Name: Shiv Pullepu | MongoDB
Uploaded: 2024-06-11T16:00:00.000Z
Duration: 20 min 40 s

Clips
News
More from AWS Financial Services Symposium

Shiv Pullepu

Principal, Industry Solutions Financial Services

MongoDB

Three insights you might have missed from theCUBE’s coverage of the AWS Financial Services Symposium

Cloud spending in the financial services industry has been on a growth curve, and Amazon Web Services Inc. has gained momentum in this sector. For financial services companies, broader cloud adoption has led to a heightened focus on data and new ways to organize and manage vast amounts of information to create intelligent solutions using AI.How the financial world is leveraging data to transform operations and increase customer value was a central focus of theCUBE’s coverage of the AWS Financial Services Symposium. TheCUBE, SiliconANGLE Media’s livestreaming studio, provided insights through interviews with company executives, customers and partners. (* Disclosure below

Supercharging financial services with real-time analytics and gen AI

The very fabric of enterprise data utilization is terraforming, as companies move from historic to real-time, operational data. With their split-second sensitivity, these binary bits serve as the heartbeat of innovation.MongoDB’s Shiv Pullepu discusses real-time data for financial services.Given the resources needed for such data and artificial intelligence operations, how are data wranglers such as MongoDB Inc

play_circle_outline Enhancing Financial Decisions: Exploring MongoDB's Real-time Analytics Capabilities for Operational Data

play_circle_outline Use of generative AI and vector search in financial applications

play_circle_outline Benefits of vector search on mobile devices

play_circle_outline Unlocking the Power of MongoDB: Enabling Development and Analytics for Real-time AI Applications

Info

Shiv Pullepu | MongoDB

Shiv Pullepu

Principal, Industry Solutions Financial Services MongoDB

>> The AWS Financial Services Symposium, presented by theCUBE.

>> Welcome back, everyone, to theCUBE's live coverage here in New York City for AWS Financial Symposium '24. I'm your host, John Furrier. We're here getting all the action, getting all the data, generating the data here on camera. We've got Shiv Pullepu, who's the principal industry solutions, financial services at MongoDB. Great to have you come on theCUBE, great to see you.

Shiv Pullepu

>> Thank you. Thank you so much.

>> Thanks for you coming on. You guys are a big partner, obviously with AWS, we cover you guys at reInvent. You have the whole SugarCane thing. Big part of the ecosystem, Mongo is with AWS, so congratulations.

Shiv Pullepu

>> Thank you.

>> Here, we're in the financial service Symposium. So, this is a competitive market and data is the heart of the value proposition in finance.

Shiv Pullepu

>> Absolutely.

>> Real-time analytics, generative AI, vector search are three things that you guys are really pulling together for these new applications that are emerging, and also taking advantage of existing applications. So, what is the real-time analytics position you guys are taking, explain Mongo, how you guys view the role of real-time analytics?

Shiv Pullepu

>> Absolutely. Like you said, the data is at front and central for any industry, more so financial services industry, and it kind of powers all these AI/ML platforms. It's kind of like a fuel to these different models. So, it is important, it's important to actually come up with the right data, if not current data, it's almost impossible to have a current data. You need to have the right data, right data in the different structures and also right data to these models in the movement in real time. That's where we come in. MongoDB provides the right data to these AI/ML models in the moment in real time. So, the gap that we are filling with MongoDB here is that historically, if you think about real-time analytics, the analytics are generally done using historical data. Not so much about operational real-time data, but those days are gone, right? And that's what we do. We do that very well. We provide that real-time operational data that is needed for these real-time analytics applications.

>> I'm watching some of the people here at the event. I saw the keynote upstairs. You got kind of a developer vibe but also kind of a leadership vibe. Solutions are now the big discussion because everyone's been experimenting with the generative AI. Developers are coding with obviously MongoDB and you guys have a great developer community. Obviously everyone knows Mongo has been adopted by developers. It's the standard for developers. But when you think data is now as part of the development process, data is in the applications. It's not just an ingredient, it's wrapping in with the software. So, you're starting to see the gen AI models becoming very instrumental in the software development process.

Shiv Pullepu

>> Absolutely.

>> That's changing the game and changing new opportunities. Can you share your vision and how you guys see that, the confluence of real-time analytics, this new gen AI paradigm, because there's new use cases that are emerging and old ones are being modernized in a new way, in a new era. Can you share, what's at the intersection of those things, what is that enabling real-time analytics plus gen AI?

Shiv Pullepu

>> Definitely. First, let me take you the second part of your question in that how gen AI is helping with core migrations, so legacy migrations. A number of these are financial institutions. They're actually going through massive transformation projects, say like mainframes or other legacy technology. Generative AI actually providing that positive catalyst type of helper to help a migrate the code. So, these larger languages can give you like, say for example you have a PL SQL code, it helps you actually create a MQL, which is like native to MongoDB, right? So, that's a help up that is just providing. And gen AI, just a lot of buzz around gen AI these days. I mean, in addition to the code migrations use case I just mentioned, there are a number of other innovative use cases, and the way I see it is there's a convergence between real-time analytics using generative AI can open up a lot of innovative use cases for our businesses.

>> Like what?

Shiv Pullepu

>> Like for example, fraud prevention, risk management, AML and KYC, right? And it can actually further improve the detection accuracies using the core technology behind generative AI, which is vector search. Using vector search, you'll be able to further enhance these real-time AI applications or predictive analytics applications using vector search, which I said as a foundation for the generative AI models.

>> First of all, first there's a lot to unpack there. So, the gamut of applications from risk management, all the stuff, all the blocking and tackling, all the in the weeds hard stuff can be augmented and helped.

Shiv Pullepu

>> Absolutely.

>> And the main app you see is fraud detection. I mean, if you get that app right, I mean everybody wants that. That's the killer product.

Shiv Pullepu

>> Yeah, absolutely. Yeah, it's a bit innovative approach because you might ask me, "Why vector search? Why can't you use a predictive analytics, real-time predictive analytics?" If you ask me the reason for using vector search, in the case of predictive analytics, there is always this delay in actually addressing emerging of fraud schemes. There is some new fraud that is emerging as we speak today. Real-time analytics and models may not be able to address that because-

>> It's too new.

Shiv Pullepu

>> Yeah, exactly. It's too new.

>> I haven't seen that yet.

Shiv Pullepu

>> You haven't seen that yet because your model is not trained using the data because your model is as good as your data, which is mostly historical data and also real-time data. But in the moment data, you may not be able to incorporate into actually detecting a fraud. That's where I think our vector search actually play a significant role in that as you actually in just the new transactions, about a card payment for example, and you can actually take the transaction information as one vector and also the customer profile as another vector and also anything around that customer. We can combine that and create these vectors and be able to find these different patterns and anomalies at the speed of light and they help address these-

>> So, Shiv, If I hear you correctly, what I'm hearing you say is vector search... Now I'm going to oversimplify it. Vector search is for retrieval. Everyone's doing RAG, retrieval augmentation generation, let's find, similarities vectors are great for that. What you're saying is because of the way vectors work and vectors search, you can see something new because it has a pattern similarity to something else that you can see, and then be directionally correct. You don't have to be a hundred percent accurate, but you'll have enough information through the vector embeds, and AKA the search or notification in this case, that gives you the advantage. Is that what you're saying?

Shiv Pullepu

>> Absolutely. Yeah. That gives you the advantage. Advantage of finding anomalies and patterns in the movement in real time. Because timeliness is so important here and kind proactively activating or proactively working on real time data is so instrumental in preventing fraud. Because in the fraud prevention, you want to prevent a fraud before it happens, right?

>> Yeah.

Shiv Pullepu

>> So, the timeliness, that that's part of vector search. The cost of implementing vector search and also deploying a solution like that on a mobile devices is very efficient. Whereas calling an external predictive analytics endpoint, you may actually add latency to the observations or inferences and may actually make it to impact your customer -

>> The ROI and the benefits are high.

Shiv Pullepu

>> Customer satisfaction as well, right?

>> All right, so what you're getting at here, I love this conversation. You're talking about revolutionizing fraud detection.

Shiv Pullepu

>> Absolutely.

>> Explain that. Explain how you see the revolution happening in fraud detection. What's being disrupted and what's enabling? Because I've always said, and we've always said on theCUBE, this is a disruptive enabler. This wave, it's disrupting, but it's also enabling value. So, tell me what's disruptive because if you're revolutionizing fraud, explain.

Shiv Pullepu

>> Well, so if you're looking at some numbers around the fraud, the amount of losses we see annually just on a caught fraud, just to give you an example in the US, we are losing around a $40 billion per year, annually. We are losing that much amount of money when you have all these latest and greatest technology tooling around like predictive analytics, but we are still losing money. Why are we losing money? I mean, if you ask it that question to yourself and then it becomes very clear, there are some limitations with the existing predictive analytics. It's like one of that, as I said, it may not be able to actually accommodate emerging fraud trends or the emerging fraud schemes and things like that because the fraud is evolving, they're constantly changing. The fraudsters, they constantly change their schemes. The predictive modeling is not actually able to do that in real-time fashion. Whereas a vector search, you can actually create these related data, group the data, and they use the data using vector search to find those patterns. And if you see a fraudulent pattern using a transaction data and the customer data, you can actually block the transaction right on the mobile device, on a web device and assign it to a case management team to actually further investigate that thing. By doing that, what you're doing is you are enhancing, or in other words, you're disrupting the existing real-time analytics, predictive analytics by incorporating vector search-based anomaly detection on top of what you already have. So, this is going to actually improve the performance.

>> When you say what you already have, you mean talking about your existing operational database?

Shiv Pullepu

>> Yes.

>> Which is Mongo, you guys do a lot of Atlas and all that stuff. And then on top of that you got an event-driven, real-time fraud detection system.

Shiv Pullepu

>> Absolutely. That's exactly what I said. Speaking about MongoDB, to do something like that, you need real-time data. That's what MongoDB provides. You have a real-time data and also anomaly detection or a vector search in general requires different structures of data, structured, semi-structured and unstructured data. And you need a database platform that can actually accommodate these types of different structures of data as well as different types of data. For example, operational data is one thing we talked about and also vector data. And you don't need to go and find a license for or buy a new dedicated vector database for this product need. And ideally you wanted to have everything in one platform. That's what MongoDB offers. You have a the-

>> A unified platform.

Shiv Pullepu

>> I'm sorry?

>> It's one unified platform?

Shiv Pullepu

>> One unified platform will help you create these innovative solutions quickly and take them to market very quickly and also even to or iTrade, right? And it improves the developer productivity, like I said. And cost is a big thing. For any type of gen AI solutions, it's going to cut down your cost and improve customer experience. Phenomenal.

>> Okay. Well, we're a huge fan of vector databases as I've always been bragging on theCUBE. All these transcripts that we do go into a database and we apply vector search to them, because the conversations can be retrieved. It's kind of very-

Shiv Pullepu

>> Summarization -

>> It's a very good way to categorize certain technical conversations with a lot of jargon. And it cluster them, the retrieval is phenomenal. So, I see the benefit. The question I have for you is vector search is kind of new and this integrated way, it's vector's been around for a while, but this new application is key. There's been customers out there that have implemented fraud detection systems in the past.

Shiv Pullepu

>> Yes.

>> What do they do? Do they rip and replace? Is there integration? Take me through... I'm sold. I like what you're selling right now. Okay, got it. Revolutionizing fraud detection. You had me at that. All right, I see the benefit, operational database, real-time event system on top. I got it. You have me. What do I do with my existing stuff?

Shiv Pullepu

>> You're not going to throw away your existing stuff out of the window. You're going to still keep using the existing stuff. What we are saying is we are saying how do you enhance existing ecosystem to actually increase the fraud detection accuracy? Like I said, right now we are losing $40 billion annually, which means we're still struggling. How do I enhance the existing ecosystem? So, you're not throwing anything, you're not throwing any rules-based systems we have in place, those are important, they need to stay. And we have a real-time analytics in place, predictive and that those are important. They need to stay. And on top of that, like I said, real-time analytics with this gen AI-based tech like which is a vector search, you can actually-

>> Another layer.

Shiv Pullepu

>> Yeah. Another layer on top of that to help increase the detection accuracy. Even if you make an improvement, in 2% improvement, actually that translates to a lot of money in general.

>> So, essentially they're taking some unstructured data, you vectorize it, put it in a predictive search. Put the client device or app, have the lightweight front end.

Shiv Pullepu

>> Absolutely.

>> And then everything's kind of tied together.

Shiv Pullepu

>> Yeah. Because-

>> On top of the existing stuff.

Shiv Pullepu

>> On top of the existing stuff, we are not throwing anything. Existing stuff is working well, but we could have been better because the fraud losses can be actually reduced better. So, if you think like a developer, you constantly and relentlessly try to improve these tooling around this fraud. Vector search actually provides one such innovative option to improve it.

>> I love this new generative AI. I do believe, I do agree with NVIDIA's CEO, it's a category. Jensen Huang said it's not a static programmed environment, it's generative, it's runtime, everything's new. So, you got to have this new capability. The old ways and what, was static-based rules, right? I mean, this was the old way to do it, right?

Shiv Pullepu

>> Yeah. So, think about it. Say you have a rule at a customer level, any transaction more than $1,000, you block them right there. You have such a rule. What if actually that rule need to be changed? And you may actually go and ask your developer each time if we can change, and it takes a while. And similarly, and even predictive analytics, as I said, and if there's a new emerging trend and there's a new data source that you want to use. For example, social media data, if you want to use alternate data sources to actually help improve the detection, you'll not be able to do that in real time. You need to actually take down your model and retrain it and feature extraction. And there's a lot of cost and a lot of time. So, you would wish there is something like that where you can actually try to actually block some anomalies or fraud patterns right off the bat in a more efficient manner.

>> So, you guys are doing it today. What's been the customer reaction and what's the relationship with AWS?

Shiv Pullepu

>> The customers are so happy. One of the customers I was talking to recently was really thrilled to hear this idea, especially vector search can work very efficiently on mobile devices where you do not have a lot of RAM or-

>> Yeah, that's the edge.

Shiv Pullepu

>> Yeah, get it on the edge. It's not like a big model that you need to make a call-out, but you can do this efficient search on a mobile device, which that's one thing. And that might help improve detection accuracies a lot. And the AWS is a partner of us and it's a . Wile we provide all the data and all the tooling that need to actually create these amazing gen AI-based tools like using vector search. AWS provides that-

>> They're the hardware.

Shiv Pullepu

>> They are the hardware and they help actually create these amazing applications and take them to market quickly.

>> It's funny, some Amazonians get kind of upset when I call them the hardware. I don't mean that in a negative way, but they do have a lot of hardware. They got the GPUs, they got the infrastructures of service. Of course they got Lambda, they have all the software. But from a gen AI standpoint, that acceleration of the computing infrastructure, and that's a term NVIDIA coined and uses, is really important to have that performance.

Shiv Pullepu

>> Absolutely.

>> You guys provide a lot of capabilities from Amazon to pump out more compute.

Shiv Pullepu

>> To quote an example, for example, Amazon, they have, what do we call it? Bedrock, and we have a partnership with the Amazon and we have a deep integration with the Bedrock. Bedrock is your ML gen AI model platform? And in this example, I was talking like a fraud detection using vector search. And any customer can actually leverage Bedrock to create these vector embeddings using a Bedrock platform and then use the embeddings and store them in MongoDB. Like I said, because you wanted to keep a vector data together with your operational data so that they're together and it's easy to search and find anomalies and all that in an efficient manner. But you can still use a Bedrock to create those embeddings.

>> And that's key for people to know.

Shiv Pullepu

>> That's .

>> The embeddings could be any API, OpenAI, Bedrock.

Shiv Pullepu

>> Any API.

>> But storing in a search, vector search-

Shiv Pullepu

>> Together, co-located in the same database, that's the key. And you create an AI data store that is having, like I said, different structures of data and different types of data, operational and vector together in one database platform. Like you mentioned previously, a unified data platform. But also, I just wanted to clarify that thing. Yes, our customers, they can bring their own vector embedding platforms like as I said, a Bedrock or OpenAI embeddings, although I recommend Bedrock.

>> Stay in the family, so to speak.

Shiv Pullepu

>> Yeah.

>> I mean, OpenAI is Microsoft. But again, this is an important architectural design element.

Shiv Pullepu

>> Absolutely.

>> Knowing the embeddings and storing it properly with the operational data-

Shiv Pullepu

>> Is a key....

>> gives you the key event-based-

Shiv Pullepu

>> Advantage....

>> enablement and the real-time is where you capture the pattern -

Shiv Pullepu

>> Absolutely. You capture the pattern real-time. Real-time unknown patterns, right? Patterns are not known, not until the point. Right? Or the trends that are not known up until the point. Or the models that you actually trained, the predictor models that are not available of the pattern yet. You don't have luxury of actually retraining them. And that's where you can actually use the group of data and use this vector search to, like I said, to find anomalies there.

>> Well, Shiv, I really appreciate you coming on theCUBE. It's a fantastic conversation. Of course, we've been very pleased to been covering MongoDB over the years. We're a customer, small customer, but of course you guys have developers as small customers, but also a lot of large enterprises. And the success continues. And I think this is just another example of you guys enabling a lot of value for building on top of these operational and other databases that you guys provide.

Shiv Pullepu

>> Yes. Yeah. If I want to actually use one closing statement, MongoDB is that one unified data platform, which can actually be just any type of a data structure. And you can store different types of data and provide that real-time sources to any type of analytics applications and the gen AI applications.

>> Awesome. We are here at the AWS Cloud Financial Services Symposium. I'm John Furrier, your host of theCUBE. We'll be right back after this short break.

Shiv Pullepu | MongoDB

>> The AWS Financial Services Symposium, presented by theCUBE.

Shiv Pullepu

>> Thank you. Thank you so much.

Shiv Pullepu

>> Thank you.

>> Here, we're in the financial service Symposium. So, this is a competitive market and data is the heart of the value proposition in finance.

Shiv Pullepu

>> Absolutely.

Shiv Pullepu

>> Absolutely.

Shiv Pullepu

>> Like what?

Shiv Pullepu

>> Absolutely.

>> And the main app you see is fraud detection. I mean, if you get that app right, I mean everybody wants that. That's the killer product.

Shiv Pullepu

>> It's too new.

Shiv Pullepu

>> Yeah, exactly. It's too new.

>> I haven't seen that yet.

Shiv Pullepu

>> Yeah.

Shiv Pullepu

>> The ROI and the benefits are high.

Shiv Pullepu

>> Customer satisfaction as well, right?

>> All right, so what you're getting at here, I love this conversation. You're talking about revolutionizing fraud detection.

Shiv Pullepu

>> Absolutely.

Shiv Pullepu

>> When you say what you already have, you mean talking about your existing operational database?

Shiv Pullepu

>> Yes.

>> Which is Mongo, you guys do a lot of Atlas and all that stuff. And then on top of that you got an event-driven, real-time fraud detection system.

Shiv Pullepu

>> A unified platform.

Shiv Pullepu

>> I'm sorry?

>> It's one unified platform?

Shiv Pullepu

>> Summarization -

Shiv Pullepu

>> Yes.

Shiv Pullepu

>> Another layer.

Shiv Pullepu

>> Yeah. Another layer on top of that to help increase the detection accuracy. Even if you make an improvement, in 2% improvement, actually that translates to a lot of money in general.

>> So, essentially they're taking some unstructured data, you vectorize it, put it in a predictive search. Put the client device or app, have the lightweight front end.

Shiv Pullepu

>> Absolutely.

>> And then everything's kind of tied together.

Shiv Pullepu

>> Yeah. Because-

>> On top of the existing stuff.

Shiv Pullepu

>> So, you guys are doing it today. What's been the customer reaction and what's the relationship with AWS?

Shiv Pullepu

>> Yeah, that's the edge.

Shiv Pullepu

>> They're the hardware.

Shiv Pullepu

>> They are the hardware and they help actually create these amazing applications and take them to market quickly.

Shiv Pullepu

>> Absolutely.

>> You guys provide a lot of capabilities from Amazon to pump out more compute.

Shiv Pullepu

>> And that's key for people to know.

Shiv Pullepu

>> That's .

>> The embeddings could be any API, OpenAI, Bedrock.

Shiv Pullepu

>> Any API.

>> But storing in a search, vector search-

Shiv Pullepu

>> Stay in the family, so to speak.

Shiv Pullepu

>> Yeah.

>> I mean, OpenAI is Microsoft. But again, this is an important architectural design element.

Shiv Pullepu

>> Absolutely.

>> Knowing the embeddings and storing it properly with the operational data-

Shiv Pullepu

>> Is a key....

>> gives you the key event-based-

Shiv Pullepu

>> Advantage....

>> enablement and the real-time is where you capture the pattern -

Shiv Pullepu

>> Awesome. We are here at the AWS Cloud Financial Services Symposium. I'm John Furrier, your host of theCUBE. We'll be right back after this short break.