 I hope you're doing well. Welcome to this CUBE conversation with Momento. I'm your host, Lisa Martin. I've got both co-founders here with me, Kawaja Shams, co-founder and CEO of Momento, and Daniela Miao, one of our alumni, rejoins us, the co-founder and CTO. They're here to talk about something really exciting that's about to drop, Momento Vector Index. Daniela Kawaja, welcome to the CUBE. Great to have you guys today. Thanks for having us. Thanks for having us. Super happy to see you again, Lisa. Yeah, likewise. Great to see you too. Daniela, I want to start with you. We're going to talk about vector databases, their role in modern software. Daniela, starting with you, what's a vector? How do I obtain a vector? And why should people care about vector databases? Well, I think before we go there, let's just talk about AI in general. Last time, we spoke about chat GPT and how it's revolutionizing the world as we know it. None of that has changed. If anything, I'm ever more excited about some of the existing applications that's already out there, like GitHub Copilot and all kinds of AI applications. The world is just so busy enabling AI in their applications. And I think now is the time to really care about time to market how quickly can each person launch their, each company launch their products as fast as possible and bring those benefits to their end users. A vector is just a tool that software engineers have in their toolbox. It's always been there, but now vector databases enable them to actually use it to speed up their application and enhance the capabilities like recommendation systems inside their applications. Speeding up and faster time to market is critical for as you mentioned, Daniela, for businesses in every industry and of course on the consumer and business side, we don't want less things slower. Kawash, let's bring you into the conversation. Help us understand how vector databases are different than traditional databases. What are some of the advantages there? Yeah, so Daniela and I come from the traditional database world. We used to be on the DynamoDB team. And one thing that we're learning in the modern day is that old databases prioritized exactness. So if I'm looking for a student's name with a particular ID, the traditional databases are really, really good at that. Or they're good at generating reports like who are the students that kind of fit ages X to Y and so forth. Vector databases are very different in that they help you search for things that are similar to what you're asking for rather than exactly what you're asking for. And if you zoom out and think about it, that is exactly how humans operate with each other. When I ask you a question, you're not gonna give me exactly, it's very rare that I ask you what two plus two is, right? That's what a machine is used for. I ask you, usually we have an interlog, we're having a conversation and that lack of preciseness, that lack of exactness that exists in the human conversation is what leads to discovery because then we often discover things that we didn't even know that we were looking for. So the primary difference between a vector database and a traditional database is that vector databases allow you to find things that are similar to what you're looking for and that accelerates discovery. And that's critical, the acceleration of discovery as patience is one of the things that has dwindled in the last few years and I don't think it's coming back. I wanna expand on what you said, Kawaja, and getting Daniela your perspective as well is how vector databases can be used to enhance recommendation engines or personalized content delivery. If I'm on Amazon and I'm wanting to search for a particular products or products that are related, what is the value that a vector database delivers in that context? Kawaja, we'll start with you and then Daniela, happy to be chatting. Yeah, last week we had Mokon and Manju from Etsy was talking about the role that vector searches plays at a company like Etsy. Now Etsy is a little bit different than Amazon, right? You've got a lot more creative options and like when I think of Amazon, I'm thinking of buying a specific product and products like it with shops like Etsy, you're there to really discover. And a lot of times you don't even know what it is that you're really looking for but you have a sense. You're trying to articulate it and these vector indexes are kind of facilitating your journey in terms of discovering the products. Now, if you zoom out and you look at a really, really simple recommendation system, now this might not be the best recommendation in the world but like you can really simplify our recommendation systems work. You can look at, well, this is Kwaja. Here are the products that Kwaja has bought in the past. Can I express that as a vector? Now, once I have Kwaja's purchase history expressed as a vector, can I find other people who are like Kwaja, who have bought, you know, whose collective purchase history looks similar, not exactly the same, but similar to the things that Kwaja has purchased. And then I don't have to go and recommend to this user exactly the things that other users have purchased. I can take it a step further and say, what are the items that are similar to the items that people like me are buying? So now you've got this really massive opportunity to help me discover items that I didn't even know that I wanted all by doing these similarity searches and taking it to the extreme. And the most powerful thing here is that you don't need to build a really, really sophisticated recommendation system to make this happen. Now everybody can experiment with things like this and enable this type of discovery for their applications. Wow, the power behind that sounds incredible. You know, when we see the rise of obviously here we are streaming today, real-time data streams, the rise of it, sensors, IoT devices, social media, how can vector databases really deliver the efficiency in terms of ingestion, indexing, querying that streaming data so that the discovery is deliverable to whoever wants it whenever they want it? Danielle, of you and then Kwaja, you. I think a lot of it can be captured in a vector which is really when you look at it it's sort of a mathematical representation. We don't need to get into the details here but it's really powerful to be able to capture it in this representation so that you can search over it quickly. You can search over a large amount of data quickly and get sufficient accuracy. If you want exact accuracy it becomes orders of magnitude harder but it is really powerful to have this concept where you get sufficient accuracy yet really blazing fast performance because that's actually how, again, like Kwaja said is how humans operate. We're not looking for the exact answer every single time we just need it to be good enough and we need it fast. We need it like yesterday is how people feel nowadays. It's true and I don't think that expectation is gonna go back at all. Nobody wants less things slower. So get into Momentive Vector Index for me. Kwaja, what is it, why did you create it and how is it different from existing solutions? So as a company, Daniel and I always try to improve the developer productivity and right now, everybody is busy building AI-enabled capabilities and the race is on and time-to-market matters more than ever. So we prioritize, we started looking at what are people spending their development cycles on? A lot of people are building AI-enablement capabilities through a system and vector databases and vector indexes underpin a lot of that. The difference is, it's really, really easy today to build an experiment with a vector index. And as soon as you go and try to operationalize it, take it to production, handle the scale, get the replication, get the availability, like there's a whole lot of knobs that start to appear. You have to take this toy project, you have to deploy it in the cloud, you have to understand how many replicas you need to make sure it's highly available, you have to understand capacity management to make sure it's scaling, you have to understand what indexing algorithms you've got to use, there's all these parameters, you've got to say how many indexing nodes, how many query nodes, how many storage nodes. This, all of that is completely surmountable, but this is any cycles that people are spending on this are cycles that are not going towards innovation in their business. And our hope with this momentum vector index is to eliminate all of that. It's completely transparent. It just has two very simple APIs. Index a vector and then given a vector, find me the top nearest neighbors for that, for that given vector. That's it. All of that complexity that I just mentioned gets abstracted away and that undifferentiated heavy lifting becomes momentous problems so that our customers can continue focusing on their business domain. And that's exactly what they want. Quadra rate customers, they need that time for innovation that are delivering and contributing to revenue. Danielle, I'll talk a little bit about from the CTO's perspective, the momentum vector index. What is in it for customers? What value is it going to deliver to developers in terms of making that innovation more of a reality in their organization? What Quadra and I started with this vision of serverless for everybody. We do mean it literally for everybody. We started in caching. The world is moving towards AI and vector index and we want to be right there helping them accelerate their speeds. I can give you an anecdote last week during our community conference, Mocon. A developer came up to me and said, I'm really excited to use vector index. I have to go and do a tutorial online on machine learning and then I'll be good to go. And I told him right there and I said, let me stop you right there. You do not need to learn any tutorials on machine learning. You're a software engineer. You've worked with databases before. You can work with our vector index. You do not need to learn terms like quantization or even indexing algorithms that Quadra mentioned. Our goal is to democratize this AI development for every developer out there. You don't need a PhD or even a medium expertise in machine learning. It should just be APIs that developers lived and breathe and they should just be building their unique applications on top of it. And that is what really excites me about this service and our ability to kind of provide that acceleration for every single developer out there. I tell people and I told that person at the conference, you can build an AI-enabled application today with momentum vector index. Wow, that acceleration, we're gonna unpack that in a minute. So it sounds like, Quadra, you talked about some of the complexities that momentum vector index is abstracting, but it also sounds like from a learning curve perspective, you've taken out a lot of additional time and resources required to get started. Talk about how you accomplish that and why that's such a value for customers. I think you can draw parallels to how databases used to work, traditional databases, you were talking about them earlier. In the early days, if you wanted a database that can do 10 million transactions a second, you needed a team of DBAs. You need to go buy very, very expensive racks. And there was only a small subset of companies that had the ability to pull that off because of the time, the dollars, and the human resources required to be able to do so. Today, if you want a database that can do 10 million TPS, you just go to the Dynamo console and you say, create table. You don't need to understand all of the distributed systems intricacies. And that has led to a whole new class of innovation because anytime somebody wants to build a product that requires that kind of throughput, they don't have to take a pause, go get a PhD in distributed systems and then come back and build it. That massively increases the pace of innovation experimentation and delivery to the customer. So just like that in a vector database, there's a whole lot of concepts. And even as we build the service, we're finding that customers have to deal with that. And a lot of them are solving, they come to the same conclusion. It just takes them much, much longer to come to that conclusion. And a lot of times wrong configurations lead to outages, performance problems and scalability issues for customers that we can just get out of their way by making the right decisions for them. And that includes things like, just simple things like the number of indexing nodes you need or at what point should you brute force versus at what point do I start to do nearest like HNSW or some type of approximation algorithms? That is something that sure, if you've got expertise in AI, you may wanna go turn those knobs. But for the most of the world, you just wanna build the application and just focus on the business logic. And that is the trade-off that these serverless services make. They provide you with a limited SIF API area, surface area, so you don't have as many knobs that you can turn, but in exchange, you get developer productivity, you get latency characteristics that you can rely on and you get scale characteristics that you can rely on to build experiences that you can deploy at scale for everybody. And I think you just mentioned one of the key points there, Quasar, is that scale, we talked a bit ago about how vector databases themselves can be made more scalable to handle all of the real-time, the high throughput data streams that every organization is doing and to handle and manage that scale with ease while enabling the innovation is really sounds to me like a game changer for customers in every organization because there's competitive advantage to being able to innovate faster, get products to market in the hands of customers, deliver those similar personalized experiences that we all want faster than your competition. Absolutely, and it's really exciting for us to be part of that journey and to kind of accelerate it for our customers. So let's talk a little bit about from an acceleration perspective, you both did a great job of talking about the value in vector databases, what Memento is delivering with Memento vector index, but in terms of abstracting complexity, training, how can Memento vector database accelerate training and querying processes for things like machine learning models that utilize vector representations? What's that acceleration of training? What does it look like? Yeah, I mean, for us, it's less about, so we, the vector index is a small part of the end-to-end kind of pipeline and our very narrow focus on optimizing this pipeline allows us to accelerate the search capability part. The training, the modeling, the embeddings generation, there are other primitives for that, and that's where there's other services that offer that, but right now, what we're launching is focused primarily on how do I store an index to vectors at infinite scale and how do I deliver the search results to you very, very quickly? And the temptation is always there for us to go on the end-to-end pipeline, but it's really, really important for us to kind of maintain that purpose-built footprint that we have, focus on just making the search experience as seamless, as easy to use, performant and scalable as possible before we get distracted by the other areas in this pipeline. And our focus remains on the training that we're eliminating on the human side to be able to do these searches. That laser focus is gonna be a game changer for a lot of organizations. So Danielle, I'll talk to us about when this is launching, when can customers get their hands on it, and what's been some of the feedbacks if you mentioned that one client that said, I've gotta learn all these things first machine learning and he said, actually, no, you don't. When can people start playing around with this? Well, we have the wait list available today. So anybody can go to our website and click to join the wait list. Everybody on the wait list gets sneak preview to our product in a couple of weeks. And then in the next month or so, it will be more generally available to everybody. But if you get on the wait list, you get a sneak preview of the vector index experience. Of course, alongside the vector index, I was very honest when I told that developer, you can build an AI-enabled application today. We will be releasing all kinds of material on how to use it, sample applications. We have a bunch already that's already available. You can build your own chat bar, recommendation system, any kind of image similarity app, like really exciting stuff that people can get hands-on with immediately. Very exciting stuff, of course. Oh, Quasya, go ahead. Oh, sorry, I was just going to say, if you reach out to me or Daniela and mention the show, we'll put you to the top of the list and we'll get you access even sooner. So there are people that are using it. Hey! You know, just let us know. And the feedback so far has literally just been, people are astonished at the ease, like the core difference between what we're building and what exists out there is operationalization. It is really easy to build toy projects and experiments with what's available today. And we're just eliminating that hurdle from toy project to something that you can operationalize, deploy at scale, and rely on. I love that, that time to market, that time to value is so critical. You heard it here, guys. Quasya said, if you reach out to them, you get put to the head of the class on the wait list to get your hands-on moment to a vector index. Because where can clients go and imagine your customers to the website to be able to sign up for the wait list or is there anything special you want them to know? If you just go to gomomento.com and there's a banner right there, you can join the wait list and you just enter an email and we get in touch. We're basically, we have the capability working, we're just letting in a few users at a time. There's been a lot of sign-ups already. So we're just, again, we're deliberately trying to let us few people in as possible so that we can give them a curated experience, but also collect a ton of their feedback. Right, that's the main thing is like, whatever you launch is not going to be perfect in terms of capabilities, in terms of the APIs that are needed. So we're using this as an opportunity to really work closely with a small subset of customers to make sure that what we have built actually works in their specific applications as well. And that's really a nice symbiotic direction that you're going in in terms of helping those, those select few really get their hands on it to help you evolve it as well. Quasi Daniella, congratulations on the evolution of Momento. GoMomento.com to sign up for the wait list for the Momento Vector Index. We so appreciate your insights and your time and goMomento because you guys have a lot of momentum. I know I said that before, Daniella, but I couldn't resist it. Thank you both so much for joining me on this CUBE conversation. I really appreciate your time. Thanks for having us. Thanks very much, Lisa. My pleasure. We want to thank you for watching this CUBE conversation. Keep it right here for more great content. You're watching the CUBE, the leader in live tech event coverage.