Hello, everyone. I'm so glad to be with you today. In next twenty minutes or so, we're going to explore new innovation around AI relevance, around generative answering, and interoperability with customers' infrastructure. All of this in the context of commerce and knowledge with impacts for specific, specific use cases with customers. So I'd like to start with highlighting the core capabilities of the platform from an end user perspective. The Coveo platform obviously supports lexical search. So it's the ability to do keyword search, either users wanna be, efficient, maybe they are lazy, maybe they wanna discover where the content is. This is critically important to support great lexical search from the platform. In the past eighteen months, we've also invested substantially in semantic search. So the ability to process natural language queries. This is different from lexical search, but still has to be built on the same platform to leverage behavioral data and content. From the same platform also, we are able to leverage, to the the content to provide AI recommendations, either for content but also product recommendations in the context of commerce. And now, using semantic search, behavioral data, content, and security, now we're able to generate answers automatically using a large language model, and the content has already been indexed with the relevance. All of this from the same platform. The all of these methods to interact with information needs to be optimized for both anonymous experiences. So think about B2C commerce. Typically, users will log in when they are about to buy, but before that, all the product discovery and content discovery is done in anonymous mode. So we have ways to optimize relevance and personalization for those experiences. And then obviously, authenticator experiences, you're logged in. So presumably, there's a user profile that is available about the user who's, navigating, who's searching, or asking question on the website. So for authenticated experiences, we also have the ability to leverage user customer information from available customer data platforms, that are available in the infrastructure. So all of this is to provide relevance. In the context of commerce, I'd like to highlight some investments that we've made recently that are quite exciting. So in commerce, obviously, there are two priorities. Right? One is to optimize relevance. So it's really to match the user intent. And the other one is to really configure business outcomes for the merchandisers. So we not only are we providing what the user wants, but we're also serving merchandisers objectives. It can be to increase conversion. It can be to increase profits. It can be to increase revenue per visits. I'd like to show you one example of what we've been doing for more than a year now in the context of anonymous experiences. We call that intent aware ranking. So how do we provide personalization in the context of anonymous experiences basically in commerce? In this example here, a user is on a website, clicks on those three things, visit those three pages on the website here, and searches for men's hoodie. These are the results. Are these results good? Well, they are matching the query obviously, but are these the better results based on what the user has done? So let's look at a different user, again anonymous, clicks on those three items in the session. Same query, Mens Hoodie, this is what he gets. Clearly, it matches the query, but the relevance is totally different. Why is that? Under the hood, what's happening is based on all of behavioral data captured for all of the sessions of all of the users anonymously, product vectors are generated inside the catalog. So at the catalog, we are we have created links automatically using deep learning between products in the context of specific queries. So we know now that if you click on those three items in top right corner, chances are you're looking for men hoodie in for the winter season, right? And automatically these winter ish hoodies will surface on top of the list. This when enabled at customers, at existing customers will increase the revenue per visit by five between five and ten percent. So this provides tremendous value just by turning this capability on. But this is only one of the ingredients of the so called relevance recipe. I'd like to show you what gets into this, this whole relevance recipe, to provide optimal results for end users and also optimize merchandisers goals. So you start with what's already basic. Right? A merchandiser will decide what are the filtering and business rules. What do I need to boost? What do I need to bury for a specific campaign, for a specific context, for a specific customers? So that's the first step. Then keyword and lexical matching will provide precision. In the example to the right here, dark gray hoodie for men, I get three results that are formally matching the keyword search here. Only three results but they are matching, right? They are highly precise. Let's increase a little bit the net and let's capture other results that are even if they are not formally matching the query, we believe that they are similar and they are good. So we by adding semantic understanding, we've now moved from three results to fifty eight results here. So, of course, we have more results that we believe are matching the query, but are they relevant? That's where you need to add automatic relevance tuning. So this will add a popularity score for a given query for this result set. So the fifty eight results have been reranked based on popularity for this specific query. We then add the intent aware product ranking, what I've just described earlier on, the in session personalization to the mix and now we have optimal ranking. Those fifty eight results are now ranked optimally based on the query. Now, this recipe is what we, make available to our customers and we're getting great results from that. Now, the next step is what else could we add, right? There are multiple attributes within the customer catalog and the customer contacts that we could potentially add to this relevance recipe. We have things like, a conversion score on the product, a propensity to be added to cart. So there are trending elements, there are newness, there's the margin that sometimes is available at the product level. So all of these elements potentially can play into the relevance recipe. So the next question is, how do we weight those elements in this new recipe? Right? First so so first we need to understand what are the business goals. So merchandisers will define their business goals. Do they want to optimize for conversion? Do they want to increase revenues? Or maybe they wanna optimize for profits? Depending on the business goals, a new model, we have created a new model that is called business aware product ranking that will take all of these ingredients on the left, what we call ranking features, and will optimize weight. So we have the perfect combination of these elements to achieve the business goals. So by optimizing for conversion, in this case, we are again having an additional level of ranking and reordering based on small weighting of those ranking weights to the left. Now is this mature? Where are we going with this? So first of all, we have tested this in the previous months at select customers in context of product listing pages. Product listing pages are obviously using search, search without a formal query. It's really category pages also known as product listing pages. So in this example here, real life example from a customer using this technique, just turning the switch, we have increased conversion by five point nine percent and revenue per visitor by six point six percent. So quite excited about this early result. We intend to optimize this even more and make this available, for general search by the end of the year. So stay tuned. This will have a big impact on results and for our commerce customers. I'd like then to switch to generative answering. This United is a great customer of Coveo, and I think that we already, highlighted some of the great work we're doing there. I'd like to drill down a little bit more on what's under the hood with United. First, United is using Coveo on their home page, so end users can type queries like this one. Can I fly with my pet? There's a result that is automatically generated from the created content index by Coveo on united dot com. So there's a great answer here. There's also curated content that is surfaced by search engine by Coveo on top of the list. And then at the bottom great results ranked by relevance, they're also super useful. But what's really unique is the ability to protect against these kinds of query. And United serves a very large and diverse audience where some of their consumers will either have either put typos in the query or, just try to play the systems, to game the systems. So in this example here, can I fly with my kids in a check bag? Right. Some of you may be smiling, but these are real queries that are being asked each and every day. You don't want to provide an answer. So this is the Coveo answer. Coveo says, we don't have a generic answer for you, and United displays it this way so there's protection against hallucination. But then at the bottom you have two great search results that are actually linked to the query. So, at the bottom if it's a typo these search results will help you, right. Bag fee calculators, flying with kids and family boarding. So, this is super useful. How do we do this? So let's explore a little bit under the hood how it works and how Coveo is unique in the market to feed those large language models in a way that are relevant and secure. So there are four stages to retrieval between providing content to ground the large language models so it can provide a great answer. The first step is business rules. As we described earlier, this is where you have filtering. This is where you have boost and bury merchandising rules, this is where you have also security trimming. So preparing the content. Then number two, it's all about the relevance. So lexical, semantic, AI based, outcome based relevance. After number two, you have the best documents, the best ranked documents available. So those two documents at the bottom of the United example, this is coming after number two. Number three is really important. We use semantic search to extract the best paragraphs or chunks or passages that are all synonyms and those passages that are excerpt of content are then re ranked based on the query. And then are do they pass the threshold? So they are good enough to generate an answer from a large language model perspective, this is where in the case of United it it stops. So we are not at number three, we're not going to provide passages to the prompt to the large language model, hence why there's no answer. And then number four, it's prompt creation from those passages and additional context and security that is sent to large language models so we generate an answer. We don't trust the large language models to deal with facts. Large language models are used to master the language. The facts are provided by Coveo in one, two, three step. Okay? So this is critically important. That's why United uses Coveo. Now customers are large customers. A lot of them have those internal projects around building their own large language model based on open source or provider like OpenAI. And they, they they are building that to serve specific use cases. They may fine tune it using their own vocabulary. They may want to use and own this for strategic reason and that's okay. What we have launched this quarter is what we call the passage retrieval API. So, the step, the three first steps of our retrieval of our retrieval, infrastructure are now available for these customers through an API, so they can use the best passages, the most relevant, secure, and contextual passages, and then use that to ground a prompt to their internal GPT or their internal LLM. So this is highly strategic. It leverages the interoperability with the internal infrastructure of those customers and then all of the content that is inside Coveo with the relevance can be used into multiple use cases that are powered by these LLMs. In addition, because we provide our own large language model in our infrastructure, these customers can AB test their own large language model versus the baseline provided with Coveo and then decide. Maybe it's better to use a passage retrieval API in their context, it will still remain with Coveo. So all of this is now available in pilot and, we expect to make it GA by the end of calendar year. So just before last Dreamforce, we have announced this new partnership with Salesforce around data cloud, around Salesforce data cloud. And what's exciting about that is that when Coveo is available inside a customer that is using Salesforce, we're going to make Salesforce better because we're going to make available all of the content and the behavioral data process by Coveo available inside data cloud. So it can be then used by all of the AI models from Salesforce. So let's see how it works. First of all, we're proud to have the support from Salesforce Service Cloud. So this is Kishan Chetan, the EVPN GM of Service Cloud Salesforce that is supporting Coveo. Basically saying for use cases that require advanced relevance or complex connectivity, Coveo is a great option for Salesforce customers in the context of Service Cloud. So if we look at the market as Salesforce market where we play, again, we look at it from two axis. So you have the content axis at the bottom, that's the x axis. On the y axis is really audience size and diversity. So on the y axis, if you have millions of users across multiple experiences, Coveo is a great option. If you have multiple content pieces with large and complex documents such as hundred pages PDFs or highly technical content, then you need Coveo advanced capabilities to be able to process that. So that's the x axis. And obviously, if you have both, so typically large and global enterprises will require, will have those requirements, then Coveo is a perfect option. Now, for those customers in Salesforce ecosystem, they may power these advanced use cases with Coveo. But then, what we're doing is we're synchronizing the behavioral data and the index content with Salesforce data cloud. So customers have this option. And by doing that, it's now available for Einstein models that are, that are outside of Coveo specific use cases to leverage this data. So one example here, this is, the representation of Coveo behavioral data that has been synchronized inside Salesforce data cloud. So it is possible for a Salesforce admin to explore this data and this data explorer inside Salesforce. This is behavioral data coming from Coveo. And then from this behavioral data and content, it becomes possible to leverage Coveo into an agent force interface like this one. Right? So you're asking to Einstein, what do you know about this product, Blue Lagoon Flip Flops? So Einstein will offer an answer that is coming from the content previously indexed with Coveo, and the behavioral data will also inform relevance on that content. So we believe that's another example of interoperability with Salesforce infrastructure and customers that will create huge value down the road. It will it will now become possible for Salesforce customers to use Coveo to feed their Einstein models for multiple use cases and make Agent Force better because it now has access to all the content from the enterprise already indexed with Coveo. So with that, thank you very much everybody.