 Welcome back to Las Vegas, everybody. This is Dave Vellante, and we're here in Las Vegas as part of SuperCloud 5. Savannah and Lisa are live in our Palo Alto studio. We got Rob and Rebecca coming in from Barcelona. Well, it's late in Barcelona, but they're probably still up drinking Sangria. And we're here in the Emerald Lounge inside the Mongo Hankspace. I'm really excited to have Ravi Damaravan, who is the founder and CEO of XFluence, XaFluence, XaFluence, XaBig. We're talking big data, we're talking AI. Ravi, welcome to theCUBE. Thanks for coming on here at ReInvent. Thanks for having me. So I'd like to ask anybody with a founder in the name, why'd you start the company? I was having the role of a data and analytics practice leader for several system integrator firms. I was also responsible for building platforms and solutions that solve problems. I found that in the healthcare space, there are quite a bit of problems that can be solved with data and analytics. That's what got me into starting the company. So healthcare is really hard. Not only is everything stove piped, but you have such a strict privacy requirements. So what gave you the confidence to take down that challenge? I can firmly believe that the market to handle the privacy has quite been improved, primarily because of the hyperscalers and the clown providers and the likes of MongoDB giving higher emphasis on security. What we do is we make sure that we give highest priority to the PII data that gets anonymized and it has various level of encryption security that we enforce. And it's very hard to get data either in rest or in motion to anybody to get it. And we have multiple controls, audit practices in place to make sure that it is something that doesn't get into other sands. So with the technology evolving or regarding the security and others, we feel confident to handle that part of the problem. Where we see opportunities also, that there is quite a bit of problems that disjointed healthcare systems don't talk to each other and there was an opportunity for us to really solve the problem. So talk a little bit more about the problem that you solve. Are you bringing in disparate data so that I can do analysis or take action on that? What are you doing that's unique that was white space problem that you had to solve? So what happened was when a typical payer system or a clinical provider system, the data stays in a very validated environment with a lot of governance and controls that only certain segment of people will be having access to and all. With the formation of the data lake and the ability to bring data from different sources. For example, we bring data from different claim systems, different authorization systems, different medical care management systems. We are able to identify the patient profile because of the master data management concepts that can be applied in MongoDB. We are able to create a longitudinal view of a patient, get the patient data and do various level of analytics on them. We do analytics to the level of doing the risk score assessment, looking at their disease methods, population health. We look at the overall what impacts them in their care gap mechanism. So wait, so you're building an analytics application on top of Mongo. Correct. Okay, explain why you would use Mongo for that and not like a redshift or a snowflake or a Databricks. Very good question. The reason why we used Mongo is one, the number one factor is that Mongo is the only platform that allows you to build lightweight master data management concept. For example, if you look at it, patient data can come from 30 systems, but I need to know which system is the system of record. To do that, I don't need to invest into large master data management products, I can leverage Mongo. In addition to that, the delivery mechanism of the data, it can be distributed on a mobile, on a cloud and it can be built as a native app. And I am reducing my total cost of it to be very centered towards one system being there. In addition, what really got us into using Mongo was the time series type of data that comes from device manufacturers that needs to be looked into from a time series aspect. And the ability for writing machine learning programs in an open way. When we built the platform, we also saw the emergence of fire interoperability becoming the de facto standard, which is primarily supported in a JSON format. So we saw that the value of Mongo being JSON centric to be the prime factor. And it's just so easy for developers to take advantage of it, so you can focus on the work that you have to do. Are you confined to healthcare? I mean, it seems like there would be some other industries where there's disparate data. Even in financial services, data's all over the place. Do you apply this technology to other industries? We need to apply to financial services too. Financial services, we have partnered with FIS to jointly take our platform to asset side management in the markets. So we have master data management that is built on Mongo for financial services, but majority of the company is focused on healthcare. Well, I saw, okay, so we've got to ask about AI. I mean, there's so many use cases potentially for AI and healthcare. Which ones are you contemplating or tackling? We already started building some of those where it started was we had a fire platform which required mapping of data sources. We saw that Generative AI, a right fitment to do mapping easily without having to build business analysts to do that. And at least it can help business analysts to make that transition faster. We also saw that there is an opportunity for doing that with clinical trial process with the protocol mapping. For patient recruitment, patient research and studies, we saw that Generative AI, we built our own LLM models and the ability for using Mongo's vector database to refine our models was wonderful. So we built a model for clinical trial protocol. We are also working in the specialty chemical industry to find out new molecular research to figure out how the molecules are formed with easiest route manufacturing methods. And we were able to do that for a couple of clients in the pharma industry, that's their ability to do the total manufacturing of the molecule. If it takes six to eight months to research that, we were able to reduce it to three to four months using Generative AI methods. Wow, okay, so that's real productivity driver. But did you say you built your own LLM? We will talk about LLM. Why did you build your own LLM as opposed to not using an open source or an anthropic or a Lama2 or something that you could get from Amazon? Customers had some kind of inhibition because when the LLMs are a lot of IP level data that resides, they want the language learning model not to do any public routes into internet or any other place. So they want to confine it to their own environment within the thing. In the space we work in, especially in the pharma industry and research and chemical industry, there are a lot of documents that are proprietary and tribal knowledge to the system that they work with. So it can't get out of it. So we took a model, we open source model, configured it into their environment and made a proprietary LLM for them. Ah, okay, so you started with open source, obviously. Yeah, open source, yes. Did you share what you used? Was it Falcon or? We used Falcon, we used a couple of more Python based models that we configured. And then shaped that for your own needs. And you're doing that in a very specific use case for healthcare, so it's a domain specific model. Is that right? We are, as an organization, we are very focused on domains. We believe that the problems can be solved by understanding the systems as well as the problem statement in a peculiar way. So the way we are working is solving the problems in the healthcare and payer and provider space. Does that work happen in the cloud or does it happen on prem? It happens in the cloud, we work on Atlas, we work on all the three clouds, primarily IWS, GCP are the primary ones. Do you have customers that say, we don't want to do it in the cloud, we want to do it on prem? There are many. And can you help them? Yes, we do, because all our mechanism of the LLM models and all are Kubernetes and Docker deployable, which is in DevOps mode. So we can bring it into an environment into MongoDB Enterprise Server or MongoDB Custom and bring it on on prem. For the customers that have not embraced the cloud strategy, we still do the on-site deployment. And it's relatively small. I mean, it's not trillion parameters, right? I mean, it's relatively small. But still, how did you train it? You got access to cloud GPUs and trained it? We have access to cloud GPUs, but the kind of parameters and the configuration settings are evolving. The more they are using, we have to increase the GPUs and consumption, but the customers are happy with it because they are able to reduce their research time and increase the productivity of it. I feel like during COVID, all we talked about was COVID. And now all we talk about is Gen AI. Do you feel like that's appropriate? Has it affected your strategy in a way that you mean just mentioned some productivity metrics that were very impressive? Do you feel like we're over rotating on this or maybe we're under-rotating that the potential is there that we should really be leaning into this conversation and affecting our strategies dramatically? When Gen AI came in, I was a little skeptic about the ROI that it would generate. But when you are tuning it for a right use case and for a right level of cost controls that you have, there is a value to it. For example, there are organizations that did it for service request automation, which reduce their overall cost of using the teams. But there are places where research documents, especially the PDFs and the generative information that takes a whole lot of manual effort to extract. You can use Gen AI in a simplified way. Without having to overdo it, you can control the configuration, compute, infrastructure and all. Define your ROI criteria and use it very intelligently. But if you overdo the hyper parameters and do a lot more of unrelated usage, then it may end up into more cost spent. And you were early into building this LLM. So you must have been using a vector database before Mongo announced its vector search. Have you migrated over to Mongo Vector Search? We have. We are the early ones who took the value because Mongo Vector Search and Mongo Vector Database allows you to confine the research to fragment and refine your research in terms of what you want the output to be. So we are able to pass more information, we are able to train more data that is relevant and leverage the usage of the value of the vector database in a big way. And you determined there was no need to have a separate vector database infrastructure that the advantages of consolidating that in Mongo, there weren't, were there trade-offs that you lost because you didn't have the dedicated vector database or no? There was a trade-off when vector database of Mongo was not there, but I kind of feel that when, we were always using Mongo for all the aspects, so we started naturally using it as an extension because we want the total cost of ownership to be minimal. What do you want from, if you could write a wish list, Dev Itchachari is sitting here, is it Dave, what is it that you would ask him to do that would make your life better? I kind of feel that the ability of more machine learning models being available to use the data in a data-friendly environment and the ability to be translating even image-related DICOM to non-DICOM, we are able to read the metadata and all, but there is an aspect of confinement and other things that you could do for more unstructured data, that would be one ask. The second one is probably having much more proprietary frameworks of language learning model that we don't need to use more public APIs. Having Mongo-owned centric data APIs would be a... More consolidation into the platform, more value in the platform. What about the cloud vendors? If you had all three cloud vendors in front of you, you said you're running on all clouds, what would you ask them for? The cloud vendors are giving their own refinement of their models and again the model outcomes or when I run it on Google's to AWS, to Microsoft, the outcomes are completely different. Consolidating it and looking at it with what outcomes OpenAI would give, the confidence level and the degree of value data that comes out of it would be the ask, like how much of it is trust, how much is bias built, how much of non-bias is built in, that transparency would be helpful. Did your customers, did the OpenAI governance meltdown and the firing and rehiring and all the drama, did that cause concern amongst your customers? Customers were always open for simple information search, they were okay with using OpenAI kind of a framework, but when it came to their own proprietary data, they were confined in asking what, where it would reside and where the IP would reside. So we went with the proprietary LLMs, that's what our experience... So you didn't have to worry about that. Right, right, because you had that, you had that fenced off. That part of it. You're the LLM vendor, so you don't, I presume you, well, let's see, I guess you do have, you have access to the metadata, but you don't have access to the data. Data itself, we don't have the access to it. And Mongo is an investor in your company, right? What's next for you guys? We are working, we are growing on a 100% growth rate, we are working on the industry, we are focusing our research, we started with payer and provider, now we are extending it to clinical trials and pharma, that's next for us. We see there is an intersection of patient engagement in all the micro-verticals in healthcare, I think that's where we want to get our... Yeah, nice. Robbie, thanks so much for spending some time with us, really appreciate it. It's, thank you very much. And best of luck to you and the company. Thank you very much. All right, keep it right there, this is Dave Vellante, John Furrier's in the house as well, he's actually upstairs now doing other interviews on the third floor here at AWS Reinvent. This is SuperCloud 5, and we'll be right back right after this short break.