and invoke and test actually, register test and invoke your agents within the admin console. You can see here we have, seven total agents running right now, kinda displays the health of it as well. I just showed you the well recompletions, but we have a few others, that that that we are actively working on, the well declined, the petrophysical analyst agent, the ecosystem agents that Michael mentioned, Databricks. Spotfire admins can also now monitor all the RAG indexes in one place. This is massive. This is a this is a big, leap from what it was previously because, ideally, the RAG indexes, depending upon how you deploy, Copilot, you'll have to pretty much go to the source and, and actually keep a track of them manually. We've kind of removed all of that and you can now, view all your RAG indexes in one place. You can also configure user management, and, security. You can also look at system logs. This was, this was fairly inconvenient in the past where you would have to go into, you know, wherever you've deployed your Spotfire CoPilot back end infrastructure. This could be in the cloud or on premise. You'd have to pretty much manually, get all of those orchestrator logs, but now it's all in the admin console. And you can keep a track of all your OAuth, clients and credentials. Hey. Great demo, Vaibhav. That's wonderful. So I'm, gonna go ahead and, stop my share and, hand it over to, Niklas. Let me just make sure. Thanks a lot, Vaibhav and Michael. So all the great things that Michael and Vaibhav spoke about here is, is mainly about Copilot 2.3, which is, our latest and greatest of Copilots. So I'm gonna now speak about what we call visual AI and the insight agents as Michael also mentioned. So visual AI is the next chapter of AI in Spotfire, and we think of it as a way to accelerate how experts work powered by AI. So as Michael said early on, we do not try to use AI to replace experts. We want to amplify the analytical capabilities of the industry experts and give them more time to use their industry knowledge. We embed AI directly into the visualization experience so AI can surface patterns and signals and suggest insights and next steps of analysis in context. It's not a black box AI. Every AI recommendation can be reviewed, accepted, refined, reused, or ignored. So the expert stays in control. And this is important when making decisions that directly impact the profitability of the company. So visual AI insights agents combine AI with domain knowledge, statistical techniques, and industry standard methods, so recommendations are really context aware. As you will see soon, we are providing industry native agents for energy and semiconductor manufacturing but we also plan to make visual AI agents extensible so your organization can embed its own models, logic, standards, and knowledge in custom AI agents that are integrated directly into the Spotfire user experience. So visual AI is built upon AI agents with domain and analytical knowledge that surface findings and recommendations to the user directly in the interactive Spotfire user experience. The agents will analyze data, look for trends, patterns, and outliers, and reports its finding, also make recommendations for the next step. So the agent asks the user to provide direction and guidance when needed. For example, may ask the user to choose between alternatives or provide input in natural language. So based on the guidance from the user, the agent may focus their analysis more and provide more relevant insights, to amplify the user's analytical expertise. AI agents are delivered in the AI agent registry that we are, also working on updating, which can be accessed through the add on browser directly in a product, just like for visualization modes and actions. And now we're building AI agents, making them available in the same way. So this means continuous AI innovation with new agents being delivered continuously to users between Spotify product releases. As mentioned, we do plan to do specific AI agents for manufacturing and energy. And in semiconductor manufacturing, Spotfire agents helps users understand patterns, anomalies, and trends in the huge datasets generated in manufacturing of semiconductors. So this helps users identifying root causes more quickly, in the end, improving yield of good products from the process. The wafer analytics agents provide wafer aware data preparation and enrichment recommendations, specific analytical methods that take the structure of wafer data into account, and using wafer specific visualizations in the recommendations. In the oil and gas asset valuation case, there is a need to quickly analyze available data to estimate the ROI for the possible investment. So the agents helps cleaning, combining, and transforming data to support the analysis, making the analysis faster and leaving more time for estimating the return on investment. The agents recommend visualizing data in well logs decline curves and other familiar oil and gas native visualizations supporting decisions with huge impact on financials of the company. Oil and gas companies must constantly analyze and adjust production variables in order to optimize recovery of hydrocarbons and keep operational costs low while ensuring safe and sustainable operations. So users constantly need to make trade off decisions that directly affect profitability of the operations. Often data that support these activities is incomplete siloed and uncertain in nature. The production optimization agents help preparing this data and analyzes aspects such as the well head pressure, flow rates, vibrations, and other data available in order to indicate possible needs for equipment maintenance or well intervention contributing to extending the economic lifetime of wells and keeping the estimated reserves accurate. So while we are providing industry native agents for semiconductor manufacturing and oil and gas, we know that some of you want to build your own agents and use them in Spotfire. So as we usually do, we open up for customers and partners to extend also the AI capabilities of Spotfire by adding custom AI agents that integrate seamlessly inside the Spotfire user experience. AI agents are developed similarly like mods and may use the LLM data in the Spotfire document, calculations of the Spotfire data engine, data functions, data from information links and other library assets. This means you can build your own insight agents that incorporate your company specific rules and knowledge and turn them into AI recommendations for the user. Let's now take a quick look at visual AI in a demo. So here we are looking at the user interface with the Insight panel on the right. You see there is a button in the top of the insight panel, which is split. So I can choose individual insight agents by clicking on the little down arrow, Or I can just push the entire button in case I want all the insight agents to use. So the data that we're looking at here is production data from oil wells. So if we look a little bit more on the visualizations we have in the analysis here, you notice that, hey, there is something which doesn't look really right. So we mark that part and you can see that we also have the option here to actually, invoke Insight agents directly from a floating toolbar in the top of the visualization. So here, we'll select to use the well production outlier Insight agent and see what it can tell us. So you can see that the agent actually immediately asks us a question whether what kind of data this is. We confirm it's oil well data. And the insight agent then actually immediately detects here that I do actually have a data table that is that could be used to explain the reason for, the outliers in the visualization that I marked. So the insight agent actually suggests that I create a relation to another data table, which we'll accept. And that means that the insight agent now is creating relationships under the hood, and the agent continues analyzing data. It's gathering insights and then comes up with a summary here that it think there are underperformance due to high motor temperature, pressure fluctuations, and so on. And we can click on that summary in order to see a much more detailed view of the different insights. So there are much more details in the top, overview of well production performance and outlier analysis. Then comes a recommendation here to create a well decline curve. But I'm most interested in the correlated factors that we see here. So high motor temperature, flow line pressure fluctuations, increased pump vibration, and the intake pressure variability. So I'm going to go ahead here and create all of these recommended visualizations. So these parameters are now visualized as box plots which is showing the difference between the inliers and the outliers in the data. And we can clearly see here that the there certainly is a difference in the, in all these, factors between the, the the marked outliers and the, the the inliers. So but how can we make sure that this is statistically significant? Well, using the new visualization properties panel, we can actually just select those visualizations and then choose to add comparison circles to the, to the visualizations. And that actually shows us that all the all these comparison circles are non overlapping, meaning that the finding here is statistically significant. So that's a quick preview of our planned AI Insight agents support integrated in the visualization experience. It's industry use case aware and with the ability to build your own custom agents. So I'm looking forward to tell you more about this in other context and other, events soon. And don't hesitate to reach out if you want to discuss. We are very interested in hearing about your AI at ease. So with that, I'll hand over to Michael for wrapping this up. Terrific. Thank you, Niklas and Vaibhav for great presentations, demonstrations. Very good. Okay. So just to to wrap up, after two great demos from Vaibhav and Niklas, what are the recent trends, and what are we looking at to the future? It's been a fascinating last six months. I get the sense that we don't really need prompt engineers anymore. That was a big thing in the early days of GenAI, but, the sophistication and the multimodal nature of communications with large language models since that that doesn't really need it. Intent based architectures are everywhere. So when you type in the chat, figuring out what the intent is and doing something appropriate, that wasn't common. We were a very early adopter of that. I think our very first version of Spotify Copilot, which is based on six intents, but now any intent is gonna accommodate it. We're seeing the evolution of RAG to agentic RAG. In fact, agentic multimodal RAG is what we're targeting with our Spotify Copilot. There's a Google developer blog written on written on that actually. So vector and graph databases, still a pro avail, as tokenized to ways of communicating with data. A lot of folks are now talking about general knowledge bases, and the ability we've got through our ecosystem agents to go get data from somewhere else. You look at that right hand side of that agenda to the multimodal RAG graph, and you can see anything that has an MCP server, we can make a query, from with an HMA compliant agent to get, you know, supplemental data. So Databricks has been mentioned a few times. Databricks has a nice MCP server, and we we've been able to supplement data through that route. Foundation models. So they're a lot more fit for purpose models coming around, you know, custom trains for specific users, is, you know, manufacturing models there. And I've been we've been working with some other energy sector, foundation models, and we'll be this is a bit energy centric, this presentation, because we have our energy forum next week. But we'll be announcing some other partners there. And then with the LLM, SLM proliferation, you know, local on prem, you know, a lot of activity with Gorilla and Llama, and the deployment with OLama. A lot of that stuff is now managed on Hugging Face. There's an enormous amount of local and, SLMs and targeted SLMs. But, definitely, we're going forward with this agentic, multimodal architecture, RAG architecture for our Spotify Copilot. Let's see. And in terms of standards, to the future, we've been really focused on, building everything with standards. You know, the Google ADK is pretty strong, developer agent development kit that we're we're standardizing on. A two a is our standardized protocol that we're using for integration and back and forth with other agents. You know, the intent prompt, is a is a way of saving, an agent. We we're looking at that, storing that, and definitely storing Python code that we're getting. And, typically, our folks are generating such agents using combination of Versus code, you know, along with, GitHub Copilot and Claude, to to develop the agents. And we're happy to show folks how to do that, but then we're storing and managing them in these different in these different ways. In addition to the coding standards that I just mentioned, there's the agent framework standards, you know, ADK strands, Microsoft Microsoft asset framework, LandGraph. So we're we're we're adhering to all those standards, and then sharing in the, our agents in in our community. We're looking at sharing agents as code, agents as prompts that I mentioned, and the number of registries that we want to use to make those available to to you and, and and the extensibility points available to you. So we've put a few links here on the presentation, to wrap up on, where you can find the latest updates on Spotfire Copilot, how where to download it, the support portal. So Copilot is now fully supported by our support organization. And so you can just file your regular tickets and, and get support, that way. So any other questions you have, please reach out to your account team or email us at data science, @spotfire.com. And thanks again, Vaibhav and Niklas, for a great demonstrations today. I'll turn it back now to Jean-Philippe to do the wrap up. Beautiful. Thank you very much, Michael. And thank you once again to yourself, Michael, to Vaibhav and Niklas for today's great presentation. Now just a few little things that we wanted to mention, before we wrap up today's session. So as you can clearly see on screen and as Michael just mentioned, we have our Spotfire energy forum taking place next week in Houston, Texas. If you haven't already registered, please do, and we look forward to welcoming you at the energy forum in Houston. Now if you wouldn't mind going over to the next slide, please, Michael. So, you know, just before we get on to the q and a segment, there's just a few things that we wanted to share. As previously mentioned, you know, we've got two great webinar series that are being out to on a regular basis. So So whether you're looking to find out more about Spotfire or looking to learn about the latest in terms of what's new, don't hesitate to register to the full series. On demand access, as mentioned earlier on, a recording of today's webinar will be available soon, so please do keep an eye on your inbox for the link. If you wouldn't mind going to the next slide, please, Michael. Now in terms of next steps, if you're interested in learning more, please feel free to visit our website at spotfire.com or contact us directly. There are lots of ways to interact with us, whether it is via our socials, through our community. Additionally, our blog site has lots of great content where we share the latest in visual industrial analytics, dive into Spotfire industry pro in more detail. And last but not least, if there are any enhancements that you'd like to see in Spotfire or have any ideas that you'd like to share with us, please don't hesitate to visit our ideas portal and log ideas in. Now we do have a couple of minutes before we wrap up today's session. So I think we've got time for one little question. And for that question, if you bear with me for one second. Perfect. How will AI capabilities evolve within Spotfire Industry Pro over time? Great question. I think I'll defer to Niklas on that one. Yeah. So, yeah, great question. So AI capabilities in industry pro will, I mean, evolve in several directions. One is obviously the, I mean, proliferation of more and more industry native agents, inside agents that actually solve specific use cases and problems in in our targeted industries and use cases. The other direction is also to, let's say, to integrate AI in many different parts of the applications, not only in insight agents. We call this embedded AI, which we in in many cases, it will probably not be so noticeable. It's just going to make Spotfire easier and faster to use and more fun to use, such as much better default values when configuring inputs to an action or to a data function or better default aggregation choices when you're selecting, when you're configuring visualizations, better support for joining and creating table relationships, and so on. So both embedded AI, but also more and more, industry agents, of course. Beautiful. Thank you very much, Niklas. And with that, we are at time. I do realize that, you know, we have a lot of information in today's session. So if there are any questions that come up in your head after we close today's session, please do not hesitate, to reach out to us directly, and we will get back to you very quickly on your questions. Now with that, I'd just like to thank you all for joining today's session, and we hope to see you at one of our future webinars. And one last thank you from me to our speakers for today as well. Thank you again, everyone.