NYSE Media Week CXO Series, theCUBE + NYSE Wired | Philip Rathle, Neo4j

Clips
More from NYSE Media Week CXO Series, theCUBE + NYSE Wired

Philip Rathle

CTO

Neo4j

play_circle_outline Unlocking Data Connections with Neo4j: Graph Databases Aiding AI and Introducing GQL Standard

play_circle_outline Graphs as a key factor in accelerating the need for connected data

play_circle_outline S-curves of use cases and the impact of adding graphs to existing systems

Info
Transcript

Philip Rathle, Neo4j

Philip Rathle

CTO Neo4j

The Cubes CXO series features an interview with Philip Rathle, the Chief Technology Officer at Neo4j. Neo4j helps companies understand connections in their data in industries like finance, retail, and aviation. Using a graph data store, Neo4j allows deep analysis of relationships for predictive behavior. The AI trend has increased demand for graph databases and knowledge graphs. Neo4j's technology is used by companies like Klarna for data consolidation.

explore Keep Exploring

What is Neo4j and how are they making graph databases more accessible with the new Graph Query Language (GQL)? add

What pattern describes the relationship between graphs, software, and AI in terms of their impact on the world? add

What are some examples of ways companies can disrupt their own products and improve competitiveness by incorporating Neo4j and graph technology into their existing systems? add

bolt Powered by CUBE AI

Philip Rathle, Neo4j

search

Dave Vellante

>> Hi everybody, welcome to The Cubes CXO series in collaboration with NYSC Wired. My name is Dave Vellante and really excited to have Philip Rathle here. He's the Chief Technology Officer and former Chief Product Officer at Neo4j. Really interesting native graph data store that is riding the wave of AI. Philip, thanks for coming into the studio here overlooking the trading floor.

Philip Rathle

>> It's a pleasure to be here. - And great to have you.

Dave Vellante

>> This is where it all happens.

Dave Vellante

>> Amazing, isn't it? Oracle rang the bell today.

Philip Rathle

>> Interesting. Of course, you know a little bit about database, but let's start with your role and a little bit more on Neo4j.

Philip Rathle

>> Sure. So Neo4j is a company that really helps a lot of the largest companies in the world and lots of startups, companies of all size, help understand connections in their data. So the world is getting more and more connected, and it's not enough to just be able to manage your data. You need to do that store and retrieve and do basic analysis. But if you really want to get to the bottom of what's going on and get better predictive behavior to get to the next level and competitiveness, then understanding your relationships and being able to mind those more deeply makes a pretty huge difference. I joined Neo4j in 2012. Company was 30 people. Now we're 800 plus, growing, and used by many of the world's largest companies. So top 20 out of 20 U.S. banks, including JPMC, Citi, UBS, and more. Top retailers, three of the top aircraft manufacturers including Lockheed. It's pretty quite a broad and a fun space.

Dave Vellante

>> I've always been super excited about graph databases, knowledge graphs, the expressiveness that you get from such a system is awesome, frankly. But it's always been hard to query, get the query flexibility of, say, a sequel. Right?

>> Yeah. - So you have that sort of trade off.

Dave Vellante

>> So that's why it's been really a home run in a lot

Philip Rathle

>> of more narrow use cases like cybersecurity as an example. There's obviously a great use case and application for it, and that's beginning to change. There's an AI tailwind which almost necessitates that you have these connections and can visualize them. Can you explain a little bit more on the technology and how you're making it more accessible?

Philip Rathle

>> Yeah, so let's start with the model, because the model, as you point out, is really natural and easy to understand in the same way that a business person will whiteboard their domain like circles and lines, and Philip owns this car, which has this VIN and is insured by this insurance company, you can just swivel the chair and put that into the graph. As far as querying, I have some good news for you, David. So this year, ISO, for the first time in 37 years released a new query database language standard, and it is now a sibling to SQL. It's called GQL, and it's called Graph Query language. It entirely democratizes graphs. The way that language has been designed is it's very similar to SQL in all the ways that matters so easy for people to learn. Also, very similar to Cypher, which is the de facto graph query language that Neo4j opened up nearly a decade ago into 2015.

Dave Vellante

>> So this was super news for the industry because rather than having a series of proprietary query languages, you now have an industry standard. So you get the flexibility of something as simple as SQL that everybody can understand, and you maintain that expressiveness of the graph. Now, let's talk about why this is so important in the AI era. Because everybody always says, "Philip, you can't have good AI without good data."

Philip Rathle

>> Yeah.

Dave Vellante

>> And it's hard to get good data. The example that I always use is if you want to have a consistent data, all the data is locked inside of applications and the data, the metadata, the business logic, and it's sort of trapped, if you will. I've always seen graph as at least a path to exposing that, harmonizing that data. And again, the example I use is revenue. What is revenue? Could be NRR, could be ARR, could be bookings, could be calendar year, could be fiscal year, and people spend more time arguing about that. Am I on the right track in terms of the tailwind that AI has provided for your business? And I'm really interested in how you're applying it to solve problems.

Philip Rathle

>> Yeah, I guess there are multiple tailwinds. So one is on the data preparation side, and then one is on the query interacting with AI side, and there's a technique called GraphRAG, which is retrieval augmented generation using a knowledge graph that has really become a pretty hot trend right now inside of AI. But those two things go together. So I was just re-listening yesterday to Sequoia's training data podcast, and there's an episode where the CTO of Klarna talks about their internal application and how it's based on Neo4j. And the reason he talks about the value he talks about that the knowledge graph brings is it's not just a way to consolidate one's data assets to normalize them, to get them correct, but also a way to provide collaboration around people as people use the data. Obviously you have the same data in 50 different places if you're a large company, and who knows which one is most up to date and correct. And so providing a means of exposing that, cleaning it up, curating it, but then using GraphRag as well on top of it. And what we're finding with GraphRag is it also bridges an important gap between human design, what we as humans want the machines to do and what the machines are doing, which otherwise would be entirely a black box. You get benefits which we can drill down into why of increased accuracy, better answers, higher fidelity, explainability, and then governance and explainability on the back side. But you also get these, I think zooming out, this provides a good way to bridge what's happening in the black box machine realm and what we humans need in terms of preserving our autonomy of decision-making, as well as corralling the AI to do what we want it to do.

Dave Vellante

>> So the Klarna example is really interesting because essentially they put forth in this podcast that, the way I would describe it is they're blowing away their whole SaaS model, and they're basically replacing-

>> Which I don't necessarily

Philip Rathle

>> recommend people go out and do that. There's a good reason for having Packard software.

Dave Vellante

>> And I'm not sure they're going to... It's not like it's going to happen overnight, but they are leader in thinking about how you can apply AI because the SaaS model has been around for a long time and there's some good parts of it, and there's some not so good parts of it, particularly people, there's a backlash on SaaS pricing and the model's sort of fossilized. And so it's interesting to hear them talk about how they can use the platforms like Neo4j and AI and GraphRAG to rethink how they approach workflows. And so this is what I wanted to get to because I think RAG is really interesting. GraphRAG is even more interesting, but it's still early days. And now we're talking about agents, this whole notion of agentic. And to the extent that you have harmonized data and it's governed and it's secure, going beyond a single co-pilot or a single agent starts to get really interesting where you can imagine multiple agents working in concert. You've got harmonized data, you've got maybe an agent control framework, which probably is not your swim lane, but maybe you touch upon it. And now you're sort of doing things that you necessarily can't do with microservices, which are largely hard-coded and somewhat fragile.

Philip Rathle

>> Yeah. - So this is a new world where entering,

Dave Vellante

>> I don't know what inning we're in, but it's not the ninth.

Philip Rathle

>> Yeah. - It's very early.

Philip Rathle

>> Yeah. - So how do you see this whole thing playing out

Dave Vellante

>> and what's Neo4j's role?

Philip Rathle

>> Yeah, so I guess going back to where you started around

Dave Vellante

>> SaaS build versus buy, actually, we have a lot of customers using Neo4j inside of their SaaS applications. And ideally, this is how, certainly the longer tail of mid- market and smaller companies, most of their software is off the shelf. Most of our software is off the shelf. We build our own systems only when we absolutely have to, other than our core product, of course.

Dave Vellante

>> Right. - But then large enterprises oftentimes have very

Philip Rathle

>> bespoke requirements, which is why they build their own, where Neo4j is also used. So we love selling to both and supporting both. And ultimately it's each company's decision for what they're trying to do. Do they build? Do they buy? We're very happy with both. On the agent side, yeah, this is the next big thing for sure. And you're right, agentic workflows themselves are forms of graphs. And you can see the beginnings of this with LangChain and LangGraph where you have basic orchestration frameworks. We have been having some conversations, again, we're not here yet, but this is maybe an inning or two ahead about the need for control graphs in more complex applications where you have many agents and you need the application to... Then the framework itself to be able to orchestrate between what the agents are doing and saying. But the thing that's happening now is inside of many of these agents, you have small models, large models, vector-based RAG, also GraphRAG. Some might be a call just to an LLM. Some of the workflows we've seen where LLMs are generating Cypher in order to be able to engage with a graph based on a human language question, then using another LLM to do a two or three shot. Okay, let's see if an error comes back. And if it does, let's validate that. And so this idea of ping ponging between different models and different agents is really compelling.

Dave Vellante

>> Yeah, it sure is. And by the way, I'm not by any means predicting the depth of SaaS. I actually think the reverse. I think the existing SaaS companies have a huge opportunity.

Philip Rathle

>> I agree. Yeah. - Because they own the business logic,

Dave Vellante

>> they have the data, they have the metadata, and to the extent that they can connect all that together, which is non-trivial, I do think there's two vectors, and I'd be interested in your opinion on this. I think there's the existing SaaS platforms, the great SaaS platform, ServiceNow, Salesforce, Workday, Oracle, et cetera, SAP, they're not going anywhere, and they've got great opportunities within their own domains. I think there's other opportunities, there's really three vectors, other opportunities to connect to those backend systems, use products like platforms like Neo4j to harmonize that data and then really empower those agents. And I think there's a third vector, which is startups that say, "Hey, we're going to completely reimagine the software stack."

Philip Rathle

>> Yeah.

Dave Vellante

>> And I think all three are viable. I think all three, it's not a winner take all. I think the market's big enough to support those three. So my question is how do you support each of those? I think you're in a good position to do so.

Philip Rathle

>> Yeah, and part of what you're describing, pattern number two, where a SaaS company ties into an existing company with data, which is part of what Klarna is doing, actually, it brings them tighter in with the ecosystem of their different SaaS vendors, that it's another way in which the world is becoming more connected, and there are more touch points between things which actually accelerates the need in the case for graph. So I've seen a pattern, a single pattern that actually describes those three, which is, so I think of it as graphs are eating the world, much like the well- known recent software is eating the world, which continues to be true, right? More in a SaaS form these days. AI is now eating the world. But the way in which graphs are eating the world is this, is for any data-driven human activity, so let's call that a use case, you can only get so good with just surface data about the interconnectivity of the system of a given system or how it connects to other systems. So systems of biology is how they relate to systems of people and buyers and doctors and providers and so on and so forth. Or systems of payments in relation to products, banking products that customers own and FinTech and so on and so forth. So there's like an S-curve. Everything moves in S-curves when it looks-

>> Life's, an S-curve.

Philip Rathle

>> Right. Life's an S-curve.

Dave Vellante

>> So most use cases are at some S-curve where they're not using connected data and so they're topping out. So time and again, my journey at Neo4j has been seeing at least one new use case every week for over 10 years where someone's discovered some new way to use the deeper connectivity two, three levels out or more, inside their existing application or inside their existing predictive analytics or forensics to do better. So then there's another S-curve that starts. So either existing companies disrupt their own products by just adding on Neo4j along to the side, add graphs to your existing use case alongside existing systems. Then it gets better, then it gets harder for their competitors to compete. The competitors have to level up. Or you have new companies come into being initially as startups, and then they get bigger and then they become kind of fast-growing, mid- market SaaS companies, which then start pulling business away or creating their own new use cases. But this is all kind of the same graph phenomenon as far as we're concerned.

Dave Vellante

>> I want to have you back and go deeper into this because you're opening up now so many other potential areas of conversation. Large language models, small language models, data that's sort of hidden inside of corporations now that you can connect to. We like to say that large language models and small language models are going to turn to LAMs and SAMs large action models and small action models, where again, you can't hard code everything. But these agents can observe what we're doing and you guys can create, observe and create those connections, or I guess take advantage of what the agents are observing and then store those connections and then reuse them.

>> There's definitely a feedback loop,

Philip Rathle

>> and we're seeing that with customers already, where they'll take the actions at different levels, like the end action that was taken, the action that was taken by an individual agent, and then even the action taken by the given query, feed that into the system, store it, and then use that to improve things. And you can do that through GraphRag. You could also do it by using the graph to train the model. So lots of different-

Dave Vellante

>> Trainings, fine-tuning.

>> Yeah. - I really would love to have you back.

Dave Vellante

>> We're out of time, but Philip, thanks so much

Philip Rathle

>> for coming on The Cube, really. And keep it right there. We got more action high above court side, as they say, from NYSE, The Cube plus NYSE Wired. I'm Dave Vellante. We'll be right back.