Alright. Good morning and good afternoon, everyone. Welcome to this month's learning series webinar. My name is Claudine Ting, and I work on the global marketing team here at Coveo. For those who are attending a learning series webinar for the first time, this monthly webinar program is where we get hands on with how to enable Coveo's features that can help you create more relevant experiences. This is the second session in our three part series on machine learning. Today, we'll focus on smart snippets and the case assist machine learning models. Before we get started, I just have a few housekeeping items to cover quickly. For this webinar, we encourage you to type your questions on the q and a box in the chat box as we go along. At the halfway point and at the end of the presentation, we'll try to provide a short recap of the questions being asked, but, guaranteed, we welcome all questions anytime during the webinar. Lastly, today's session is being recorded, and you'll receive the presentation in the next few days. Today, we're welcoming back Jason Melnick, director of customer onboarding and Lipika Prama, customer success architect. And I guess, we're ready to get started. Jason, do you wanna kick it off? Sure thing. Thanks a lot, Claudine. I'm really looking forward to this session, today as is Lipica because we're gonna be talking about two of Coveo's, newest, machine learning models, smart snippets and case assist case classification. So I'm gonna be covering the part about smart snippets, and then I'll turn things over to Lipika who will talk with you about the case assist model, after that. And as always, please feel free to ask questions in the chat along the way. We'd like to keep this, conversational. So let's get started. So I'll start things off with smart snippets. And before we talk about smart snippets, I wanted to just quickly examine, like, the typical search experience. You know? So end users might search for something by asking a question. We do it all the time in our personal lives. And when we do, we may get greeted with a set of results that potentially has the answer to the question. But where is the answer? How do we know that one of these results provides the answer, and which one of these results do we choose to actually get the answer to our question? So these are the problems that smart snippets model is seeking to solve. Which document actually contains the answer to your question? Where within that document is the answer written out? And, you know, if I'm the end user, maybe I didn't ask the question in a way that is really cohesive with the way the content is written. And so, with having a machine learning model will help accommodate for this, and make our questions more intelligible given the context of the content that we're looking at. And, you know, in short, if you were to ask an end user what they really want out of a search experience, don't be surprised that they say, just show me the answers. That's what we're after after all. And so this is where Covideo smart snippets machine learning model comes in. Now this is a machine learning model that leverages deep learning infrastructure to extract the answers from index content, and that, like, refines this and poses some follow-up questions. So you probably are familiar with this concept of people also ask. This is another component of the smart snippets model, that runs in tandem with providing the actual answer to your question. And so in the next few minutes throughout this session, we're gonna talk a little bit more about the model and how it works, and then we're actually going to configure it and deploy it into a live interface, so you can see just how easy it is to get started with smart snippets right away. So let's first talk about how the model works. Well, it starts by, like, identifying headings within index content, and then it embeds those headings, in a vector space along with the user's queries. Now, we calculate the similarity score to display, a snippet or if we don't, you know, find that similarity there, then it won't display a snippet for it. But it's basically matching up, the user's queries with headings within your content itself. This process actually comes after the search API is applied, and it refines the top ten results. So the smart snippets will be applied to what would normally be the top ten results of your query anyway. And it does leverage deep learning. So, again, in a nutshell, this model seeks to match your query with headings within a document. So that's one of those, prerequisites about this model in the first place. It's really more about the content than anything else. So let's talk about these. First and foremost, it doesn't, unlike our other machine learning models we've talked about so far, automatic relevance tuning and query suggestions, this model does not take usage analytics events to to train it. It learns off of the content itself. So to that end, the formatting is pretty important of what kind of content you wanna use. So, we'd like to look at semi structured or or structured HTML content. So we've got your h one, h two, and h three tags appropriately tagged out with the content beneath them. This is really the building blocks for the smart snippet model. It also does require the Coveo enterprise license. So you'll need to be an enterprise license customer in the enterprise org, to access this feature. The enablement process of this is actually really simple, and we're already just a a few minutes into, today's presentation, and I think I'm gonna be at the demo already. So we'll get into this. But, basically, what you're gonna do is, you know, define the document source within your catalog or within your source, your index that you wanna use, to create smart snippets with as well as any particular, document types and fields in that content. And then in the admin console, you're going to build the smart snippets model. Next, we're gonna associate that smart snippets model to a query pipeline. So so far, all of this is very similar to, like, the automatic relevance tuning, query suggestions models. You're just creating that model and associating it to a query pipeline. And then from the front end perspective, it's actually one line of Coveo, and what we're gonna do today is two lines of code. So one for the actual smart snippet and one for the people also ask component. We just pop those two lines of code into your search interface. And as long as you're on the right Coveo JavaScript version, which is the latest version, or or, the latest version of Coveo atomic headless framework, then, you know, voila, your smart snippets will will be working. So we're gonna delve right into that with the quick implementation demo. So let me just change my screens. Alright. And, Claudine, keep me honest. Can you guys see my, Coveo cloud administration platform now? Yep. I can definitely see it. Okay. Great. So, here we are in the Coveo cloud platform. And, I'm gonna jump right to, the machine learning section over here on the on your left side side navigation. And then you'll see in here is a section called models. Again, if you have a enterprise organization, you're gonna be able to find the smart snippets model. All we're gonna do here is now click this button that says add model. And we'll give it a name. Online smart snippets two. And then you'll see the different model options you have to choose from. We're just going to select the, smart snippets model here and hit next. And now as I mentioned, you're going to wanna define the source of content that you wanna use. You can use multiple sources, but when you select on this, you're gonna see the different sources that you have available. So I'm gonna choose my Coveo document source because I know that this is, an HTML source. It's content that's structured the way that I intended to be structured. I could use the Coveo website. Maybe we can use the COBEO answers site map source as well, so on and so forth. So you'll select your sources here. We can move on to next. Now these fields are actually optional. So if you don't put anything in here at all, there's gonna be a default document type and, field association that you're gonna use. So the default is document type is HTML, and the field is what we call the body. So this part is really important. You want to make sure that you're indexing and mapping a body, to your source because that's where all the content actually lives. So if you think about your content metadata, you'll have a title, which is gonna be the title of the document, the document type, in this case, HTML. You may have a a ton of other metadata that's in there based on how you have configured your source, but the body is actually the HTML of that content itself that includes the h one, h two, h three tags, and all the associated text within them. This is the important part that you'll need before, answering smart snippets. Now if you have a source that has different types of content in there, and maybe you're indexing something like site core, for example, and you have some HTML content, document types and then some other, like, file types and things like this, then you might wanna specify the the type of document you want this to look at, and any specific fields that you wanna use. But for the default configuration, I'm just gonna proceed forward. And then the last step is it'll ask you if you wanna add any include exclusions. So there are some CSS selectors that you might wanna exclude from your source. Things like, you know, if we haven't fully mapped everything out, you might wanna exclude, like, your, the headings and footers and things like that. Sometimes there's gonna be other selectors in there that are used for specific functionality that don't really make sense to include in a smart snippet, and you can exclude them by putting the CSS selector names in here. As you go to the next part, you'll get into the final this is the really the last screen of it. It's gonna show you a summary, the model name, the model type, what sources that are in there. And it should be noted that it takes about six to twelve hours to build this source. It's gonna be actually looking at or build a model as it's gonna be looking at your source content. So I'm gonna hit add model, and that's gonna start going. Now like a good cooking show, of course, I've already done this all ahead of time. I have a model in here already. It's called smart snippets test. And just so you can see, I'll I'll run-in here and look at the configuration. You can see I've built it essentially the same way. I didn't really specify any document type or fields. I didn't specify any exclusions. And I used a couple of different sources in here that built the model. It did take about a day for this model to to build. That's why I did it ahead of time. Otherwise, this will be a very long webinar. So there it is. It's that step one out of the gates. So we'll move on to step two. So step two, I want to associate this model to a query pipeline. And that query pipeline will, of course, be associated to the search interface where I want to enable my smart snippets. In this case, I do have a a search page. It's called Jason's test, and I have a special query pipeline for that called Jason's page. I'm gonna click into this, and I'm gonna go right to the machine learning section here, and then I'm gonna click on associate model. Now this is going to give me a listing of all the models that I've created already, and I'll find my smart snippets test model right here. I can click on this and associate the model. And now my model is associated to my query pipeline. The model has been built and configured correctly. And now I'm here on my test page, and, I still don't have smart snippets. So there is still something one more step that I have to do, the the final step to enabling smart snippets in here. And this is gonna take a little bit of very light front end development work. I'm just gonna need to add a couple of lines of code in here. And those codes let me go grab those real quick. What I'm gonna do is go into edit mode on my search interface. This helps when you have an interface editor. This is not a drag and drop type of a feature, though. So for this, when you're in the interface editor, you're you're gonna have to go into code view to look at this. What I'm actually looking for is a particular section right down here. This could be different on your site, but I'm looking for the line, this class for Coveo hidden query and Coveo error report, which comes right before your Caveo result list. This is where I'm gonna embed the codes. So I'm gonna pop this down. I'm going to add my class. And if you're using the Coveo interface editor, you must be in the most recent version of KOBEEO for, KOBEEO JavaScript framework. As you start to type in, you can start typing in smart snippets, and you're gonna see smart snippet is one of the, the helper will bring it right up like this. We're gonna close out this class. And this code right here for smart snippet, this is actually what's going to render the snippet results as your your top result. Now you saw that there is another version of this as well, which is called the, the smart snippet suggestions. So I'm gonna add this as well. So you can see Kaleo smart snippet suggestions. This is that people also ask component that appears below your smart snippets. So the same thing. I'm just gonna close it out. So now I've got my code on the page. It's in the right place right before the Coveo result list right after the error report. I can switch over to UA or UI view just to make sure my UI still looks okay, and I hit save. But if I do, like, a search I did before for machine learning, now I have my smart snippets. So this is kinda how it renders. You can ask a question, like, I'll pop a question in here, such as what is a query pipeline? Common question we get. There's an answer for it. I get my smart snippet in here. The all of that, it becomes with this show more and show less, feature down here so you can open it up and actually review your results without having to click into the content itself. And then that smart snippet suggestion component that I added is down here below. Below. This is where you can see people also ask. You can get more information, similar queries, and so forth. And we actually included this, was it useful? Yes or no, feedback mechanism as part of this as well. This is kind of, like, for, just tracking purposes. Like, you can actually track this usage analytics to understand how people are using it. At this time, it doesn't actually change, like, the model learning in any way, so it's way more of a feedback mechanism for you to learn from and so forth. But we have considered for future development of this model to be able to, leverage this as part of, the model training elements. So, that may be a a future iteration. But that's pretty much about it. So I went through a lot of it really fast. I saw that there are a couple of questions in here. Again, just to recap, we built the model. We added the model to the query pipeline. I added a couple of lines of code into my search interface, and now I've got smart snippets enabled. It's a pretty quick process to do with a pretty powerful result that gives you a really nice, modern field to your search interface. So I guess we'll stop there and see what kind of questions that the OEMs come up with. Alright. Thanks for that, Jason. We do have some questions that came up in the chat and the q and a box. So, Chris has a question. Is it possible to tweak the model if you find that an answer it chose is not correct? That's a great question, Chris. So the model itself, this is the strange thing about it compared to other models. It's not really based on end user input or at least analytics at all. It's looking at the content. So the way that you would tweak this is through boosting rules. So if you find a rule that doesn't like, a a result that doesn't really quite answer the question that was asked, then we can use a ranking expression to negatively boost that down. If you get it down below the tenth result, then smart snippets won't act on it because it only is gonna, like, look at the top ten results that are returned, and it will create a snippet out of one of those results. So, again, if that result that's not good is below the top ten, then it won't be considered as a snippet, response. Okay. Here's a thank you for that, Jason, as well. We got we got one here from Sham. If we have a featured result for a keyword, does smart snippets consider or ignores to extract the content from the featured URL? Yeah. It's, kind of another way like, really, the the last answer I gave is it's it's kind of pertinent here too. So if you're boosting any content, if you are using a featured result to put a content at the top, if it's within the top ten results that are displayed without smart snippets, that's those are the candidates that the smart snippet model is going to look at to be able to pull the snippets from. So that also means automatic relevance tuning. So if you have a art model in there that is dynamically boosting content to the top of your list, these that content within those top ten results, that's those are gonna be the candidates that smart snippets will be looking at, to pull a snippet out of. Awesome. And then, here's another one. Does does the snippet suggestion use the same model, or do we need to create a new model? That's a great question, Sham. Again, it actually uses the same model. So, it's, all part and parcel the same thing. And, again, it'll only look at that, the the top ten results that would have come back from your query anyway. Okay. Thank you for that, Jason. I just wanted to share, another, question that was raised earlier and was answered, via textpadlip because so, just Sham hopes it's a smart snippet reacts. Smart snippet respects the permission of document and shows, what the user is entitled to. And Lipica said, Hi, Sham. Yes. Of course. Coveo follows late binding security, which means we don't render any results that the customer should not be able to view. We check the permissions before showing any results. Okay. Alright. And, Sham, you're very welcome. Keep the questions coming, everybody. There's one more question, before we move on to the next part, I think. I I see that. It's a it's a good one. So can smart snippets get used in, combination with, case deflection functionality? And, actually so as long as, again, it's within the top ten results that would be displayed back, it will work, and it will work in any interface as well. So I demonstrated it today in a full search kinda panel, but it is applicable. You can add this into IPX. You can add it into an insight panel or into a case deflection panel as well, and it renders, pretty much the same way. So, yes, it would work in this way. Cool. Thank you, Jason. So we'll move on to the to the next part of today's webinar, which is about case assist and the case classification models. I'm gonna quickly share my screen. Just let me know if everyone can see it. Jason, can you see my screen? Yes. Okay. Awesome. Awesome. So today we're gonna talk about, you know, another deep learning model very similar to what, Jason spoke about the smart snippets. We call it the case classification model. But before going into the case classification model, why is it needed, when would you need it, what use case would you apply it to, I wanna form a little bit of foundation as to where it is applicable, what use case so you know exactly, you know, how to use it. So before going into the case classification model, I wanna talk about a feature called case assist, which is which is where we use the case classification model. So, and if you've been a customer of of Coveo for a while or even if you're a new customer, you would know that Kaver has the case deflection component, has a case deflection component, which what it does basically is, you know, any case creation page, where you see the, you know, the the the form. Usually, there'll be a form where you enter the product name, the product family, all of that, And then you see the recommended solutions on the on the right hand side. So these the case deflection component takes advantage of, two other machine learning models, which are the ITD and the ART models. And they basically look at user behavior. They look at large amount of the ITD model looks at large amounts of texts and figures out, you know, certain certain keywords that are important even if even if the text is too long. And that's how we are able to deflect cases. We're able to provide suggestions to customers to for for them to figure out, okay, I can solve this on my own. I don't need to, you know, log this case. But one of the things that, that our customers started telling us and we also got feedback from, you know, customers who log cases on on our support side as well is that, you know, a lot of times the case submission page can be can be long and it can get really cumbersome to fill out all the different fields. And sometimes you don't exactly know what fields you should be selecting. For example, sometimes, you know, you would have to enter the product family, the product version, and you're not always sure what version of the product you're using. You're not always sure, you know, if it's a complex product, you're not always sure what subcategory it falls into. So it can get a little complicated before you can submit a case so the the feedback we got was that you know just let me submit my case already So, which is why we we designed this feature called case assist. And what case assist helps you do is, you know, remove that complexity of customers having to do that manually. So the problems we're trying to solve here is, obviously, the first one is cases being created because we want to deflect cases. We want customers to self serve. Really, customers themselves want to self serve. They don't want to create a case, and cases getting routed incorrectly. So if you would remember the first screen that I was showing where, you know, you would see the product family, the product version, and all of that within your organization. Most service organizations, they have a structure based on, you know you know, subject matter experts. So there are different teams that are that do cases for different products, for different product versions, or for different types of problems, you might have different teams. So because a customer might have selected the wrong product on the case, these cases get routed, you know, to the wrong teams. And what ends up happening is it's a frustration for the customer as well because there's a lot of back and forth where you're trying to really understand, you know, what the customer's problem is actually to route them to the right team, to route them to the right specialist to solve their problem. So, the Coveo case assist, and this is how the UI looks like. And I'll do a very short demo, and I promise I will get to the case classification model, but I think this is important to lay the foundation of what we're gonna learn. The Coveo case assist feature is obviously, it is designed to get high quality incoming cases, and it's streamlining the customer's experience. So if you look at the screen here, and I'll just point here, you see that the case subject and the case description are already entered. And you'll see that, typically, the customer would have to select the product. They would have to select the category, and they would have to select the topic. But in this situation, we are analyzing through our NLP, models the the description of, you know, the description and the subject, and we are able to predict that this case is related to this product, to this category, and to this topic. And like promised, I'm gonna do a very short demo here just to show you how it really works. I'm just gonna reload the screen to make sure that I am logged in. Okay. So what we are doing here is we we are looking at a case assist page, and I'm gonna, you know, enter a case subject. And I have a case description already here, so I don't waste time typing. And I enter the case description. And if you look at the screen right now, the select cat related categories, basically, a customer would have to select the case reason and the product that they are logging the case for. And if you look at the screen right now, we are already predicting what the case reason might be. You know, the reason might be, you know, the tracker isn't set up properly or the device isn't paired properly, and the product we're also get you know, based on what's been written there, we know that, you know, this is the problem with the speed with Blaze. So this is a hypothetical company that sells fitness products, you know, fitness watches, and whatnot. So we're we're guessing from what the customer has, has entered in the description that this is the case reason and this is a product. So this is how case assist really works. So going back to, to our, to our slide here. So case assist really is and I'm gonna do this really, really quickly. Case assist is really an API based solution, and it can be used in any any any kind of UI. It can typically, it is used along with the case, you know, the case creation UI. But what it does is it has two components. One you saw already, which is the case classification where we are trying to classify the cases into the different product, issue type. And what I showed was what is in my organization, but the case classification model will work according to what is, you know, what is required for your organization. So the case classification and the document recommendations. For today's webinar, we're gonna focus on the case classification, which is where we need the ML model to be built. So if you look at case classification, there are two types of case classification that happen in the case assist space. So two types of ways that we recognize the different categories or the different pick lists, that the customers have to select. The first one is case similarity. So and this is by default. Case similarity, we look at different cases and it it is very similar to, you know, you already have an index where you already have indexed your cases. And we will look at those look at those cases and we will say that, okay. Typically with this subject and this description, this is the category that the customer selects. And that's that's what that is all about, the case similarity. Then comes context recognition, which is the machinery model, which is the NLP technique that looks at, you know, the common vocabulary. It looks looks at contextual nuances. It also looks at, you know, how the user is changing those categories if it has suggested something wrong, or something incorrect that is not related to what the customer is looking for. Typically, this ML, you know, ML model, this new case classification ML model works really well for organizations that have, support cases that have more than eight words. Most organizations have support cases that have more than eight words because customers sometimes, you know, they can attach log files, they can give you, you know, step by step procedure on how to figure out what their problem is. It can be anything. Right? So, typically, the context recognition is which is used for cases that are really long, and most organizations have longer cases. So now let's look at I'm gonna take you back to the to the UI here. And this is a Coveo platform, the cloud admin platform. So any customer that has an enterprise, service license, the case assist functionality is available for you. So the case assist functionality is right here, and I'm going through the case assist, how to set it up, and then I will show you how to go about creating the ML model. Model. So I have one here already, so I'm just gonna click on that because we're not learning about case assist today. So if you're setting up case assist, it is all API based. So once you create that case assist, model here, what it does is it generates a key, and you can use that key to use it in in because it is API based, you can use it in in your UI where you have your case creation component. There are two aspects to it like we discussed before. One is the document suggestion where we are basically telling these are the articles that might help solve your problem. And the second one is case classification, which is the Ml model that we're gonna learn about today. So the case classification model, if I say configure case classification, you can see that by default, it goes to, look at case similarity, which is which which is static, and it will look at all the cases that you have indexed, and it will figure out the case classification based on that. But you can change that. You can say, I don't want case similarity. I would rather go with context recognition. So the context recognition needs you to build a model. And why does it need you to build a model is because it's gonna build a model based on what is what will work for your organization. And why it is needed is for every organization, you might want to look at different fields for the machine learning model to understand the context of your of your business. Right? You might want it to look at the case subject, the case description, but sometimes you might want it to look at, you know, different fields that you have. And every organization has different field names. So what we're gonna do in this model is we're gonna go through step by step and make sure that all of those field names are proper and that we are learning from the right information. So I'm going to go back here. So this is a case assist setup. To build a case classification model, we go into machine learning and click on models. And I have a couple of models here already, but just like, just like Jason's model, this model also requires around six to twelve hours to build because it looks at all the cases that are getting indexed. So it would take a really long time for me to show that here, which is why I went the opposite route. I first showed the demo, and then I'm gonna show how to build the model. So if you click on add a model, and this is the regular interface where you build all the other models. So if you scroll down here, you will see case classification right here. You can click and select it. You can add a name to it. Click next. The first and the most important thing is to select what the case classification model is going to learn from. So the case classification model, like I've been saying, it learns from cases. So you the first prerequisite for this to work is you have to index cases. So if you're not indexing cases, there's nothing for it to learn from. So you have to index cases. And there are a couple different prerequisites that I will also talk about is that the source that you're selecting or the case source that you have should have at least five hundred cases. And that is the optimal you know optimal level where the model becomes good and it has good amount of information to learn from Obviously, it can have much much more than that. But any less than that, it it doesn't it doesn't start learning. It doesn't start creating that model. So five hundred cases minimum. And you can also say how much, you know, how much further back you want to go to as well. So just like any other model, so I will select the source here. So I'm gonna select, you know, one source here. And what you can say is you can select the date range. So you can say the model your model will learn from support cases within this time period. So you can say last six months, last three months, last year, and you can even have a custom date. So we can say learn from last year and a half or something like that, or loan from the last three years per se. So you can, you can select that. The reason you have this option here is this. The further that you go back, you have more cases. Right? Because, say, last six months, there were around five thousand cases. If you went back a year, you would have even more cases. The recommended is six months, but you could go back even more than that. Then comes the advanced filtering options. Why is this needed? This is not needed for most customers, but some customers, you know, if they have, you know, additional advanced settings where they're using multilingual content. And and this is just an example. And say for example, they only want it to learn from the the cases that are English language and then I can select this advanced filtering settings to make sure that it is learning from all the English language cases. The other setting that you would put here is, make sure that your your ML model is learning from closed cases because, it can learn from open cases. It can learn from all cases. But closed cases, the accuracy with which the classification or the product category is selected is higher. So you would want it to learn from closed cases. So you can put that advanced filtering option here. So it's very simple. You just put your filter name over here. So you can say SF language, because I was talking about language and you can say is equal to, and you can put in the value here. All of these values, please make sure, that they they are appropriate for your organization. You can check them by going into the content browser and looking at all the different field values that you have there and make sure that you put the accurate ones here. I'm gonna click next here. So a couple of things couple of more things that this model needs, to be able to predict accurately. The first thing it needs is the case ID for the training. And why does it need the case ID? Because the case Id is typically unique to each case. Right? So each or it could be the case number because case numbers are also unique. You cannot have more than one case that have the same case number. So what you need to do is you should just put the name of the field here so you can say s f case number. Then the next and the most important thing, and we've discussed this, today is that it looks at fields. Right? It looks at fields that it has to learn from. So it can learn from the subject or the description. Typically, those are the fields that we look at to figure out the classification. So for example, here, I will put the SF subject, s f description, and then click next. You can select more than that as well. You could only go with description. It's totally up to you. And last but not the least is you select which fields will we provide the predictions for. So now if we go back into our UI here. So what we did was we said, look at the subject, look at the description, and figure out what the classification is. But in this situation, if I was creating a model for this UI, I'm gonna predict the case reason, and I'm gonna predict the product. So the things that I would enter here is s f case reason. And again, please make sure to check the exact names of the fields I am entering here just to show you an example, an s f product. So once these are entered, you click next, and it'll show you a place where, you know, it has, you know, accepted all of the settings that you have here. And once you say, add a model, it'll start building the model, and it can take up to twelve hours for it to build a model. And once you have built the model, you it you know, it's live, and then you can see it working on your on your case assist page. So I'm gonna, go back to the deck here, and I think I already discussed this, is the prerequisites. I think I already spoke about one. The other one is there has to be sufficient amount of classified support cases. I already spoke about this, at least five hundred cases and sufficient amount of data in the training field. So if you remember, we selected the subject and the description field. So the training must contain at least ten characters to be considered by the model. This is a fairly loose, you know, prerequisite because if you're selecting a case subject or a case description, they will have more than ten characters. So, the reason it's there is so you don't select a field that doesn't have more than that. So make sure that it has if you're selecting case subject and description, ninety nine point nine nine percent of the time, you wouldn't have to worry about this prerequisite. So that is how you build, the case assist model. And once the model is built, what you can do is you can look at the configuration of the model. The only thing I want to point out here is once your model is built, you're not able to change any of the configurations here. You're not able to change the fields that it is looking at. You're not able to change the fields that it is predicting. The only thing you can change is the name of the model. So say, for example, you you are not very sure of what you selected as a configuration for your model, you can always delete the model or discard the model and then select the one and and create a brand new one. And it might take a while, but it should still be okay because you will be learning from the same data. So you should be able to get up to, you know, where you were before very, very quickly. So I'm gonna stop really quickly over there because that was all I had for the case creation model, and I see a question here. Alright. Thank you for that, Lipika. We do have a question from Jennifer. If one doesn't have support cases, would user feedback be a valid source for the model to learn from? Yeah. I would think so. It so it basically has to have a structure has to have a structure. I only said cases because cases typically have a subject and a description, but it would be the exact same way. There is a source. You tell what it has to learn from. You tell the field that it has to learn from and and what field it has to predict into, but it should definitely work. Okay. Thank you for that. Another question is, hold on one second. What is the difference between case assist and the Coveo case deflection component? Mhmm. So, the difference between case assist and Coveo case deflection is case assist is is using what we call as the NLP or the deep learning models, and it is predicting the case reason and the case product. Case assist has two different functionality. One is the classification that we see here. The next one is the document prediction, which says, hey. Look at these recommendations. These might solve your problem. Case assist has two components. The case deflection model typically has, you know, your form and then your recommendations, and that's it. It doesn't do anything about classification of your cases. And it also uses a different machine learning model, the and this one uses a different machine learning model. Okay. Thank you for that, Letika. Any other questions before we wrap up? I think that's all we have for today. Thank you so much everybody for joining us. I hope we hope you learned a lot, and you're gonna be seeing us, the three of us again, next month in September for the last part of this series. Thank you again, and have a great rest of your day. Bye for now. Thank you, everyone. Bye.