Philip Rathle of Neo4j, chief technology officer, joins theCUBE hosts in San Francisco to examine how graph technology underpins modern artificial intelligence. Rathle discusses architectural patterns such as GraphRAG and retrieval-augmented generation, the enterprise knowledge layer, the role of ontologies and schema flexibility and the path from research to production-ready systems. They provide practitioner and customer perspectives throughout the conversation.
Rathle cites University of Newcastle research showing GraphRAG improves accuracy by up to 80 percent and doubles answer rates compared with vector-only RAG. They note GraphRAG can reduce inference cost while increasing answer quality, reinforcing its value for enterprise AI implementations.
Rathle emphasizes the enterprise knowledge layer as essential for governance, explainability and data sovereignty and argues that knowledge graphs often determine production success. The discussion highlights factors to consider when moving from research to production, including architecture, ontologies, schema flexibility and governance.
Relevant topics include graph database design, knowledge graph engineering, GraphRAG architectures and production-ready AI systems.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems.
Thanks for confirming your account. Now you can access theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems with this email address.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Philip Rathle, Neo4j
Philip Rathle of Neo4j, chief technology officer, joins theCUBE hosts in San Francisco to examine how graph technology underpins modern artificial intelligence. Rathle discusses architectural patterns such as GraphRAG and retrieval-augmented generation, the enterprise knowledge layer, the role of ontologies and schema flexibility and the path from research to production-ready systems. They provide practitioner and customer perspectives throughout the conversation.
Rathle cites University of Newcastle research showing GraphRAG improves accuracy by up to 80 percent and doubles answer rates compared with vector-only RAG. They note GraphRAG can reduce inference cost while increasing answer quality, reinforcing its value for enterprise AI implementations.
Rathle emphasizes the enterprise knowledge layer as essential for governance, explainability and data sovereignty and argues that knowledge graphs often determine production success. The discussion highlights factors to consider when moving from research to production, including architecture, ontologies, schema flexibility and governance.
Relevant topics include graph database design, knowledge graph engineering, GraphRAG architectures and production-ready AI systems.
>> Hello, I'm John Furrier here with theCUBE in San Francisco for Neo4j Graph Talk. This is an event that brings all the industry insiders, technologists, and customers together to talk about the innovations in AI at the graph level, really powering the secret sauce for what we're seeing scaling AI. And that's ontologies, that's getting data fast, feeding the AI, and having intelligence. It's becoming the key piece. Philip is back on theCUBE. We just saw each other in Paris at the Raise Summit, gave the keynote talk this morning, CTO of Neo4j. Philip, thanks for coming back on theCUBE. Appreciate it.
Philip Rathle
>> My pleasure. Thanks for being here, John.
John Furrier
>> I love the Graph Talk, very intimate events with customers and industry insights, really getting into the technology and the innovation side. There's going to be one in New York in September. We'll also be covering that. We're going to hear from practitioners, but you gave the keynote, set the table. First of all, what did you talk about and what is this event about from your perspective? What are people talking about?
Philip Rathle
>> Yeah. So we made a shift from having one big global event to just spreading them out across 30 or so major cities throughout the world. And it gives customers more of an opportunity to be on stage and share their stories, which is ultimately what it's all about. So we've got a bunch of that today. And I know you're going to be interviewing a number of speakers. I kicked off the day by putting things into perspective. We're now four years post the release of ChatGPT, so well into the gen AI era. And when people first started building solutions, people were just throwing spaghetti on the wall and trying everything possible. And now we know enough about the patterns to be able to come back and come up with a shared vocabulary and a shared set of architectural patterns. So I shared those. And then leading up to that, I shared, "Well, okay, there's still a foundation that's been built up over a decade plus pre-2022." And really getting into that, the origin of graphs leading into them becoming an ISO standard and used across all the world's top companies and many startups.
John Furrier
>> Talk about the historical nature of graphs. I think it's important to do a little retrospective to get kind of the modern current because you know me, everyone knows I'm a huge fan of graph. I always talk about it. I think it's super important. Now the world sees it. So maybe misunderstood for many years, but now people are getting it. But I still hear people talking about, "Oh, graphs are great." They don't actually don't know the players, who does what, what scale. "Oh, they're the best." I mean, there's always claims. I think what I like about Neo4j is you guys have been there for a very, very long time. You can have a perspective. You have a lot of data. Share the data you learned over the past four years with the ChatGPT, but why it's important and where it came from and why Neo4j? Because there's a lot of confusion around what to choose and why to choose it as people are just jumping in and doing it.
Philip Rathle
>> Yeah. So it really started with us asking the question way back in the day of, "Look, how does the real world and the digital world shows up... It shows up as complex systems and it doesn't show up naturally in tables unless maybe if you're an accountant, it looks that way." But most systems show up as networks, be it payments, biology, ecology, telco, hierarchies of supply chain, HR, bill of materials, so on, or paths and journeys. And so that asked the question, "Well, what if we built a system that represented those in the way that they show up instead of putting them into tables? And do we think we could do a 10X better job and provide significantly more value? Because there's no point in starting a startup if you don't feel you can do at least an order of magnitude better than the other way."
So that was the origin story. And then some of the things that we foresaw, which have become really important for AI, is, number one, if you have to decide ahead of time all the kinds of data that you're going to need to store, then you're going to be really limited. You want to be able to evolve really easily. In the database world, that means evolving your schema. So having schema flexibility, having a language that you can use to both describe your data, but then also query your data. And so we're a big fan of open. We see ourselves as most of our work is creating a new category of technology. That means inviting in people to compete with us. And then we get to win over people's business because we do the best, not because of lock-in.
John Furrier
>> What's the modern update as of today? Okay, four years in, what's the state of the art? What's the top conversations happening here in the room and around the world with graphs?
Philip Rathle
>> Enterprise knowledge layer is the big topic. In fact, my colleague Jesús Barrasa just published an amazing piece on this that in some ways is the spiritual successor to the GraphRAG manifesto, which I wrote two years ago. So GraphRAG describes the pattern of having an LLM call out to a knowledge graph so that you externalize your knowledge in context. It doesn't live in the model. It lives in a system of knowledge. And that gives you better accuracy, explainability, and governance. But fast-forward, what we've actually seen is that take shape is an important architectural element called the enterprise knowledge layer. And so he lays out essentially the anatomy of that layer. And that post just came out this week.
John Furrier
>> There was a research report from the University of Newcastle had results that compared GraphRAG to RAG. What are some of the findings that you're seeing? Is it more accurate? Is it faster? What are some of the updates in terms of the benchmarks? And then we'll get into how that ties to economics.
Philip Rathle
>> Yeah. So we've had lots of evidence through our customers that GraphRAG improves accuracy, provides governance, improves explainability. But it's always really nice when you can have a third party world-class academic institution do some research because then not only does that give you some objective results so you know what to expect, but you can understand the why behind it. So University of Newcastle, which is top university for computer science globally, built up multiple data sets, created over 500 complex queries, ran them with LLMs with just vector-based RAG, and then did the same with GraphRAG, and found that you get 2X improvement in queries that have an answer versus the model saying, "No, I actually don't know. I don't have enough information." 80% improvement in accuracy and just significantly better qualitative results and more precise answers as well, which are more useful. And then last but not least, fewer tokens. We're all talking about tokenmaxxing. But it turns out if I can be more surgical and precise about what data I bring back and feed to the model, not only do I get better results, but it's far cheaper than just flooding the zone with context and then churning away with a very expensive model.
John Furrier
>> So price has come to factor in there. All that for aside, but you get accuracy and speed with the GraphRAG. How is it translating into economics? I mean, you're a CTO, you have the technical piece. I was just commenting yesterday at the AMD Advancing AI event that the winners in that game of Nvidia and AMD and all the Intels is that whoever can redesign the computing architecture to get there first to enable AI to align with how humans think and work. And that's really the new user experience, how do people think and work? That's the brain. That's why I love graphs. But economics became the big thing. I want to see benchmarks around not just technology, but economics. What are some of the value pieces you're seeing come out of this?
Philip Rathle
>> Yeah. There are two levels of this. One is the system has to make it to production before you can even get any economic benefit. And then number two, once you've done that, then it's a question of refinements. One of the other speakers earlier this morning was kind of quipping about how there's the saying that no AI systems are making in a production, which is obviously an exaggeration, but many still don't. And he said, "Well, the difference between the ones that do and the ones that don't is the ones that do actually have a graph, this part of the picture." And I think some of this can be. Jensen has the five layer cake and that describes the business from his perspective and what it takes to run models. But when you're building AI systems, it's not just that stack. You have the whole data layer.
John Furrier
>> It's the NVIDIA view.
Philip Rathle
>> Yeah, it doesn't-
John Furrier
>> There's no data in there.
Philip Rathle
>> There's no data in there. There's no data infrastructure in there. So what you actually need to make it work is to externalize your data and have a system to do that. So having a knowledge graph actually is often the difference between no value and a lot of cost and value just making it into production. But then what this report by the University of Newcastle says is, "Well, assuming that I can get good enough into production with just vector-based RAG, then there's significant refinement, 2X improvements by going from just vector to GraphRAG."
John Furrier
>> Okay. So take me through a day in the life. You know I love to kind of walk down the checklist because I always think maybe I could use that for theCUBE data. But people generally want to see the engagement path. How do I get started? Let's just say I've got a bunch of vector embeds laying around, doing some RAG on it, and I want to go to a graph. What happens? Take me through the setup, the build out, standing up knowledge graphs. What does it take?
Philip Rathle
>> So to answer this, let me actually talk at a high level about this anatomy of the enterprise knowledge layer, which Jesús laid out. You have three kinds of graphs that you can have at different layers. There's everything that's in the ontology bucket, which is your business meeting mapped to your technical meeting where we use the term semantic layer or semantic map for that. So that's one bucket. Another bucket is the data. So you have your reference data, your own internal data that might be graph shaped. And then the third one is memory. So by this it's like long-term memory. So back to the brain analogy, neurons, synapses like nodes and relationships. Ontological memory, memory of your procedures and your rules, which goes back to context graph and decision traces. So first, I want to decide what do I want to start with? You don't have to boil the ocean, you can just start with one. And then depending on what I want to start with, all right, where is the data right now? Is it in a relational database? Then I need to map it in from there. If is it unstructured, then I map it in from unstructured. And ideally I have an existing ontology that I can use to convert that. And the good news when it comes to building these is this used to be a significant point of friction, but now AI comes to the rescue and we've been able to supercharge a bunch of our tooling with AI so that an LLM can take a relational database schema, click a button, and in under two minutes, you end up with a graph schema compressing what used to be a week of work down into two minutes and then you're off to the races.
John Furrier
>> What's the coolest thing you're seeing here being discussed that you had to kind of point at kind of the concepts of what people are showing as people get more graph reps in, building out, getting into production. What's the big thing that jumps out at you? What are you most excited about?
Philip Rathle
>> It's the diversity, the horizontal diversity of the kinds of problems that are solved. I just get so much joy out of having been doing this for close to 15 years and seeing people solve problems that I've never imagined even existing.
John Furrier
>> When I saw the Hadoop phase of data storage and large scale big data, the first wave of big data, people were so excited. And then there was like, "Oh, this is so hard." And then you start to see data lakes. And then there's a general enthusiasm in the data world of data people who love data like us. It wasn't like a total standing ovation, but it was still euphoric. "Ah, yeah. Data lakes are cool. I can do pipelining." Still a lot of stuff to rigid. But with graphs, I've noticed that people who are working with graphs are seriously pumped up. They have such a high from playing with them and working with them that whenever I do an interview with someone who's working with graphs, it's almost like relief meets the future. Explain the phenomenon because it's a common thread. I know you work with a lot of folks, you see the same thing. There's general excitement, but it's not just, "I'm excited." They're super pumps. They're sharing and they're telling all their friends. They're telling everybody. Why is that the case? Is it just so refreshing? Is it the scales, the flexibility? Why are dating nerds and geeks getting super pumps on graphs?
Philip Rathle
>> It's because if I'm trying to solve an AI problem that's engaging with the world, and it's in nearly 100% of cases, it's this world that shows up as a complex interconnected system. And as soon as you're dealing with it in tables, it's like having some homework.
John Furrier
>> It is. It's like doing homework that you don't have to do.
Philip Rathle
>> It becomes drudgery because you're going from a 3D color system into 2D black and white and trying to re-extract it and do all the color touch up and re-extrapolate it back into 3D. And with the graph, you retain that. So it's so much more empowering. And I love your... Yeah, it's a combination of relief and this is the future and I'm untethered and I can move forward. I'll just say this year I've gone out of my way a little bit to give back in academia. I've done some guest lecturing at U Chicago and Stanford and other places. And because for many reasons, including the fact that there's an ISO standard behind this, professors are now... It's an obligatory part of their database curriculum. And students come to me and say, "Look, I learned Postgres. I learned my SQL. I learned Neo4j. Man, graphs are like, it's so much easier." Now, that's not the reflex of someone that spent their entire career thinking in tables because you forgot how to see color. You forgot how to deal in 3D.
John Furrier
>> grinding. It's like the bark of the tree. You don't see the whole forest. I think that's true. One of the things I will point out, I'd like to get your reaction to, is that also there's something about achieving something. When people use graphs, they get different outcomes. And it's almost like found money. It's like, "Oh wow, we discovered this new way to do things." And that seems to be consistent. Graphs bring you into a whole nother realm of, "Wow, I can actually do new things. My capabilities are more headroom for things to do." And so they're knocking down business goals. Folks I've talked to are like, "Hey, I just keep the expectations low. I just deliver the hardest problems."
Philip Rathle
>> I'll give you an example. Just a couple weeks ago, we were doing a POC with a, I can't say which country, but let's say a national tax agency. 48 hours from the start of a POC, they identified more than $100 million in tax fraud. That's just in 48 hours with a subset of data. And this phenomenon, if you generalize it, it's if you've been walking around just kind of limited with your blinders because you're looking at this data in 2D, you bring the data into a graph view and all of a sudden all these things become blindingly obvious, which before you just simply couldn't see.
John Furrier
>> Yeah. And it was hidden. You had to query it out and extract it out. It's like pulling oil out, fracking oil. It's hard. And it's not always the best. Final thoughts. As you look at the second half of the year, I know we were talking about in Paris sovereignty. We're trying to get an event going, get a small group of leaders at the NYC here in San Francisco or in New York. I think it'd be in New York. But as the community comes together, what do you hope to see with other leaders as they come together and look at things that are going to be big problems to solve or opportunities to capture around sovereignty, AI production, value creation? What's your view and what should the conversation be?
Philip Rathle
>> Evolving our view of what the AI stack is and recognizing the need for an enterprise knowledge layer now that we have a clear definition of it. But then that actually gives organizations their own sovereignty. There's been a lot of talk in the news lately about, "Well, hey, does some AI LLM foundation model provider have all my data?" If you keep your data separate and in a separate layer, then it becomes sovereign to you. And then nation states, sovereignty is a big topic. And we had this whole great conversation with Amit in Paris a couple of weeks ago at Raise. This is a really big topic on the world stage where the definition of sovereignty entrains the definition of what does my AI stack look like. And I'll add one more thing. Our CEO, Emil, just revealed recently that our net new revenue last quarter was greater than our net new revenue all of last year. There's just so much growth in this space and we're just one vendor and it's one microcosm and a lot of the growth is community-based.
John Furrier
>> Yeah. Yeah. I really think it's going to be a great tailwind for you guys and the whole industry because as these computer architecture from AMD and Nvidia and everyone else gets better, they're redesigning compute architectures to the way people think and act. And AI brings agency to users and they can be empowered. The interface is no longer gooeys or windows, images. It's natural language. And so that's going to give the empowerment to the user. So whoever can enable that as fast as possible in the way they think, which is a brain, which is basically neurons and graphs, our brain is a graph, I think that's the future. And
Philip Rathle
>> In a way that's understandable. You can look at a graph and understand it and then a machine can execute on it. It becomes a shared language between human and machine.
John Furrier
>> It's like TVs, 4K is great, but 8K is coming around the corner. So we'll see. Phillip, thanks for coming on theCUBE. Appreciate the conversation.
Philip Rathle
>> Thanks, John. Always a fun time.
John Furrier
>> All right. I'm John Furrier, here in San Francisco for Neo4j Graph Talk San Francisco, of course. New York City, the next stop will be in September. We'll be right back.
>> Hello, I'm John Furrier here with theCUBE in San Francisco for Neo4j Graph Talk. This is an event that brings all the industry insiders, technologists, and customers together to talk about the innovations in AI at the graph level, really powering the secret sauce for what we're seeing scaling AI. And that's ontologies, that's getting data fast, feeding the AI, and having intelligence. It's becoming the key piece. Philip is back on theCUBE. We just saw each other in Paris at the Raise Summit, gave the keynote talk this morning, CTO of Neo4j. Philip, thanks for coming back on theCUBE. Appreciate it.
Philip Rathle
>> My pleasure. Thanks for being here, John.
John Furrier
>> I love the Graph Talk, very intimate events with customers and industry insights, really getting into the technology and the innovation side. There's going to be one in New York in September. We'll also be covering that. We're going to hear from practitioners, but you gave the keynote, set the table. First of all, what did you talk about and what is this event about from your perspective? What are people talking about?
Philip Rathle
>> Yeah. So we made a shift from having one big global event to just spreading them out across 30 or so major cities throughout the world. And it gives customers more of an opportunity to be on stage and share their stories, which is ultimately what it's all about. So we've got a bunch of that today. And I know you're going to be interviewing a number of speakers. I kicked off the day by putting things into perspective. We're now four years post the release of ChatGPT, so well into the gen AI era. And when people first started building solutions, people were just throwing spaghetti on the wall and trying everything possible. And now we know enough about the patterns to be able to come back and come up with a shared vocabulary and a shared set of architectural patterns. So I shared those. And then leading up to that, I shared, "Well, okay, there's still a foundation that's been built up over a decade plus pre-2022." And really getting into that, the origin of graphs leading into them becoming an ISO standard and used across all the world's top companies and many startups.
John Furrier
>> Talk about the historical nature of graphs. I think it's important to do a little retrospective to get kind of the modern current because you know me, everyone knows I'm a huge fan of graph. I always talk about it. I think it's super important. Now the world sees it. So maybe misunderstood for many years, but now people are getting it. But I still hear people talking about, "Oh, graphs are great." They don't actually don't know the players, who does what, what scale. "Oh, they're the best." I mean, there's always claims. I think what I like about Neo4j is you guys have been there for a very, very long time. You can have a perspective. You have a lot of data. Share the data you learned over the past four years with the ChatGPT, but why it's important and where it came from and why Neo4j? Because there's a lot of confusion around what to choose and why to choose it as people are just jumping in and doing it.
Philip Rathle
>> Yeah. So it really started with us asking the question way back in the day of, "Look, how does the real world and the digital world shows up... It shows up as complex systems and it doesn't show up naturally in tables unless maybe if you're an accountant, it looks that way." But most systems show up as networks, be it payments, biology, ecology, telco, hierarchies of supply chain, HR, bill of materials, so on, or paths and journeys. And so that asked the question, "Well, what if we built a system that represented those in the way that they show up instead of putting them into tables? And do we think we could do a 10X better job and provide significantly more value? Because there's no point in starting a startup if you don't feel you can do at least an order of magnitude better than the other way."
So that was the origin story. And then some of the things that we foresaw, which have become really important for AI, is, number one, if you have to decide ahead of time all the kinds of data that you're going to need to store, then you're going to be really limited. You want to be able to evolve really easily. In the database world, that means evolving your schema. So having schema flexibility, having a language that you can use to both describe your data, but then also query your data. And so we're a big fan of open. We see ourselves as most of our work is creating a new category of technology. That means inviting in people to compete with us. And then we get to win over people's business because we do the best, not because of lock-in.
John Furrier
>> What's the modern update as of today? Okay, four years in, what's the state of the art? What's the top conversations happening here in the room and around the world with graphs?
Philip Rathle
>> Enterprise knowledge layer is the big topic. In fact, my colleague Jesús Barrasa just published an amazing piece on this that in some ways is the spiritual successor to the GraphRAG manifesto, which I wrote two years ago. So GraphRAG describes the pattern of having an LLM call out to a knowledge graph so that you externalize your knowledge in context. It doesn't live in the model. It lives in a system of knowledge. And that gives you better accuracy, explainability, and governance. But fast-forward, what we've actually seen is that take shape is an important architectural element called the enterprise knowledge layer. And so he lays out essentially the anatomy of that layer. And that post just came out this week.
John Furrier
>> There was a research report from the University of Newcastle had results that compared GraphRAG to RAG. What are some of the findings that you're seeing? Is it more accurate? Is it faster? What are some of the updates in terms of the benchmarks? And then we'll get into how that ties to economics.
Philip Rathle
>> Yeah. So we've had lots of evidence through our customers that GraphRAG improves accuracy, provides governance, improves explainability. But it's always really nice when you can have a third party world-class academic institution do some research because then not only does that give you some objective results so you know what to expect, but you can understand the why behind it. So University of Newcastle, which is top university for computer science globally, built up multiple data sets, created over 500 complex queries, ran them with LLMs with just vector-based RAG, and then did the same with GraphRAG, and found that you get 2X improvement in queries that have an answer versus the model saying, "No, I actually don't know. I don't have enough information." 80% improvement in accuracy and just significantly better qualitative results and more precise answers as well, which are more useful. And then last but not least, fewer tokens. We're all talking about tokenmaxxing. But it turns out if I can be more surgical and precise about what data I bring back and feed to the model, not only do I get better results, but it's far cheaper than just flooding the zone with context and then churning away with a very expensive model.
John Furrier
>> So price has come to factor in there. All that for aside, but you get accuracy and speed with the GraphRAG. How is it translating into economics? I mean, you're a CTO, you have the technical piece. I was just commenting yesterday at the AMD Advancing AI event that the winners in that game of Nvidia and AMD and all the Intels is that whoever can redesign the computing architecture to get there first to enable AI to align with how humans think and work. And that's really the new user experience, how do people think and work? That's the brain. That's why I love graphs. But economics became the big thing. I want to see benchmarks around not just technology, but economics. What are some of the value pieces you're seeing come out of this?
Philip Rathle
>> Yeah. There are two levels of this. One is the system has to make it to production before you can even get any economic benefit. And then number two, once you've done that, then it's a question of refinements. One of the other speakers earlier this morning was kind of quipping about how there's the saying that no AI systems are making in a production, which is obviously an exaggeration, but many still don't. And he said, "Well, the difference between the ones that do and the ones that don't is the ones that do actually have a graph, this part of the picture." And I think some of this can be. Jensen has the five layer cake and that describes the business from his perspective and what it takes to run models. But when you're building AI systems, it's not just that stack. You have the whole data layer.
John Furrier
>> It's the NVIDIA view.
Philip Rathle
>> Yeah, it doesn't-
John Furrier
>> There's no data in there.
Philip Rathle
>> There's no data in there. There's no data infrastructure in there. So what you actually need to make it work is to externalize your data and have a system to do that. So having a knowledge graph actually is often the difference between no value and a lot of cost and value just making it into production. But then what this report by the University of Newcastle says is, "Well, assuming that I can get good enough into production with just vector-based RAG, then there's significant refinement, 2X improvements by going from just vector to GraphRAG."
John Furrier
>> Okay. So take me through a day in the life. You know I love to kind of walk down the checklist because I always think maybe I could use that for theCUBE data. But people generally want to see the engagement path. How do I get started? Let's just say I've got a bunch of vector embeds laying around, doing some RAG on it, and I want to go to a graph. What happens? Take me through the setup, the build out, standing up knowledge graphs. What does it take?
Philip Rathle
>> So to answer this, let me actually talk at a high level about this anatomy of the enterprise knowledge layer, which Jesús laid out. You have three kinds of graphs that you can have at different layers. There's everything that's in the ontology bucket, which is your business meeting mapped to your technical meeting where we use the term semantic layer or semantic map for that. So that's one bucket. Another bucket is the data. So you have your reference data, your own internal data that might be graph shaped. And then the third one is memory. So by this it's like long-term memory. So back to the brain analogy, neurons, synapses like nodes and relationships. Ontological memory, memory of your procedures and your rules, which goes back to context graph and decision traces. So first, I want to decide what do I want to start with? You don't have to boil the ocean, you can just start with one. And then depending on what I want to start with, all right, where is the data right now? Is it in a relational database? Then I need to map it in from there. If is it unstructured, then I map it in from unstructured. And ideally I have an existing ontology that I can use to convert that. And the good news when it comes to building these is this used to be a significant point of friction, but now AI comes to the rescue and we've been able to supercharge a bunch of our tooling with AI so that an LLM can take a relational database schema, click a button, and in under two minutes, you end up with a graph schema compressing what used to be a week of work down into two minutes and then you're off to the races.
John Furrier
>> What's the coolest thing you're seeing here being discussed that you had to kind of point at kind of the concepts of what people are showing as people get more graph reps in, building out, getting into production. What's the big thing that jumps out at you? What are you most excited about?
Philip Rathle
>> It's the diversity, the horizontal diversity of the kinds of problems that are solved. I just get so much joy out of having been doing this for close to 15 years and seeing people solve problems that I've never imagined even existing.
John Furrier
>> When I saw the Hadoop phase of data storage and large scale big data, the first wave of big data, people were so excited. And then there was like, "Oh, this is so hard." And then you start to see data lakes. And then there's a general enthusiasm in the data world of data people who love data like us. It wasn't like a total standing ovation, but it was still euphoric. "Ah, yeah. Data lakes are cool. I can do pipelining." Still a lot of stuff to rigid. But with graphs, I've noticed that people who are working with graphs are seriously pumped up. They have such a high from playing with them and working with them that whenever I do an interview with someone who's working with graphs, it's almost like relief meets the future. Explain the phenomenon because it's a common thread. I know you work with a lot of folks, you see the same thing. There's general excitement, but it's not just, "I'm excited." They're super pumps. They're sharing and they're telling all their friends. They're telling everybody. Why is that the case? Is it just so refreshing? Is it the scales, the flexibility? Why are dating nerds and geeks getting super pumps on graphs?
Philip Rathle
>> It's because if I'm trying to solve an AI problem that's engaging with the world, and it's in nearly 100% of cases, it's this world that shows up as a complex interconnected system. And as soon as you're dealing with it in tables, it's like having some homework.
John Furrier
>> It is. It's like doing homework that you don't have to do.
Philip Rathle
>> It becomes drudgery because you're going from a 3D color system into 2D black and white and trying to re-extract it and do all the color touch up and re-extrapolate it back into 3D. And with the graph, you retain that. So it's so much more empowering. And I love your... Yeah, it's a combination of relief and this is the future and I'm untethered and I can move forward. I'll just say this year I've gone out of my way a little bit to give back in academia. I've done some guest lecturing at U Chicago and Stanford and other places. And because for many reasons, including the fact that there's an ISO standard behind this, professors are now... It's an obligatory part of their database curriculum. And students come to me and say, "Look, I learned Postgres. I learned my SQL. I learned Neo4j. Man, graphs are like, it's so much easier." Now, that's not the reflex of someone that spent their entire career thinking in tables because you forgot how to see color. You forgot how to deal in 3D.
John Furrier
>> grinding. It's like the bark of the tree. You don't see the whole forest. I think that's true. One of the things I will point out, I'd like to get your reaction to, is that also there's something about achieving something. When people use graphs, they get different outcomes. And it's almost like found money. It's like, "Oh wow, we discovered this new way to do things." And that seems to be consistent. Graphs bring you into a whole nother realm of, "Wow, I can actually do new things. My capabilities are more headroom for things to do." And so they're knocking down business goals. Folks I've talked to are like, "Hey, I just keep the expectations low. I just deliver the hardest problems."
Philip Rathle
>> I'll give you an example. Just a couple weeks ago, we were doing a POC with a, I can't say which country, but let's say a national tax agency. 48 hours from the start of a POC, they identified more than $100 million in tax fraud. That's just in 48 hours with a subset of data. And this phenomenon, if you generalize it, it's if you've been walking around just kind of limited with your blinders because you're looking at this data in 2D, you bring the data into a graph view and all of a sudden all these things become blindingly obvious, which before you just simply couldn't see.
John Furrier
>> Yeah. And it was hidden. You had to query it out and extract it out. It's like pulling oil out, fracking oil. It's hard. And it's not always the best. Final thoughts. As you look at the second half of the year, I know we were talking about in Paris sovereignty. We're trying to get an event going, get a small group of leaders at the NYC here in San Francisco or in New York. I think it'd be in New York. But as the community comes together, what do you hope to see with other leaders as they come together and look at things that are going to be big problems to solve or opportunities to capture around sovereignty, AI production, value creation? What's your view and what should the conversation be?
Philip Rathle
>> Evolving our view of what the AI stack is and recognizing the need for an enterprise knowledge layer now that we have a clear definition of it. But then that actually gives organizations their own sovereignty. There's been a lot of talk in the news lately about, "Well, hey, does some AI LLM foundation model provider have all my data?" If you keep your data separate and in a separate layer, then it becomes sovereign to you. And then nation states, sovereignty is a big topic. And we had this whole great conversation with Amit in Paris a couple of weeks ago at Raise. This is a really big topic on the world stage where the definition of sovereignty entrains the definition of what does my AI stack look like. And I'll add one more thing. Our CEO, Emil, just revealed recently that our net new revenue last quarter was greater than our net new revenue all of last year. There's just so much growth in this space and we're just one vendor and it's one microcosm and a lot of the growth is community-based.
John Furrier
>> Yeah. Yeah. I really think it's going to be a great tailwind for you guys and the whole industry because as these computer architecture from AMD and Nvidia and everyone else gets better, they're redesigning compute architectures to the way people think and act. And AI brings agency to users and they can be empowered. The interface is no longer gooeys or windows, images. It's natural language. And so that's going to give the empowerment to the user. So whoever can enable that as fast as possible in the way they think, which is a brain, which is basically neurons and graphs, our brain is a graph, I think that's the future. And
Philip Rathle
>> In a way that's understandable. You can look at a graph and understand it and then a machine can execute on it. It becomes a shared language between human and machine.
John Furrier
>> It's like TVs, 4K is great, but 8K is coming around the corner. So we'll see. Phillip, thanks for coming on theCUBE. Appreciate the conversation.
Philip Rathle
>> Thanks, John. Always a fun time.
John Furrier
>> All right. I'm John Furrier, here in San Francisco for Neo4j Graph Talk San Francisco, of course. New York City, the next stop will be in September. We'll be right back.
John Furrier
