Transcript for "Revolutionizing enterprise search with AI: unlocking the potential of your data": Thank you so much and warm welcome to everybody here. I am Philippe Martensil, the President of Teneo in France, and I will be leading a discussion on revolutionizing enterprise search with AI, tall order. I'll be moderating this discussion over the next twenty minutes, and we'll explore how AI is transforming the way companies unlock the most valuable of the asset, the data. With the rise of AI agents and data driven decision making, enterprise search is no longer just about retrieving documents. It's about enabling intelligent action to fully leverage the AI revolution. Enterprise data is often fragmented, hidden across systems. Adobe and Cohere are partnering to solve this with powerful AI, unlocking value, driving productivity and laying the foundation of next generation of AI agents. Joining us today are two exceptional leaders, actually, yes, driving innovation in this space. Abhijit Singh is the Senior Engineering Manager at Adobe's Document Cloud, where he leads Gen AI and AI services initiatives powering Acrobat and services APIs. His work focuses on enhancing document intelligence for millions of users. Nick Morales, has global customer experience and partnership at Cohere, a leading AI company. He brings a deep background from Alteryx and IBM and now helps businesses integrate advanced language models to drive real world impact. So Nick, let me start with you. Can you give us a quick sort of lay of the land as, what you see as the, obviously, the opportunities, that Enterprise AI offers businesses, but also some of the challenges that may hinder the unlocking of those opportunities? Yeah. Thank you. And, it's no surprise. Right? It's not a spoiler alert to say what a massive opportunity there is with, enterprise search. You you think about the amount of data that enterprises are sitting on, not just years, but potentially decades' worth of data, of unstructured data that, is in the form of emails, documents, PDF, presentations in different document stores and images that, AI is now providing the capabilities to be able to unlock, unlock to be able to make the workforce smarter, to be able to inform decisions. But it's not an easy thing to do. You think about the different challenges that come with working with a variety of types of data and data sources, it becomes a much more complex challenge. Now this conversation, let's first ground it just like a good rack system. Right? This is a conversation around enterprise search, which is fundamentally different than consumer search. Now yesterday, I was googling, the score of the Real Madrid, Paris Saint Germain Club World Cup game, which because of my jet lag, I'm from San Francisco, I I thought it was today. But Google brought me back a response, and they said, okay. Here's the score in 2022. Well, I shrugged. I said, okay. Well, it's it's a it's a few years old game. It's it's low stakes consumer search that you would do. Right? Just like you would look for restaurant reviews or news articles. But when you think about the enterprise and with a new technology like AI that's still building trust, working with proprietary data, working and needing to use context to bring high value insights to the enterprise, you're not allowed to make those type of mistakes. So with the amount of potential for value that enterprise search systems provide, there are some common challenges. And what we're seeing is the, the amount of different, data silos that exist between large enterprise organizations, whether it's sitting in a file store like an s three or in Salesforce or SharePoint or Google Drive. You have to be able to bring that data together. You also see that traditional search is focused on keyword or or lexical search. So it's looking for exact, matches, between words or even metadata and what you're actually searching for. And in enterprise, you need more contextual based, retrieval of information. And the third is security and and compliance. So you can have the best search system in the world, but if it doesn't have access to your key information because your IT team doesn't trust how it was deployed, then you are able to actually have the full picture or three sixty degree view of of the enterprises. So that's what Cohere focuses. We focus on, deploying models and solutions that are secure and allow you to really unlock that data that you're sitting on. Thank you very much for that overview, Nick. There's a great setup actually for my next question, turning to you, Abhijit. Could you provide an overview of the partnership that Adobe and Cohere, have just, gotten to? How did it come about? And more specifically, what is it that you're bringing to market? What is different with this partnership? So, Adobe brings Adobe Acrobat AI assistant to the market. It's an amazing product that helps users tighten the loop between comprehension of the document to action, whether it's for consumption or for creation workflows. So, and Coher is a key partner in one of our trust pipeline, which attributes the answers back to the sources. The embedding models are an amazing technology that we have been able to leverage to really establish trust in our generative AI assistant. That's something that Adobe is very focused on being responsible with AI, and that's where we leverage their technology to make sure that the answers that you get, you are able to refer back to what components in your source document were used to generate those answers. So that's where, Cohesity's technology has been very, really helpful. And, again, we came across their technology after a lot of evaluation in terms of what's available in the market, And their model was really able to hit the the the performance, the cost, and the quality factors, and we, that's why we selected their model. And and the company has been really innovative, as soon as a new use cases, new use case emerge, they have a new amazing model out there that we can leverage to unlock those use cases. So yeah. Thank you very much. Nick, you in your earlier answer, you mentioned some of the current challenges, with enterprise search. I like the 2025, 2022 you mentioned, and yes, no company can actually live with that. How does AI, and specifically, Coheria Secure, AI models, address these challenges and improve the search functionality? Once you've said what you mentioned about Paris Saint Germain Real Madrid, okay, how do you make this go away? Yeah. Yeah. So, let's start with first the models and then talk about what how you we uniquely bring them to market. So the goal of our search and retrieval models is to ensure that there is context and intent behind the results of your prompt or your query. So we wanna make sure that we understand the meaning of your question regardless of the language it was it was posed in. And that starts with the embedding model to be able to, look at your query and understand, what was the meaning. So if I ask it for what is the annual leave policy, well, it could be stored under, holiday schedule, or it can be stored under, what is PTO? It's a it's, not matching the exact words, but it's the the same meaning. So you will ensure that the embedding model is able to bring that meaning back to, on on the response. And then the re rank model is then able to sort these results into a list, a generated list, where you have more accurate, more contextual relevant results at your fingertips. Who wouldn't want that first result you get back to actually be the right one? So we bring these two models. We deploy them in the cloud through our partners such as AWS, who is working with us in Adobe, but also on on prem. And the unique, ways we bring this to go to market is we bring these models to where the customer is. As mentioned, with Adobe, we partner with AWS, but we also deploy privately full capability in any cloud or data center in any region. We also connect to the different data sources where you're going to want to provide context to that result. So whether it is requires a connector into Salesforce or document storage, You need to be able to, look across your different data sources to be able to have that complete answer. And then when you get to the data sources, you need to be able to retrieve from the different data types within slides, within PDFs, within documents and spreadsheets. Everyone's a favorite. Once you find the spreadsheet in the deck, then you need capabilities around multimodality. So not just text, but also the images and the context behind the images, which sometimes aren't accompanied by captions. Other times, you see images or diagrams that have text overlay, and you want also context within that. So our models bring that multi multi modality capability. Now speaking of multi, I have a multilingual. Right? A search, query may be in English. Well, I so I manage the support team also, and we get support tickets from all over the world. Well, I shouldn't only retrieve quest support tickets or inquiries based on English or in the language that I asked. So these embedding and re rank models are able to, provide context over in over a 100 languages. So you take all of that, and then now you have a RAG system where you're able to provide citations back to if I'm asking the system a question, where did that reference or source come from? The source and the actual full passage. Because, again, we're AI as an entire space is still building trust within enterprises. You need to you're gonna love the answer, but you're gonna wanna know where where it came from. And, you know, we think about search not as a a single search bar that's smarter, but a full AI system that is helping you build trust, and accurate responses. Nick, I can tell you, I'm really happy, and I I'm trusting you that I'm gonna find my twenty twenty five PTO and not gonna get twenty twenty two results. I'd be, devastated. Abhijit, can you share some examples of how Adobe and Coherus technology in their application, either internally or for customers, how are you leveraging that? Sure. I can definitely tell you, Adobe Acrobat AI assistant will give you the answer for your holidays. We have tried it out. So, I think Nick really touched upon multilinguality as a a use case. Using their embedding model, we were rarely able to unlock multilingual use cases. We now sup support French, Spanish, Italian, Japanese, Brazilian, Portuguese. These are the languages that we were able to unlock with with the help of the technology and a lot of, and a very complex pipeline behind the scene. And the flexibility that really gave us in terms of reaching out to our customer has shown tremendous results for us, and we are seeing that in how the users are using our product. And and that's one of the great use cases that we have been, able to support. And and you also touched upon multimodality. So PDF documents, you all know, are it can be a simple, text document. And a number of times, you have lot of charts, images, figures within your document. We also support other formats like images, Office open Office documents. Those kind of documents have all these complexities, and we are rarely able to leverage the multi modality of these models to give you references back to your answers so that you establish the trust in the answer that you are getting. You are able to tighten the loop in terms of the action that you want to take and get things done. So, those are the use cases that have really been enabled by the technology we have been using. Thanks a lot, Abhijit. And again, delighted to know that between Cohere and Adobe, my holidays are safe. Nick, can we focus a bit on AI agents, if you don't mind? How does Enterprise Search empower this next generation of AI agents and enhance their capabilities? Yeah. So the rapid innovation of agents and the accessibility to build and deploy agents, it's really incredible. Cohere's North platform allows, any user to quickly and within minutes build and deploy an agent, that they can customize to their specific use case. Now, when you think about search systems and agents, you know, a very common analogy is a GPS system, that if the GPS system isn't updated with the most current maps, it's not gonna get to you to your direct to your, to your location. It's it's incomplete, and it's only as good as the data. And agents are are the same way. Agents with, with access to this multimodality, to multilingual data, to, all these different capabilities from the the embedding are going to be the most valuable agents that, that you can quickly and and successfully deploy. So if you're able to then power your agents with this type of, complete, accurate, up to the data information, then you essentially have eyes and ears to the your company's knowledge because now you're giving it the context it needs to provide to be valuable. I was working with a customer, last week building a compliance agent and deploying it on our our North platform. And we worked with their staff to ensure it was securely connected to its regulatory data, its policy documentation. But now their staff is able to have conversational access or conversational experience to these critical policy documents that include diagrams, financial tables, etcetera, and quickly get the complex answers around, you know, can I share what are the data sharing policies in the EU that, you wanna ensure you have the most up to date answer with the right citations? So, these agents really are the future. And over time, as they become more autonomous, that's where Cohere continuously stresses how important the governance and management of these agents are. I can't wait we bring this to life at Teneo too. Abhijit, how do you envision the partnership between CORE and Adobe evolving to meet future market needs? I think the the most important thing for us to have flexibility and being able to be at the edge of, what's coming in. We all know how new models, new technology in this field is coming up every few few weeks. I'll not even say months. So, Adobe at Adobe, we have we have been able to re set us, set us as, in a manner that we are able to leverage those those jumps in technologies, and we are able to unlock new use cases as soon as they are available, sometimes even leading the curve. And that's where our partnership is really, is something that we really appreciate in terms of as soon as new use cases come up, either Coher is already there with those, requirement of those requirement of those technologies, or they are coming out very soon with with with what we need. An example would be embed four model, which, rarely consolidated, having a single model for all use cases across multilingual use cases or multimodal use cases. So these are the kind of innovations that we need, going forward, and, we are really kind of, ahead of the curve on application develop generative AI application development. And that's where I think the partnership will continue to move forward. Thank you very much. I think the buzz today is very much about Gen AI, AI agents, and that makes sense. As a result, sometimes embedding and re rank AI models are perhaps sailing a little under the radar. They are actually at the very core of the engine if we understood well what you just talked to us about, although they can clearly be overlooked. Being able to retrieve and use mixed document types, you both, insist on that text, I believe, images, perhaps audio and video tomorrow, is key to power business growth and to unleash the full potential of Enterprise AI. Thank you very much, Nick. Thank you, Abhijit, Adobe Encore here for having walked us through this, and thank you very much for having, paid attention. Appreciate it. Thank you. Thanks, everyone.