AI-Driven Business Success with NetApp’s Hoseb Dermanilian

Hey, everyone. Welcome back here to Textron tv. Got, uh, another Textron TV interview coming at you right now.

I want to introduce you to Jose Derian Derian. I hope I pronounced that right, Jose, welcome. Welcome to Text Drunk tv.

It's great to have you on. Thanks, Slan. Thanks for having me.

Did I mess your butcher your name up too badly? No, it did great. You absolutely did great.

Hey, I've heard worse version than this. Don't worry about it. Oh, I'm sure.

Look, we all, you know, that's the thing about people's names. We've all heard it misspoke and misstated. You know, my name's Shimmel to me, it's very plain.

Some people try to be fancy, they say Shimel or you know, some or simmel or all kinds of things. Anyway, Jose is the senior director, global head of AI sales and go to market with NetApp. Oh, said, that's a great title, but talk to us, let's hear a little bit about your background and what, what's actually the job here that goes to this title.

Absolutely. Um, you've been with NetApp for almost 11 years now, and I started the AI journey with NetApp almost seven years ago when Nvidia started making these whatever called dgx one At that time, probably your audience will know about it. This is the small computer, small computers from Nvidia that have the GPUs in them.

So, uh, when we started selling this, we were selling to line of business owners like, uh, cardiologists and, and, and, uh, and cyber security personnels and all that. Um, over time, obviously we grew this business tremendously, that we built a specialist function within NetApp, we call them AI specialists, and that represents my team and I lead this business globally. People might ask NetApp is storage company, what do you guys have to do with, uh, ai?

And I have to remind them all the time that data is the, the heart of ai. And if you don't have the right data, enough data, the clean data, if you don't secure your data, you're not gonna advance your AI projects. You're not gonna have clear AI models, you're not gonna have functioning AI models.

And that's where NetApp comes into the, uh, game by providing our rich data management functionalities to customers who either want to build their own models internally, or they want to leverage the open source models from the cloud and apply technologies like RAG and fine tuning and the different things that we, I'm sure we'll talk about it later on. Excellent. And that is, I, you kind of headed that right off at the pass, right?

What is, what is a, a data storage company, data company, you know, doing with ai, but of course today AI touches on everything, right? It, it, the, the tendrils are, are deep into it. Um, you know, like every other technology I said that I've seen come across the radar screen for the last 30 years, sometimes the hype, the hype runs far ahead.

We get a little too far ahead of our skis, right? We, the hype runs far ahead of the reality. And there's a lot of people already who said, yeah, no, I've tried to say I think it's not working for us.

It's not, we we're not getting the results. We thought we're not replacing headcount. I haven't been able to fire anyone, you know, as a result, because AI's doing their job.

Um, I'm not sure if it'll ever live up to the hype and, you know, look, it's soden in my, it's so new. Yeah, it's been on the scene now for a few years, but it's still, we're still just at the beginning stages of this. But eb as we sit here today talking to the audience, they wanna be successful with their ai ai, they want to use it with cybersecurity sustainability, especially when we talk about data or data storage, right?

You know, all these hundreds of millions of dollars being set aside for new data centers and stuff. But we need power to run them. We need water to cool them.

We need, you know, uh, everything, uh, connectivity, the, the actual hardware itself. How, how can we, how can we use or leverage AI to help us here? That's a very good question.

I think there is an important thing that you said is it's been around for many years. We started with machine learning and then it became deep learning, predictive ai, prescriptive ai, whatever you want to name it in that, in those days, now we're in the gen ai, uh, cycle. It is still very early for lots of enterprises.

And I think you, you mentioned all the things needed to have AI functional, but data is the one thing that the reason, listen, Alan, the reason we go into a lot of customers today and they tell us this is not working for us, is because if you take this models, be it Lama, be it grok, be it, uh, open AI or you know, be it GPT, whatever, they were trained on generic data, and you can't have that and put it now to answer into your customer success chat bot because it might not give answers based on your own needs, based on your own experience. You know, when we hire people, we don't put them into the job on day one, we sent them into a training, right? When I have a new hire, even if it is from the same industry, we sent them into a new hire training to teach them on our company's specific aspects, our products, our support model, our, you know, our own customers.

And the same thing needs to be done for these models. When you deploy them, you have to take them through a new hire training where you train them on your own specific data so that they become specialized in your own specific products. And now they become more useful than just bringing an open source model and saying, Hey, go and fix this thing for me.

Like, I can't just take a model and deploy it. Let's take a use case. It's a very simple use case, which is the customer support, right?

Everyone wants to automate that. Everyone wants to make it more ai, um, you know, generated one, but you, you know, that that bot might answer to your customers in a way that might trigger them to go into a competitor, which you don't want that. So that's why it's very important.

People will build these models and enterprises have to take that model and now fine tune it or, or retrain it or use technologies like RAG or CAC to make it specific to our own businesses. And this is where we expect to see more usefulness of AI in the world, rather than you and I just saying, Hey, give me a recipe of, uh, chicken nuggets that I want to cook today. Right?

I could have done that with Google probably in three minutes. Now I can do it with AI maybe in 30 seconds. There are other useful, um, use cases today, but for the generic enterprise to consume ai, there is a fear today that if I bring a model and I put my data into it, then it's gonna get compromised.

Right? So, to give you an example, we have done a, a survey where we interviewed almost 1300 tech executives, 80% of them, like 79, 80% of them said, we need to first unify our data, we need to bring them all together so now we can deploy ai, right? Uh, without that, it's not gonna, we're not gonna see any, any meaningful outcomes, right?

It might help us here and there. It might write a blog here, it might generate a marketing content there, but without bringing the, the data and unifying it's gonna be a problem, 40% of them were worried about the security of the data being exposed to these models. So, um, you know, long answer, uh, we will see a ai, I think we have crossed the, the, the, the bubble.

We have crossed the hype in my opinion. Um, we can clearly see the use cases ahead of us. It's our turn now to make those use cases and implement them and get beyond the POC.

And to do that, we need to cross two major things, data and security. I, I, I agree with you a hundred percent. You know, speaking on the data and the data unification aspect of it, I think one of the things we're doing is, I forgot the term, everyone's used the frontier models.

Mm-hmm. For these large LLMs that, you know, the, the big ais were trained on, they're great for like parlor tricks writing a blog post, they're making a picture, you know, the kind of stuff that kinda thrills gets oz or Nas outta people who don't really understand how, how it's working in there. But real work is getting done with companies who've been able to take their data and train the AI on That smaller set of data, right?

And I think that's where you come into the whole data unification thing, right? You've gotta have, you've gotta have your unique data set, you know, trained. You've gotta have the, the, that has to be, whether it's still called an LLM or a small language module or whatever you want to call it, it, that's training the AI on that is where you start seeing amazing results, right?

With the, we recently covered a, uh, a new AI that's excels in code because it's been trained on just some really great code coding data sets. Yeah. But even with coding Alan, right there, there's, uh, companies have their own standards in writing code.

They have their own frameworks in writing code. There's some, some, some, some, some things allowed, some things not allowed, even with the coding. If, if you plug in your own source code into the tool, then you might have 99% accuracy.

It might write a code based on your standards. But again, it basically taking 30 years of your IP and work and dumping it into a model. Now, I'm not saying that every tool out there is not secure.

There's a lot of enterprise versions of, of these, you know, the, the tools that are available out there that might you even deployed on premises. Because here's what also happening. People are also creating on-premises versions of these clusters of ai, and that is air gapped, right?

They can just train the models internally. There's no, um, adversaries or it doesn't leak into the outside world. And you just, you can have a very useful AI use case there.

But how many enterprises can do that? How many of them can afford, back to your point, the power, the cooling and all that. Um, we see a lot of our customers go to the cloud because these models exist in the cloud.

They can turn on a couple of GPUs in the cloud, don't need to, to spin up a whole infrastructure. But then there's the question, do I bring the mo the data to the cloud? Do I bring the model back on premises?

And it's a hybrid world. We have, uh, obviously customers who are, have been on this, you know, unification journey for many years now. You know, Johnson Johnson is a very, uh, good example of one of our customers who have been in this AI journey for many years with us.

And now we, even with gen ai, they're trying to leverage the hybrid cloud strategy. Um, that's one example. We have other examples in automotive.

Um, automotive generates a lot of data. And it's a perfect example of, um, a customer who would need to plug in their own data that is coming from their own cars and getting dumped into a, in a unified lake, and now passed to the model so that the model will be trained, right? Like you can have, you can be driving an X model of a car and then deploy, um, deploy that trained model into a y version of a car.

So, so, or a y brand of a car, it might not give you the same performance, it might not give you the same, um, customer experience. Again, as you said, personalized data is very important. And here's the thing.

I'm not saying something that people don't know. Um, I think a lot of, a lot of different organizations today know this. A lot of the vendors also, speaking of the same, hey, you need to bring your own data.

What, where we come in as NetApp is we say, Hey, because we've been storing data for many years for our customers today, for almost 30 plus years because of NetApp's existing in the three hyperscalers as first party service, we are perfectly positioned to make that happen for our customers in a secure way. Meaning, um, if a customer comes to me today and says, Hey, I've been storing data on NetApp and I'm using now this model on Vertex AI from Google, how can you help me? Then we have a perfect way to manage that data for this customer, push some of that data to the cloud, train the model rehydrated back, use caching technologies.

So we are making it happen. We're not just staying, or I'm not another person coming to your podcast and saying, yes, you need your own data. I think we know that, but how do you make it happen is what organizations or enterprises want to hear today.

Um, and that's where we come in as the defacto storage vendor for a lot of enterprises today in the world and also our existence in the three hyperscalers. Um, what we don't do is we don't sell models, we don't sell algorithms, but we connect to those, right? And we connect to those in a secure and efficient way, so that this notion of you need your own data to make the model work happens.

I love it. What's up? I just wanna get one thing straight.

You are selling the storage. You could, we could have people store whether in the hyperscaler on their own prem or what have you. Are you, when you say you're act you're helping them train the models and so forth and, and, you know, do these things, is that sort of like a pro services type of engagement or you just kind of giving them best practices and they they do it.

So what we do is we run an AI design workshop. So we Oh, you do? Uh, yeah.

We run AI design workshop. We partner with a lot of our, uh, partners today, um, like, I don't wanna forget any names, but worldwide technology, CDWE plus insight, and oh, mark, three different names, right? In, in the industry.

We partner with them because they provide the full stack. What we do is we provide an AI design workshop where we come in and say, this is the data you need. This is where your data has been seeding on.

Um, this is how you can connect these different data sources together. Uh, we also have an, an ai, you know, tool bench on the cloud that is provided by our cloud team that can help customers connect the data sources to this model. So we are not the ones basically hands-on training the models, but we are the ones at least getting the data closer to the station where the bus needs to leave.

And then that, that's where the, either the data scientists within the organization takes over, or we partner with the names I mentioned. They have a, their whole data set of data science teams that now can take over the baton and say, okay, now I've got the data. Fine tuning the model or ragging the model is, is the next step and we will do it.

But getting the data to that step is actually 80% of the process, because once you have the data, it becomes how much compute you have and how much faster you can do, right? Um, because the tools are there, the models are there, what you need is GPUs. What you need is, uh, a place to store the data.

But once you get that data to that point, I, I think you're the majority of the problem is solved. Um, and holistically what, what organizations need to think about is overall security of this whole platform, right? Not only the security of the data, but the security of the whole platform itself.

Excellent. Let's talk a little bit in our remaining time about security and then sustainability. What is NetApp specifically doing around the security of the data kind of built in, if you will?

Yeah. Um, so we are specialized in the data security itself. So we are providing ransomware protection today, which is guaranteed ransomware protection.

We've got aaa, um, certificate for that. So that's one part of the, of the equation, because whether it's AI or not, ai, you, ransomware is everywhere today. And with, with the AI being so powerful nowadays, adversaries are using it to, to create newer attacks, to, to do attacks in a faster shape.

Uh, even, even the people that we surveyed, they said, like 80% of them said they are expecting some sort of new attack to hit them every day. So ransomware is one, is, is one aspect. What what we are also doing in the realm of AI is when I am able to not let your data scientists copy data left and right and have this data land in some unprotected place, whether it's cloud or on premises, whether it's on a USB stick or on a hard drive that is not monitored or protected.

If I can do that, I am saving you. Your, your, your, your job. I'm saving you, your enterprise, I'm saving you your ip.

Um, so when we say we can move that data securely between the three hyperscalers and on-premises, it's a major advantage of NetApp over, over anything else, whether it's manual or non-manual, because by doing that, it stays under the protected systems of ontap. It stays under the same ransomware protection. So the data is not going anywhere outside of this protection umbrella if you would like.

So that's one way. The second way is we are adding soon, like rag in the box functionalities that will also limit moving the data outside of this system. So basically we'll be able to vectorize the data in line, in place.

What, what people do, Alan today is, I don't know if you heard of the rag, which is retrieval augmented generation. Sure. Where it needs the data to be vectorized.

So let's say I have hundreds of PDF documents, 10 terabytes worth of data, I need to vectorize them the easiest way, copy this into the cloud, vectorize it, and then, you know, the embeddings goes into the model. But what happens to this copy 10 terabyte of data? How is it protected?

Where is it protected? Right? What we are doing is we're saying, don't move that data, vectorize it in place, and now just move the embeddings out, which is still, you know, ones and zeros.

It's it to the outside world, it looks garbage. Um, some people can still do reverse engineering on it, but it's very hard. But it's, it, it, it adds protection right now.

You don't need to move that data into somewhere else. You just move the vectors. Um, that's one area.

And then the whole, um, you know, doing AI responsibly is another way where we are adding, um, traceability and versioning, right? One of the things that we are doing or helping our customers is we integrated with tools like Jupyter Notebooks and, uh, domino Data Lab and all the different ML ops platforms so that customers or data scientists today can version the model along the code, along with the data so that they can go back to the data that we used to train this model. So we don't come into a situation where we deploy a model into a real life scenario, like driving a car by itself and something happens and we need to go back to the data that was used to train this model, and someone says, oh, I can't find this data where it was, or what it is or what it was.

So we can version that and we provide versioning, we can provide traceability, so traces the data back. So, um, from a security perspective, those are the major aspects. We are, um, uh, we are also working on some quantum proof encryption as well.

And, you know, that's gonna come in. Uh, I don't wanna, that's a whole separate episode on itself, but Yes, it is. We, yeah.

So we're working on that one too. Excellent, man. Excellent.

Jose, we're about outta time, but for someone maybe who wants to go visit the NetApp site, get up to speed on this, maybe engage your team, check out what's there, where within NetApp should they go? com/uh, artificial or ai, uh, they can use that one. Uh, it it lends them into, into all the different, uh, tools and, and blogs and webinars that we have.

That's important. Josee, keep up the great work, man. Come back and keep us posted.

I mean, the, the crazy thing about the world we're living in with AI right now is, you know, six weeks from now, you may have a whole, a whole new set of stuff you're showing us, right? It's crazy. Every day I wake up and I'm like, oh, let me see what happened in the world.

And, uh, now I see Sonet getting connected to cursor. People are, here's the other interesting part, right? Um, you know, I'm a big X user and I go on my X and I'm like trying to find out customer's feedback or people's feedback, and people are saying, Hey, you know, I, this is not responding the way we want.

It's actually feeding data, the different data that we, it's, it's very interesting to see what's happening in the world today. And this is gonna grow even at a faster speed. And that is what is also, you know, giving us a hard time.

Because if you want to cope with this challenges and changes, you need to have a faster moving teams too internally from developers and from engineers, because if you're not coping with it, you're, you're late. You know, the old Irish verb, right? May you live in interesting times.

Correct. These are certainly interesting times. They are crazy, but it's exciting.

Jose Derian, senior director, global head of AI sales, and go to market at NetApp here on Techstrong tv. Thanks for joining us. We're gonna take a break.

We'll be right back.