This conversation explores graph-powered migrations and enterprise knowledge layers for artificial intelligence, AI-enabled transformation. Alejandro D'Andrea of BotMan-AI, founder and chief technology officer, appears in the Neo4j AI Luminaries series on theCUBE. John Furrier of theCUBE hosts the discussion.
D'Andrea explains how BotMan-AI combines AI pipelines and graph-based ontologies to ingest code, database semantics and unstructured communications to enable fully autonomous migrations from legacy stacks such as Delphi, Visual Basic and Oracle Forms into modern architectures. They describe automated entity linking, semantic layers and graph traversal that provide precise context for agents and platform-level AI systems.
John Furrier of theCUBE frames the conversation with insights from theCUBE Research and industry analysts. D'Andrea states that knowledge graphs create deterministic relationships and low-latency context retrieval that significantly reduce hallucinations and improve agent accuracy. theCUBE Research and analysts highlight faster lower-cost migrations, stronger traceability and pragmatic steps to execute migrations: prioritize ontology design, entity linking and semantic layers and start with focused pilots to demonstrate value and manage change across organizations.
Watch the full conversation to learn how knowledge graphs, ontology design and semantic layers support scalable AI infrastructure and reduce migration risk.
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
Alejandro D'Andrea, BotMan-AI | AI Luminaries with Neo4j
This conversation explores graph-powered migrations and enterprise knowledge layers for artificial intelligence, AI-enabled transformation. Alejandro D'Andrea of BotMan-AI, founder and chief technology officer, appears in the Neo4j AI Luminaries series on theCUBE. John Furrier of theCUBE hosts the discussion.
D'Andrea explains how BotMan-AI combines AI pipelines and graph-based ontologies to ingest code, database semantics and unstructured communications to enable fully autonomous migrations from legacy stacks such as Delphi, Visual Basic and Oracle Forms into modern architectures. They describe automated entity linking, semantic layers and graph traversal that provide precise context for agents and platform-level AI systems.
John Furrier of theCUBE frames the conversation with insights from theCUBE Research and industry analysts. D'Andrea states that knowledge graphs create deterministic relationships and low-latency context retrieval that significantly reduce hallucinations and improve agent accuracy. theCUBE Research and analysts highlight faster lower-cost migrations, stronger traceability and pragmatic steps to execute migrations: prioritize ontology design, entity linking and semantic layers and start with focused pilots to demonstrate value and manage change across organizations.
Watch the full conversation to learn how knowledge graphs, ontology design and semantic layers support scalable AI infrastructure and reduce migration risk.
Alejandro D'Andrea, BotMan-AI | AI Luminaries with Neo4j
Alejandro D'Andrea
Founder & CTO at BotMan-AI.comBotman AI
search
(INTRO)
John Furrier
>> Hello, I'm John Furrier, your host of theCUBE here in the Palo Alto studios of theCUBE. Of course, we have our New York Stock Exchange CUBE studios with NYSE Wired, of course, connecting Silicon Valley to Wall Street. This is our Neo4j AI Luminary Series. We highlight leaders who are building the next-generation AI software systems, taking advantage of the data, the graph databases, the role of ontologies. as intelligence comes in, you're seeing more and more activity around how data is being organized on top of existing platforms. Alejandro D'Andrea here, founder and CTO of BotMan AI, BotMan-AI.com is the URL, but love the name. Alejandro, thanks for coming on theCUBE all the way from Chile. Thanks for taking the time.
Alejandro D'Andrea
>> Great to be here, John. Thank you for the space.
John Furrier
>> Well, we're big fans of what's going on because one, we see the technology shift. we see the AI commercialization, it's happening really in two theaters, on the tech side and also the commercialization side. We have, of course, we're covering AI infrastructure, faster tokens out, more revenue, right? So that is grounded in the data. the role of graph databases is coming in to be very mainstream because at a platform level, it provides a really good utility to bring things together. It's aligned with the technology of AI systems. Everything's kind of clicking. So before we get into that, just explain what you do, because I want to set the context because you're having success with this. So explain what BotMan AI is doing, some of the things you're trying to solve and where you're at.
Alejandro D'Andrea
>> Yeah. Yeah, so we started a couple, maybe two years ago, a little more than that. I came from the corporate side. I was a CIO for some companies here in Latin America, also for Walmart International, I was CIO for Chile. So I struggled with legacy systems all my life. So one of the things that came naturally for us was trying to deal with that. And we started two years ago migrating legacy systems with an AI -based pipeline plus graphs. And I'll go into that later on. But we successfully migrated tens of legacy systems by now. And at this point, we have a very mature platform that we're using to migrate legacy systems. So that's the key of what we're doing. And so far, I think that the platform we have is pretty unique at the global level because I've been having conversations with folks in the U.S. and all around the world. And so far, I know at some point someone will say, yeah, we have the same thing, but it hasn't happened yet. So I hope I bring something new to the table.
John Furrier
>> So if I get this right, you basically have a platform that helps bring AI to legacy systems or migrate them. What specifically are you working on?
Alejandro D'Andrea
>> It's basically taking, migrating from an old code base, maybe Delphi, Visual Basic, Oracle Forms, and transform, rewrite that same system in a new architecture. Let's say React and Spring Boot, for example. So it's re -architecting the entire thing. A lot of people are doing that with agents. So a lot of people are using human teams that are accelerated by agents. In our case, it's different. We ingest the whole thing at once and we migrate it fully autonomously. And that's where graphs play a big part.
John Furrier
>> I brought up Palantir as kind of the intro because they're in the news a lot and this founder is very dynamic and controversial. But they do have a platform. They actually talk about ontologies. And so the knowledge layer for them is super valuable the way they've approached that business. That's becoming, I won't say a North Star, but a template for companies. And when enterprises try to do that kind of coolness, they bump into brittle legacy systems. And it's not always easy to just throw it away and bring in the new. You got to do the migration, you have to rethink it as a system. And if you look at the AI infrastructure and the software stack that's emerging, it's a systems problem. It's a systems opportunity to get that system right. This is where the knowledge layer and knowledge graphs and context play a very big role. What are your thoughts on that? How do you explain that and how does that render in practice?
Alejandro D'Andrea
>> Yeah. So for me, when we started two years ago, using graphs was very natural. It was a natural way of thinking how to deal with the limited context of AI. At that point, the context window was about 32K. Now we're talking about a million tokens, right? But 32K was... So you had to deal with how do I give the exact amount of context to the AI so that it can perform a particular task. In our case, it was a piece of a migration. It was like a layer of one particular API. So it was very limited tasks, but for that you wanted to give it the exact context. And when you need to ingest a whole legacy system, you can think about it. You mentioned the ontology. So for a legacy system an ontology might be modules and classes and routines. And then the database, you have packages and stored procedures and things like that. So if you come up with an ontology that represents that, and then you map all the relationships, all the dependencies between things. So if you have a routine that uses a library or writes on a table, you need all of that. you need, if it was a human doing that, they would need to understand the entire picture, right? So this routine, where is it used? What is it using? What is it writing? What is it reading from? So if you come up with the right context, it's probably enough for the particular task. So ontology for me have to do with how do you represent reality? How do you represent relationships between things? And then how can you extract in a hopefully deterministic way exactly what you need for each task, right? So it was natural for us and then we evolved and then from code we ingested semantics about the database and now we are ingesting the migration tasks and the issues. And once you start thinking about this, you realize that the more things you have linked, the more things you can do.
John Furrier
>> You know, the SiliconANGLE and theCUBE Research, we've been looking at this, we're seeing that graphs are low latency on context retrieval in terms of GenAI. We're seeing that the capability of how you can recurse through the graph is actually a great thing. There's other studies that show that if you ground the data with knowledge graphs, you reduce hallucinations by 44, 45%. This brings a platform discussion to the table because the enterprises that's unlocking their data that's specialized to their domains isn't in the general frontier models. So AGI and general intelligence, that's the frontier model. That's ChatGPT, that's Claude, that's Gemini. But when you get into the enterprise, they got data and they have data that's sitting there and it's about to be unlocked. And they want it to be relevant to the AI technology. They want the context to be surfaced super fast and they don't want hallucinations. You talk about robotics, it's even more strict. You can't have an autonomous car be wrong, right? You got to have robots that's safe. So you're starting to see safety tied to, no hallucinations. You're starting to see the value of the AI being fast data context and low latency. Lay that out. Cause I think that is a general perspective. How do you make that happen? first of all, I'm sure you'd agree, but how do you make that happen?
Alejandro D'Andrea
>> Yeah. So I would say that most people are doing work without graphs that seems good enough at the surface level. So why? Because you might have, let's forget about code for a while. Let's say you have a bunch of documents and you want the agent to retrieve the right information for a task, right? So it might go to a document, then look around and find another one, and then maybe find another one. But no one will guarantee, there's no way to guarantee that it will retrieve everything that's relevant, right?
Alejandro D'Andrea
>> Right.
Alejandro D'Andrea
>> So agents are doing better and better work of retrieving and reading things fast. But the problem is that that discovery work is being done at the time that you assign the task to it.
Alejandro D'Andrea
>> Right.
Alejandro D'Andrea
>> It has to figure out what the information is and it will retrieve as much as possible. But models are lazy also. Right. So at some point you will just stop and say, yeah, this is all that there is. But it's not. There's a lot of information that's being left out. When you use graphs, you are doing that work up front because you are creating those relationships up front once, just one time. It's much more effective, much faster because you get, let's say, a document, a legal document. At that point, when you get it, you can create all the relationships to other documents, right? You can say, okay, this relates to this particular project and this particular loss and this particular lawsuit and whatever. And this person in the organization that handles this and this project where it happened. And you do that once. And then any agent, when you assign a task, it will know automatically from the graph that all that information also might be relevant. It might not, but at least it has a chance to visit those nodes in the graph and figure out whether it's relevant or not. So it makes a great difference when we migrate code. We find it sometimes very easy to migrate huge code bases, but we have other cases where it's very hard. We have a particular customer with each form, each UI form has like 700 associated files, it will always miss many of those if you just send the agents roaming free in the code base and see what they find. But if you have already mapped that in a graph, there's no way you will miss anything. So it's a huge difference on speed, cost, efficiency, and how good the results are.
John Furrier
>> Explain to me, let me rephrase, let's talk about the knowledge layer, because there's two conversations. There's the technical, kind of like the deep tech conversation. Then there's the human average person. So, knowledge layer on the tech side, we've heard things like it's a connective tissue between organizations, data and humans and systems, gives the context memory for reasoning that makes better decisions for agents and humans. Okay, great, check. There's a playbook for that. But when you talk to the average person, I'm sure we all have been there explaining what agents are and reasoning. It's like a brain. You got the left brain, right brain, you got the cortex. you can almost look at it like the brain of an organization. And that's what executives want on their business model. They want to have an enterprise with all their data and domain expertise. That's their competitive advantage, the moat. They just want to have a brain that can reason and do things, take actions. So explain in your words, this connective tissue piece and how does that fit in the architecture?
Alejandro D'Andrea
>> Yeah. So let me explain with something that's maybe more familiar for anyone. It's something that we are using internally for our own effectiveness in BotMan. So I have a particular brain that looks at all my interactions in the digital world. That is emails, WhatsApp text, WhatsApp audios, Discord, Teams, whatever happens on my computer at night, there's a process that will fetch all of that and build a raw layer with just my emails, just my conversations as they are, right? So that's the raw layer. It doesn't give you knowledge. It's just information. It's everything that I talked about during the day, right, on any digital channel. But then another process runs that creates that knowledge layer. What is that? Let's imagine that I have an article for a particular project, right? So it doesn't matter whether I had an interaction with the project manager via email or I had a Teams meeting where we discussed the project or someone updated the project in GitHub, so the code for the project. All of that feeds into that same article. So in one piece of information, I have everything I need to know about what we're building, how the project's going, politics, payments, whatever related to the project, I can put together in one article or maybe many articles, but together in a folder, let's imagine. So I have a folder for the project that those pieces of knowledge are the distillation of all these inputs that are scattered around. Right. So that's for me, an example of what is a knowledge layer. So then your agents will work with that layer, not read every email. And again, they will lose something because there are thousands of emails, but they can look at five files maybe and have the entire picture of where we are. Right.
John Furrier
>> So you can map the graph at the right place for the right data. Talk about the role of data because now we're seeing with graphs and we're hearing this a lot, certainly on theCUBE as well, is that there's a lot of data types that are undocumented that databases can't handle. Domain expertise, for example. Talk about that piece of it because in a lot of these automations, these workflows, there's a lot of stuff that gets done that's somewhat maybe documented, but it's not like formalized in a table or database.
Alejandro D'Andrea
>> Yeah. Yeah, so I think we are finding this difference between structured data, we could call it, which are the traditional databases that have rows with information, like an Excel file maybe. And then you have this unstructured data, which is a lot of emails and documents and conversations and all of that. And for me, graphs have the ability to even connect those kinds of things. So one of the things that we do, for example, aside from reading all the code and creating this graph, we also read the database and we have processes that will try to understand the semantics of the database. So, for example, if I have a table with customers, it will describe that these are the customers that are used in such a system for this reason. And this column in particular is the social security number and what it's for and what it's related to and all of that. So aside from having your physical database with all the rows and the tables and everything, you have in the graph a way to understand what that data is about and also to relate that with things that are not in a database. Like you said, maybe you have documents and all of that, but they are also relevant to understand a project as a whole. So you might think about pricing. You can have the semantics of the tables of the databases that contain the prices and the historic prices and all of that. And also you will have documents related to the strategy of pricing or to marketing campaigns. And so you can now have an agent navigate through all of that and come up with much better insights than just one or the other, right?
John Furrier
>> I mean, I think that's important. I wanted to bring that up because I'm hearing people connect their Slack channels, emails, again, all this unstructured data and it's nuanced. And I want to get your thoughts on this because most people think when I call a database, it's a search function. It gets an answer. When reasoning is in play, the agents are actually doing reasoning, which is recursive, it could be multi-step. That's a key piece, and that's why I think graphs are getting a lot of attention. What's your reaction to that?
Alejandro D'Andrea
>> Yeah, I think it's very important how you organize information, because you can have a huge graph. We have graphs with millions of nodes, literally, and it's impossible for an agent to find the right information if you don't have a good ontology, a good way for the agent to understand where the information is. One of the things that worked for us in terms of processing all this unstructured information, like I said, emails and files and everything, is link them through entities. So an entity for us will be a person, a project, or a product, right? Mostly those. So if I have a conversation on WhatsApp and I mention you, for example, the process that gets that into the graph will first figure out who is John, right? So it will look you up, it will see all the connections that you already have, the projects that you're working on, the interactions that we had and everything. And then it will figure out where that piece of information, that WhatsApp audio or text should fit. If it should, maybe it's garbage, but maybe it's an important update on one of the projects that are linked to you. So having entities in this case you, maybe the project is another entity that's related to other things. So for us, that's the way to make sure that you get to the right point in the graph. So we first figure out, OK, what entities are involved in this piece of information, in this document or email or whatever? And then how do we traverse the graph in order to figure out where that information fits? The same with retrieval, right? So if I ask you about a project or whatever, it will first identify the entities related to my question, in this case, you, and then it will go to you in the graph and then figure out what things you're involved with.
John Furrier
>> I mean, this is really the understanding piece, because without that context, memory, and reasoning, the outputs can't be trusted or explained, or even governed. and you have an answer, but it might be incomplete, creates hallucinations, drift, there's a lot of consequences. So I think the context piece is huge. And I want to ask you, because I think this is the secret sauce in my opinion, the enterprises today have an old school mentality, a lot of legacy, a lot of practices. This is the opportunity for say Neo4j and others in this area, because if you can nail the context with the things you just mentioned, you get more accurate information, you understand relationships, you got the ability to reason, and then you could track the lineage and the traceability of things. You could say, hey, he said that on email. He said that. That's the connection. So there's a lot of things going on that could be very valuable. Why is it stuck in pilots? So why are the enterprises not all over this? Or are they? What's your take?
Alejandro D'Andrea
>> I think it's a maturity thing. They are just getting started with understanding what agents can do for them at the individual level mostly, at the productivity level. But even then, people are seldom already thinking about all these connections that I told you about. So it's very natural for most people at this point to drag and drop a file into a GPT or Claude and ask about the file. But thinking that Claude could also have a knowledge base like a graph that has a relationship between that file and the project and the people and the status and the database and the product, all of that doesn't even cross their minds. Because it's a new way of thinking that we are developing. We probably started before and now we are thinking more ahead. But most people are just in the first stages. When I think of the future, when you see the news, OpenAI just released those 10 mathematical problems that were there for decades and now have been solved with $2 ,000. And you get more and more news like that. So intelligence is not anymore the problem. It's not a problem of intelligence. The problem is context. You have to explain to the agent everything he needs to understand to do a particular task. What if you didn't have to explain that? What if it already knew all that you know? So that's the key now. So if an organization comes to a point where they have a knowledge graph that has everything that ever was ever spoken in a meeting or sent in an email, of course, if it's not private. So that's another concern that you have to solve for, but it's solvable. Once you have that brain that has all the information linked, you can ask of an agent, probably just about anything, and it will do it. Some companies are already realizing this, but many are not even thinking about this. I also consult for some companies, and they're not there yet, but it's something that obviously will be the next step for boosting productivity.
John Furrier
>> I have a series that I'm doing out of our New York Stock Exchange studio called Transformation Edge. It's about competitive edge in the modern era, basically competitive strategy for AI. Everyone that I talked to at the senior leadership level would all agree that knowledge is their competitive advantage. Domain expertise, how we do business, how we organize our culture, et cetera, et cetera. The systems that have been deployed. There's almost no debate on that. When you say knowledge layer, you get into the data platform conversation and all organizations have been going down that road. You see Databricks, Snowflake, Microsoft Fabric. These are examples of IT systems. So how do we connect that? How do you connect that? In San Francisco at the Neo4j GraphTalk, we were talking about the Virtual Graph technology You can reason over existing data platforms So there's a solution in there. Explain the thought process to connect that IT fabric, whether it's modern graphs with legacy systems, because the end game is right in sight for the technologists because there's a hundred percent consensus almost that, our knowledge is our competitive advantage in the AI future. So you got the data in place. In some cases they're moving faster on data lakes, but now you got lake -based transactional and operational systems and analytics systems. Talk about that piece. How do the technologists connect?
Alejandro D'Andrea
>> Yeah. Yeah. So we have another product that creates this semantic layer. And in this one, we are not the only ones globally, like in the migration side. But there are products like ours that will connect to a database, try to figure out the semantics, what the data are, how they are related. And that's one bit of it. For some of our customers, we also use the source code of the applications that sit on top of the database to understand the database better. Because in the source code, there's all the information about what the table is, how each field is calculated, where it comes from. So database plus source code gives you a lot of information about your databases. With that information, of course, you can use agents then to query in natural language or to ask for analysis and things like that. And that is one big piece of the organizational knowledge that you can deliver or draw value from. But that's only one. Because when you think about business processes, for example, if you want to replace part of a business process, it's not enough, right? The database information by itself is not enough because a business process will traverse some workflow on that system, but then something goes through an email and then something goes through an Excel file and then you send someone something else. And that is the process itself, right? So that knowledge is not captured there. And you also need that knowledge, right? To understand the day -to -day activities of the people and the processes and everything, that's another one. How you evolve your technological systems draws from that. So if you could understand what people are doing on a daily basis, you would have AI also suggest improvements to the systems because people are doing this manually, they could be doing it automatically. So it's many dimensions of knowledge and the structured database is one of them. But if you keep thinking about it, it's like, okay but if we add this and then if we add this and that's where a lot of knowledge graphs come in because you can represent almost anything there is there's no limit, there's no structure, right?
John Furrier
>> Yeah, and you have millions of nodes. All right, this is obviously the data infrastructure is part of the AI infrastructure. It's not just GPUs as well as GPUs and memory scarce right now, we're seeing that build out go on. So that's a check. The data side is an infrastructure play, but now you have this new wave coming in where it's like, okay, we're seeing more dynamic environments, workflows that are kind of defined deterministically, probabilistically, you have personalization, autonomous systems. This is the framing that we're seeing. So the final question for you is, how do you pitch that to senior people? What would be your pitch to say, look, I think we can solve this. How would you boil that down into business speak to get that approval? Because the people who are doing graphs are rising right to the top because they're getting results.
Alejandro D'Andrea
>> Yeah, I would say I have a lot of ideas about how to draw more and more value from distilling this knowledge. So far, we have been focusing more on the migration side because it's something that's very tangible for people. It's something that they have been trying to do for decades and they didn't know how to, and now there's a possibility. But if you ask me, I would love to go deeper in what I said before. Let's have the data layer, let's have the processes layer, the technology evolution layer. I am starting to have some conversations with some CEOs about that, the big picture that comes ahead. But I would say I think most are not ready for that conversation yet they still have to realize the value from these first steps like the migration like the agents or like automation of certain processes but it will come naturally if you find someone that has a more forward uh visionand you can paint this picture of okay how why for example but why is your organization structured in silos right now? Why do you have like a treasury department and an HR department? Because no one can have the entire context of how everything works, right? So you need to divide things in simpler things to understand because we're human. But what if all that knowledge was centralized and one agent could understand the end-to-end of how to hire someone until they pay that person, right? It's absolutely doable. but for that you have to break your notion of how an entire company works. And you've been working with that for 30 years now. So it's hard.
John Furrier
>> Yeah, for the change management, Alejandro, it's so hard because there's so much impact. It's interesting, I hear a lot, I'll just share something I've heard with you and other folks watching. It's how pilots get done is one easy, when you mentioned get some wins, migration, whatever the use case is, get a win and see the results fast. But what's interesting, people are doing pilots differently. In the old days it was, okay, let's take the system, put it on the edge of the environment, test it, no blast radius really, we're going to do a pilot. And a lot of people get their fingers in it, slows down. It maybe works, maybe not. Most of the time it doesn't. And it takes months to get stuff going. Today with agents and with knowledge layers, they have the data in a data lake, they have the base platform. They're taking on the hardest business problems that they never would do before. They're all going after the revenue. Well, we've been trying to do this, but it would have taken a huge supercomputer. Well, they got that now. Okay, you got a super computer, you got intelligence. So what's interesting is smart technologists are going after the hardest problem and they get a win. And then the result is something material, revenue, cost. I migrated that whole system in a week, right? As an example, I'm sure it's different timetables, But we're hearing, years to months, months to weeks to days. So they're starting to see the acceleration of real value. So I think that, to me, because it's too hard to go in and wave your hands and say, hey, we're going to be a modern AI -enabled company. It's really difficult. But you can get the wins.Would you agree?
Alejandro D'Andrea
>> Yes, in every transformation. Where we work with agility, you would have to do pilots, and then with data also. So, it's like you said, we are humans. Change is hard for us. People are scared. The incentives even are most times not aligned with changing. They are more aligned with operating efficiently what you already have. So, yeah, yeah, it's the same recipe. I used to do innovation and agility and digital transformation. That was what we called it. It was the same recipe. You have to start small. You have to find people that are enthusiastic. You have to show results to everyone. You need a CEO that asks for this and spend some time asking for it and then enabling people. It's the same recipe. Even if we achieve AGI, I think we are already there probably, but it will take some years for us to adapt.
John Furrier
>> Now it's business transformation because the money's on the table. I've seen value with revenue. Alejandro, thank you for this conversation. Great insight, expertise, great to have you on the Neo4j AI Luminaries series. Thanks for coming in all the way from Chile.
Alejandro D'Andrea
>> Thank you, John. It was a pleasure. I'm a big fan of the work you're doing.
John Furrier
>> We appreciate that. from all over the world, from Latin America, North America, EMEA and Asia Pacific, it's a global phenomenon. It's a distributed computing systems opportunity and we're covering it here on theCUBE. I'm John Furrier, thanks for watching.
Alejandro D'Andrea, BotMan-AI | AI Luminaries with Neo4j
search
(INTRO)
John Furrier
>> Hello, I'm John Furrier, your host of theCUBE here in the Palo Alto studios of theCUBE. Of course, we have our New York Stock Exchange CUBE studios with NYSE Wired, of course, connecting Silicon Valley to Wall Street. This is our Neo4j AI Luminary Series. We highlight leaders who are building the next-generation AI software systems, taking advantage of the data, the graph databases, the role of ontologies. as intelligence comes in, you're seeing more and more activity around how data is being organized on top of existing platforms. Alejandro D'Andrea here, founder and CTO of BotMan AI, BotMan-AI.com is the URL, but love the name. Alejandro, thanks for coming on theCUBE all the way from Chile. Thanks for taking the time.
Alejandro D'Andrea
>> Great to be here, John. Thank you for the space.
John Furrier
>> Well, we're big fans of what's going on because one, we see the technology shift. we see the AI commercialization, it's happening really in two theaters, on the tech side and also the commercialization side. We have, of course, we're covering AI infrastructure, faster tokens out, more revenue, right? So that is grounded in the data. the role of graph databases is coming in to be very mainstream because at a platform level, it provides a really good utility to bring things together. It's aligned with the technology of AI systems. Everything's kind of clicking. So before we get into that, just explain what you do, because I want to set the context because you're having success with this. So explain what BotMan AI is doing, some of the things you're trying to solve and where you're at.
Alejandro D'Andrea
>> Yeah. Yeah, so we started a couple, maybe two years ago, a little more than that. I came from the corporate side. I was a CIO for some companies here in Latin America, also for Walmart International, I was CIO for Chile. So I struggled with legacy systems all my life. So one of the things that came naturally for us was trying to deal with that. And we started two years ago migrating legacy systems with an AI -based pipeline plus graphs. And I'll go into that later on. But we successfully migrated tens of legacy systems by now. And at this point, we have a very mature platform that we're using to migrate legacy systems. So that's the key of what we're doing. And so far, I think that the platform we have is pretty unique at the global level because I've been having conversations with folks in the U.S. and all around the world. And so far, I know at some point someone will say, yeah, we have the same thing, but it hasn't happened yet. So I hope I bring something new to the table.
John Furrier
>> So if I get this right, you basically have a platform that helps bring AI to legacy systems or migrate them. What specifically are you working on?
Alejandro D'Andrea
>> It's basically taking, migrating from an old code base, maybe Delphi, Visual Basic, Oracle Forms, and transform, rewrite that same system in a new architecture. Let's say React and Spring Boot, for example. So it's re -architecting the entire thing. A lot of people are doing that with agents. So a lot of people are using human teams that are accelerated by agents. In our case, it's different. We ingest the whole thing at once and we migrate it fully autonomously. And that's where graphs play a big part.
John Furrier
>> I brought up Palantir as kind of the intro because they're in the news a lot and this founder is very dynamic and controversial. But they do have a platform. They actually talk about ontologies. And so the knowledge layer for them is super valuable the way they've approached that business. That's becoming, I won't say a North Star, but a template for companies. And when enterprises try to do that kind of coolness, they bump into brittle legacy systems. And it's not always easy to just throw it away and bring in the new. You got to do the migration, you have to rethink it as a system. And if you look at the AI infrastructure and the software stack that's emerging, it's a systems problem. It's a systems opportunity to get that system right. This is where the knowledge layer and knowledge graphs and context play a very big role. What are your thoughts on that? How do you explain that and how does that render in practice?
Alejandro D'Andrea
>> Yeah. So for me, when we started two years ago, using graphs was very natural. It was a natural way of thinking how to deal with the limited context of AI. At that point, the context window was about 32K. Now we're talking about a million tokens, right? But 32K was... So you had to deal with how do I give the exact amount of context to the AI so that it can perform a particular task. In our case, it was a piece of a migration. It was like a layer of one particular API. So it was very limited tasks, but for that you wanted to give it the exact context. And when you need to ingest a whole legacy system, you can think about it. You mentioned the ontology. So for a legacy system an ontology might be modules and classes and routines. And then the database, you have packages and stored procedures and things like that. So if you come up with an ontology that represents that, and then you map all the relationships, all the dependencies between things. So if you have a routine that uses a library or writes on a table, you need all of that. you need, if it was a human doing that, they would need to understand the entire picture, right? So this routine, where is it used? What is it using? What is it writing? What is it reading from? So if you come up with the right context, it's probably enough for the particular task. So ontology for me have to do with how do you represent reality? How do you represent relationships between things? And then how can you extract in a hopefully deterministic way exactly what you need for each task, right? So it was natural for us and then we evolved and then from code we ingested semantics about the database and now we are ingesting the migration tasks and the issues. And once you start thinking about this, you realize that the more things you have linked, the more things you can do.
John Furrier
>> You know, the SiliconANGLE and theCUBE Research, we've been looking at this, we're seeing that graphs are low latency on context retrieval in terms of GenAI. We're seeing that the capability of how you can recurse through the graph is actually a great thing. There's other studies that show that if you ground the data with knowledge graphs, you reduce hallucinations by 44, 45%. This brings a platform discussion to the table because the enterprises that's unlocking their data that's specialized to their domains isn't in the general frontier models. So AGI and general intelligence, that's the frontier model. That's ChatGPT, that's Claude, that's Gemini. But when you get into the enterprise, they got data and they have data that's sitting there and it's about to be unlocked. And they want it to be relevant to the AI technology. They want the context to be surfaced super fast and they don't want hallucinations. You talk about robotics, it's even more strict. You can't have an autonomous car be wrong, right? You got to have robots that's safe. So you're starting to see safety tied to, no hallucinations. You're starting to see the value of the AI being fast data context and low latency. Lay that out. Cause I think that is a general perspective. How do you make that happen? first of all, I'm sure you'd agree, but how do you make that happen?
Alejandro D'Andrea
>> Yeah. So I would say that most people are doing work without graphs that seems good enough at the surface level. So why? Because you might have, let's forget about code for a while. Let's say you have a bunch of documents and you want the agent to retrieve the right information for a task, right? So it might go to a document, then look around and find another one, and then maybe find another one. But no one will guarantee, there's no way to guarantee that it will retrieve everything that's relevant, right?
Alejandro D'Andrea
>> Right.
Alejandro D'Andrea
>> So agents are doing better and better work of retrieving and reading things fast. But the problem is that that discovery work is being done at the time that you assign the task to it.
Alejandro D'Andrea
>> Right.
Alejandro D'Andrea
>> It has to figure out what the information is and it will retrieve as much as possible. But models are lazy also. Right. So at some point you will just stop and say, yeah, this is all that there is. But it's not. There's a lot of information that's being left out. When you use graphs, you are doing that work up front because you are creating those relationships up front once, just one time. It's much more effective, much faster because you get, let's say, a document, a legal document. At that point, when you get it, you can create all the relationships to other documents, right? You can say, okay, this relates to this particular project and this particular loss and this particular lawsuit and whatever. And this person in the organization that handles this and this project where it happened. And you do that once. And then any agent, when you assign a task, it will know automatically from the graph that all that information also might be relevant. It might not, but at least it has a chance to visit those nodes in the graph and figure out whether it's relevant or not. So it makes a great difference when we migrate code. We find it sometimes very easy to migrate huge code bases, but we have other cases where it's very hard. We have a particular customer with each form, each UI form has like 700 associated files, it will always miss many of those if you just send the agents roaming free in the code base and see what they find. But if you have already mapped that in a graph, there's no way you will miss anything. So it's a huge difference on speed, cost, efficiency, and how good the results are.
John Furrier
>> Explain to me, let me rephrase, let's talk about the knowledge layer, because there's two conversations. There's the technical, kind of like the deep tech conversation. Then there's the human average person. So, knowledge layer on the tech side, we've heard things like it's a connective tissue between organizations, data and humans and systems, gives the context memory for reasoning that makes better decisions for agents and humans. Okay, great, check. There's a playbook for that. But when you talk to the average person, I'm sure we all have been there explaining what agents are and reasoning. It's like a brain. You got the left brain, right brain, you got the cortex. you can almost look at it like the brain of an organization. And that's what executives want on their business model. They want to have an enterprise with all their data and domain expertise. That's their competitive advantage, the moat. They just want to have a brain that can reason and do things, take actions. So explain in your words, this connective tissue piece and how does that fit in the architecture?
Alejandro D'Andrea
>> Yeah. So let me explain with something that's maybe more familiar for anyone. It's something that we are using internally for our own effectiveness in BotMan. So I have a particular brain that looks at all my interactions in the digital world. That is emails, WhatsApp text, WhatsApp audios, Discord, Teams, whatever happens on my computer at night, there's a process that will fetch all of that and build a raw layer with just my emails, just my conversations as they are, right? So that's the raw layer. It doesn't give you knowledge. It's just information. It's everything that I talked about during the day, right, on any digital channel. But then another process runs that creates that knowledge layer. What is that? Let's imagine that I have an article for a particular project, right? So it doesn't matter whether I had an interaction with the project manager via email or I had a Teams meeting where we discussed the project or someone updated the project in GitHub, so the code for the project. All of that feeds into that same article. So in one piece of information, I have everything I need to know about what we're building, how the project's going, politics, payments, whatever related to the project, I can put together in one article or maybe many articles, but together in a folder, let's imagine. So I have a folder for the project that those pieces of knowledge are the distillation of all these inputs that are scattered around. Right. So that's for me, an example of what is a knowledge layer. So then your agents will work with that layer, not read every email. And again, they will lose something because there are thousands of emails, but they can look at five files maybe and have the entire picture of where we are. Right.
John Furrier
>> So you can map the graph at the right place for the right data. Talk about the role of data because now we're seeing with graphs and we're hearing this a lot, certainly on theCUBE as well, is that there's a lot of data types that are undocumented that databases can't handle. Domain expertise, for example. Talk about that piece of it because in a lot of these automations, these workflows, there's a lot of stuff that gets done that's somewhat maybe documented, but it's not like formalized in a table or database.
Alejandro D'Andrea
>> Yeah. Yeah, so I think we are finding this difference between structured data, we could call it, which are the traditional databases that have rows with information, like an Excel file maybe. And then you have this unstructured data, which is a lot of emails and documents and conversations and all of that. And for me, graphs have the ability to even connect those kinds of things. So one of the things that we do, for example, aside from reading all the code and creating this graph, we also read the database and we have processes that will try to understand the semantics of the database. So, for example, if I have a table with customers, it will describe that these are the customers that are used in such a system for this reason. And this column in particular is the social security number and what it's for and what it's related to and all of that. So aside from having your physical database with all the rows and the tables and everything, you have in the graph a way to understand what that data is about and also to relate that with things that are not in a database. Like you said, maybe you have documents and all of that, but they are also relevant to understand a project as a whole. So you might think about pricing. You can have the semantics of the tables of the databases that contain the prices and the historic prices and all of that. And also you will have documents related to the strategy of pricing or to marketing campaigns. And so you can now have an agent navigate through all of that and come up with much better insights than just one or the other, right?
John Furrier
>> I mean, I think that's important. I wanted to bring that up because I'm hearing people connect their Slack channels, emails, again, all this unstructured data and it's nuanced. And I want to get your thoughts on this because most people think when I call a database, it's a search function. It gets an answer. When reasoning is in play, the agents are actually doing reasoning, which is recursive, it could be multi-step. That's a key piece, and that's why I think graphs are getting a lot of attention. What's your reaction to that?
Alejandro D'Andrea
>> Yeah, I think it's very important how you organize information, because you can have a huge graph. We have graphs with millions of nodes, literally, and it's impossible for an agent to find the right information if you don't have a good ontology, a good way for the agent to understand where the information is. One of the things that worked for us in terms of processing all this unstructured information, like I said, emails and files and everything, is link them through entities. So an entity for us will be a person, a project, or a product, right? Mostly those. So if I have a conversation on WhatsApp and I mention you, for example, the process that gets that into the graph will first figure out who is John, right? So it will look you up, it will see all the connections that you already have, the projects that you're working on, the interactions that we had and everything. And then it will figure out where that piece of information, that WhatsApp audio or text should fit. If it should, maybe it's garbage, but maybe it's an important update on one of the projects that are linked to you. So having entities in this case you, maybe the project is another entity that's related to other things. So for us, that's the way to make sure that you get to the right point in the graph. So we first figure out, OK, what entities are involved in this piece of information, in this document or email or whatever? And then how do we traverse the graph in order to figure out where that information fits? The same with retrieval, right? So if I ask you about a project or whatever, it will first identify the entities related to my question, in this case, you, and then it will go to you in the graph and then figure out what things you're involved with.
John Furrier
>> I mean, this is really the understanding piece, because without that context, memory, and reasoning, the outputs can't be trusted or explained, or even governed. and you have an answer, but it might be incomplete, creates hallucinations, drift, there's a lot of consequences. So I think the context piece is huge. And I want to ask you, because I think this is the secret sauce in my opinion, the enterprises today have an old school mentality, a lot of legacy, a lot of practices. This is the opportunity for say Neo4j and others in this area, because if you can nail the context with the things you just mentioned, you get more accurate information, you understand relationships, you got the ability to reason, and then you could track the lineage and the traceability of things. You could say, hey, he said that on email. He said that. That's the connection. So there's a lot of things going on that could be very valuable. Why is it stuck in pilots? So why are the enterprises not all over this? Or are they? What's your take?
Alejandro D'Andrea
>> I think it's a maturity thing. They are just getting started with understanding what agents can do for them at the individual level mostly, at the productivity level. But even then, people are seldom already thinking about all these connections that I told you about. So it's very natural for most people at this point to drag and drop a file into a GPT or Claude and ask about the file. But thinking that Claude could also have a knowledge base like a graph that has a relationship between that file and the project and the people and the status and the database and the product, all of that doesn't even cross their minds. Because it's a new way of thinking that we are developing. We probably started before and now we are thinking more ahead. But most people are just in the first stages. When I think of the future, when you see the news, OpenAI just released those 10 mathematical problems that were there for decades and now have been solved with $2 ,000. And you get more and more news like that. So intelligence is not anymore the problem. It's not a problem of intelligence. The problem is context. You have to explain to the agent everything he needs to understand to do a particular task. What if you didn't have to explain that? What if it already knew all that you know? So that's the key now. So if an organization comes to a point where they have a knowledge graph that has everything that ever was ever spoken in a meeting or sent in an email, of course, if it's not private. So that's another concern that you have to solve for, but it's solvable. Once you have that brain that has all the information linked, you can ask of an agent, probably just about anything, and it will do it. Some companies are already realizing this, but many are not even thinking about this. I also consult for some companies, and they're not there yet, but it's something that obviously will be the next step for boosting productivity.
John Furrier
>> I have a series that I'm doing out of our New York Stock Exchange studio called Transformation Edge. It's about competitive edge in the modern era, basically competitive strategy for AI. Everyone that I talked to at the senior leadership level would all agree that knowledge is their competitive advantage. Domain expertise, how we do business, how we organize our culture, et cetera, et cetera. The systems that have been deployed. There's almost no debate on that. When you say knowledge layer, you get into the data platform conversation and all organizations have been going down that road. You see Databricks, Snowflake, Microsoft Fabric. These are examples of IT systems. So how do we connect that? How do you connect that? In San Francisco at the Neo4j GraphTalk, we were talking about the Virtual Graph technology You can reason over existing data platforms So there's a solution in there. Explain the thought process to connect that IT fabric, whether it's modern graphs with legacy systems, because the end game is right in sight for the technologists because there's a hundred percent consensus almost that, our knowledge is our competitive advantage in the AI future. So you got the data in place. In some cases they're moving faster on data lakes, but now you got lake -based transactional and operational systems and analytics systems. Talk about that piece. How do the technologists connect?
Alejandro D'Andrea
>> Yeah. Yeah. So we have another product that creates this semantic layer. And in this one, we are not the only ones globally, like in the migration side. But there are products like ours that will connect to a database, try to figure out the semantics, what the data are, how they are related. And that's one bit of it. For some of our customers, we also use the source code of the applications that sit on top of the database to understand the database better. Because in the source code, there's all the information about what the table is, how each field is calculated, where it comes from. So database plus source code gives you a lot of information about your databases. With that information, of course, you can use agents then to query in natural language or to ask for analysis and things like that. And that is one big piece of the organizational knowledge that you can deliver or draw value from. But that's only one. Because when you think about business processes, for example, if you want to replace part of a business process, it's not enough, right? The database information by itself is not enough because a business process will traverse some workflow on that system, but then something goes through an email and then something goes through an Excel file and then you send someone something else. And that is the process itself, right? So that knowledge is not captured there. And you also need that knowledge, right? To understand the day -to -day activities of the people and the processes and everything, that's another one. How you evolve your technological systems draws from that. So if you could understand what people are doing on a daily basis, you would have AI also suggest improvements to the systems because people are doing this manually, they could be doing it automatically. So it's many dimensions of knowledge and the structured database is one of them. But if you keep thinking about it, it's like, okay but if we add this and then if we add this and that's where a lot of knowledge graphs come in because you can represent almost anything there is there's no limit, there's no structure, right?
John Furrier
>> Yeah, and you have millions of nodes. All right, this is obviously the data infrastructure is part of the AI infrastructure. It's not just GPUs as well as GPUs and memory scarce right now, we're seeing that build out go on. So that's a check. The data side is an infrastructure play, but now you have this new wave coming in where it's like, okay, we're seeing more dynamic environments, workflows that are kind of defined deterministically, probabilistically, you have personalization, autonomous systems. This is the framing that we're seeing. So the final question for you is, how do you pitch that to senior people? What would be your pitch to say, look, I think we can solve this. How would you boil that down into business speak to get that approval? Because the people who are doing graphs are rising right to the top because they're getting results.
Alejandro D'Andrea
>> Yeah, I would say I have a lot of ideas about how to draw more and more value from distilling this knowledge. So far, we have been focusing more on the migration side because it's something that's very tangible for people. It's something that they have been trying to do for decades and they didn't know how to, and now there's a possibility. But if you ask me, I would love to go deeper in what I said before. Let's have the data layer, let's have the processes layer, the technology evolution layer. I am starting to have some conversations with some CEOs about that, the big picture that comes ahead. But I would say I think most are not ready for that conversation yet they still have to realize the value from these first steps like the migration like the agents or like automation of certain processes but it will come naturally if you find someone that has a more forward uh visionand you can paint this picture of okay how why for example but why is your organization structured in silos right now? Why do you have like a treasury department and an HR department? Because no one can have the entire context of how everything works, right? So you need to divide things in simpler things to understand because we're human. But what if all that knowledge was centralized and one agent could understand the end-to-end of how to hire someone until they pay that person, right? It's absolutely doable. but for that you have to break your notion of how an entire company works. And you've been working with that for 30 years now. So it's hard.
John Furrier
>> Yeah, for the change management, Alejandro, it's so hard because there's so much impact. It's interesting, I hear a lot, I'll just share something I've heard with you and other folks watching. It's how pilots get done is one easy, when you mentioned get some wins, migration, whatever the use case is, get a win and see the results fast. But what's interesting, people are doing pilots differently. In the old days it was, okay, let's take the system, put it on the edge of the environment, test it, no blast radius really, we're going to do a pilot. And a lot of people get their fingers in it, slows down. It maybe works, maybe not. Most of the time it doesn't. And it takes months to get stuff going. Today with agents and with knowledge layers, they have the data in a data lake, they have the base platform. They're taking on the hardest business problems that they never would do before. They're all going after the revenue. Well, we've been trying to do this, but it would have taken a huge supercomputer. Well, they got that now. Okay, you got a super computer, you got intelligence. So what's interesting is smart technologists are going after the hardest problem and they get a win. And then the result is something material, revenue, cost. I migrated that whole system in a week, right? As an example, I'm sure it's different timetables, But we're hearing, years to months, months to weeks to days. So they're starting to see the acceleration of real value. So I think that, to me, because it's too hard to go in and wave your hands and say, hey, we're going to be a modern AI -enabled company. It's really difficult. But you can get the wins.Would you agree?
Alejandro D'Andrea
>> Yes, in every transformation. Where we work with agility, you would have to do pilots, and then with data also. So, it's like you said, we are humans. Change is hard for us. People are scared. The incentives even are most times not aligned with changing. They are more aligned with operating efficiently what you already have. So, yeah, yeah, it's the same recipe. I used to do innovation and agility and digital transformation. That was what we called it. It was the same recipe. You have to start small. You have to find people that are enthusiastic. You have to show results to everyone. You need a CEO that asks for this and spend some time asking for it and then enabling people. It's the same recipe. Even if we achieve AGI, I think we are already there probably, but it will take some years for us to adapt.
John Furrier
>> Now it's business transformation because the money's on the table. I've seen value with revenue. Alejandro, thank you for this conversation. Great insight, expertise, great to have you on the Neo4j AI Luminaries series. Thanks for coming in all the way from Chile.
Alejandro D'Andrea
>> Thank you, John. It was a pleasure. I'm a big fan of the work you're doing.
John Furrier
>> We appreciate that. from all over the world, from Latin America, North America, EMEA and Asia Pacific, it's a global phenomenon. It's a distributed computing systems opportunity and we're covering it here on theCUBE. I'm John Furrier, thanks for watching.
John Furrier
