Future of Data Platforms Summit | Frederic Van Haren, HighFens

Clips
More from Future of Data Platforms Summit

Frederic Van Haren

CTO & Founder

HighFens

play_circle_outline Introduction of Frederic Van Haren, CTO and founder of HighFens.

play_circle_outline HighFens focuses on AI and data-centric solutions for organizations.

play_circle_outline AI solutions offered by HighFens combine hardware, software, and services.

play_circle_outline Organizations utilize multiple data platforms, complicating data integration efforts.

play_circle_outline Optimizing Data Lifecycle Management: Addressing Data Gravity Challenges for Enhanced Efficiency in Modern Enterprises

Info
Transcript

Frederic Van Haren, HighFens

Frederic Van Haren

CTO & Founder HighFens

In this Future of Data Platforms Summit interview, Frederic Van Haren, CTO and founder of HighFens, joins theCUBE’s Rob Strechay to discuss how enterprises are evolving their data platforms to meet the demands of AI-driven applications. Van Haren shares his perspective on shifting from compute-centric to data-centric strategies, highlighting the role HighFens plays in guiding organizations across consulting, services and full AI solutions.

The conversation explores how enterprises are balancing infrastructure decisions between CPUs, GPUs and storage ... Read more

explore Keep Exploring

What is HighFens and what approach do they take towards building applications for end user organizations? add

What inspired the creation of HighFens and what expertise does it aim to provide to organizations? add

What are the key components and challenges of providing AI solutions in today's dynamic IT environment? add

What role does HighFens play in assisting organizations with their data challenges? add

What are the challenges associated with data collection and management in organizations? add

bolt Powered by CUBE AI

Frederic Van Haren, HighFens

search

Rob Strechay

>> Welcome back to the Future of Data Platforms Summit on theCUBE. I'm Rob Strechay with theCUBE Research, managing director and principal analyst. Today we're going to continue talking about data platforms and how they are changing and evolving to meet modern data-driven applications needs. In this episode, I'm joined by Frederic Van Haren, CTO and founder of HighFens. Welcome on board, Frederic.

Frederic Van Haren

>> Well, thanks for having me.

Rob Strechay

>> Well, we've known each other for quite a while. I mean, again, you've been my customer when I've been at one of the data platform companies, and you've really focused in this area really around AI, how AI, traditional AI and ML has really evolved into now agentic and types of workflows and how that is really bringing things together. And you have some really interesting perspectives on working with end user organizations and how that they're getting to build these applications and solve for that kind of data gravity issue. First off, why don't you tell the folks at home kind of what HighFens is, what you guys do so they get an understanding because I think you have a different approach to these applications and how you help them.

Frederic Van Haren

>> Right. So let's start with the beginning. So before I started my own company, I used to work for a speech company for a long time, and the experience I gained over there was to really go from a CPU centric or compute centric to a data centric approach. In 2016, I did a little bit of AI, large speech recognition. I mean, in those days we're also calling it large language models, but clearly different than the large language models today. And in 2016, I said, "Why don't I help other organizations get up to speed and use our expertise to get them to a point where they can be really efficient?" And so that's what HighFens really was created for. And so there are a couple of things we do to help customers because not all customers are the same. So the first thing we do is consulting. And so what is really consulting is if people don't really know what they're doing or what they need to do in order to be successful, that's really consulting. You can look at it as a roadmap, having regular conversations, what your weakness, what your strength, what you need to get. And then services is really more hands-on. That's really where it gets really interesting. And our customers can be mature, meaning they already have a lot and they have some problems, they want some fine tuning, or they have really not a lot and they're asking us to help out. And so the final thing we're doing, which is really our main business today, is what I would call AI solutions. And so AI solutions, it's all of the above plus hardware plus software plus services. And so why is this important? It's because today it's not a matter of buying a storage device or a GPU server or a piece of network. It's about the combination of it. And so the challenge is traditionally IT would buy equipment for three years and it would be relatively static. AI is so dynamic that it's a moving target, and certainly around data, right? And so where we spent a lot of time today is on one side with mature organizations who heavily focus on data. And then you have the other organizations where they're less mature and they heavily focus on, "I need to buy GPUs." And so there's a little bit of everything in between. And then the final thing I would like to say about what we do is there is training, which is batch processing, which really where all the data lives.

Rob Strechay

>> You mean AI training?

Frederic Van Haren

>> AI training people, that's right. Exactly.

Rob Strechay

>> Training the models.

Frederic Van Haren

>> Training the models. That's right. And so that's one type of challenges, right? And then you have the second challenge, which is the inference where you take the output of training and they're also... It's not a single line. There's a feedback cycle, right? You put the models, you deploy them, you get some feedback, you have to update the models. And so you're really never done. And that's really where HighFens comes into play, training inference at all stages.

Rob Strechay

>> Yeah. And like you said, you kind of work across a whole lot of different industries, help people, again, like you said, they're at different levels of maturity. Even in big companies sometimes they're not as mature as you would think they are in dealing with the deluge of data that they have, and sometimes they don't even know what they have. So it kind of help people understand some of the challenges that you see when you get into these organizations and you help them build these roadmaps out, especially when it comes to that data layer, that data platform.

Frederic Van Haren

>> So the challenges is that it used to be where it's very difficult to get access to data. And so nowadays, because there's so much data, there are a couple of challenges. So the first challenge is to try to teach organizations that getting the data in from wherever the sources and different sources, different formats, that's one problem, but that storing that data. It's not extracting value, right? Storing the data is buying a bunch of books and not reading them. It doesn't mean you build up expertise, right?

Rob Strechay

>> Right.

Frederic Van Haren

>> And then the second thing is really about scalability, is once you have stored the data and you want to process it, there's a whole data lifecycle that sits behind it. And so we have customers who are dealing with data silos where they're saying, "We have a lot of data, but we have different groups who want to use the data differently." So an example would be you could have financial data being collected and then there is one team who is really caring about price of stocks today. And so those people are really caring about the data that came in the latest. While the trending people who are using the same data sets, they are interested in more historical data. They really don't care what the price was yesterday or the day before. And so if you combine all of that, a lot of organizations are saying, "How do we do this?" Because it's in one case, the cleaning for example of the data happens through a CPU, so you're not going to use a GPU for that. But the next stage where you want to process and extract value, you use GPUs. And then once you want to test, maybe you want to test on CPUs or GPUs, and every time, the storage could be different types of storage. So there is a need for organizations like us, I mean this is kind of self-advertisement, but there's a need for organizations to help enterprises understand that there are options and that a lot of organizations have a lot of different flavors, and putting these flavors together is not necessarily easy.

Rob Strechay

>> I was going to say, that to me is the striking piece is that, I mean, some of these organizations, pharmaceuticals, the automotive and traditional AI types doing a lot of different things in the back end have been around AI for a while. They have massive datasets. Pharmaceuticals in particular have just oodles and oodles, and then you have healthcare and others that get into these different verticals where they're collecting so much data. I was talking to the head of platform engineering for one of the large automotive companies, and we were talking about just how they don't only have Databricks or Snowflake, they don't only have a Redshift or a BigQuery or what have you. They also have a lot of on-premise data which they're trying to gain access to in an ongoing basis. And they're using file and they're using object, and then they may be using NVMe up in this actual servers for go fast types of things to feed the GPUs. Are you seeing it just as a all-of-the-above strategy for these organizations as you go and approach it and helping them because it's cost optimization thing as well?

Frederic Van Haren

>> Right. I think that the challenge starts with the collection of the data, right? So in a lot of cases, the data is very fluid, and so if you don't catch it, you lose it. And so in some cases, you don't really have a choice on picking the right solution for the right target. The other problem is data gravity, right? So if you have five petabytes on a storage device that's five years old, moving to a new storage device is very, very difficult. So you have a tendency for people to stick with hardware longer than they should be while at the same time other groups and in the same organization are using newer technology. And so you still end up with a flavor of main latest and minus one and version minus two where you have to deal with. I mean, the other thing you said was really key is when we started with HighFens in 2016, people were asking us, "Which vertical do you play in?" Because the assumption was that we would have absolute knowledge within a certain vertical. Nowadays, people don't ask these questions anymore. The question is about, "How can you help us with our data?" And so the point there is, unless you're really talking to hardcore low level data scientist, it's really a data problem, right? And the data scientists, they don't necessarily care about the hardware, right? And so somebody has to care about the infrastructure and it's not always that easy. So you have a lot of ancient problems and all kinds of bottlenecks that you have to deal with. And I wouldn't like to be in the shoes of companies today because it's going so fast. I mean, look at my own experience. We doubled in capacity every 18 months. This was like 2006 to 2016, doubling every 18 months. And a new storage device might come out every two years. So you really need to have some kind of a transition plan. Data lifecycle management, I mean, we talk about data platforms and all that stuff, but it's really key. It's a big word to say there's a lot of data. It's very fluid, very dynamic, very siloed, and no one solution does it all.

Rob Strechay

>> Right. And I think no one solution catalogs at all as well. Because as you talk about getting up above that storage layer and things of that, now people are bringing Iceberg into the mix. And so putting that on top of file systems and objects and making it more integrated, I guess, you could say. Do you see that as helping some of the organizations where maybe they are sweating assets even longer than they should?

Frederic Van Haren

>> I think it's a necessity. So if you look at the traditional infrastructure in the early 2000s, we're using a database and maybe at 10 events a second, then a database could easily handle that. And so the traditional databases were great, but they had some issues, updating and deleting, you would be blocking rows, right? And so you needed infrastructure and solutions that kind of still use the same methodology, the concept of a database, as you mentioned, Apache Iceberg is a good example where it doesn't eliminate, but it reduced significantly the problems regarding to blocking and at least locking updates and deletes. And it does it in such a fashion that now you can handle millions and millions of events in the same database, right? So I think those innovations are really key for enterprises, but it adds another level of abstraction, right? It's not easy, right? People want to buy a database and does it all. While in the reality nowadays, you kind of have to piece this together. Here's the interesting piece is the interesting piece is that there's a lot of tools out there that can help you out. All right, the question is which one do you want to pick and which one will continue to evolve? I mean, w talked a little bit about the different data platforms out there. Some of them started out as being heavily focused on compute. Others started out as heavily on data. While the reality is that all of these data platforms have to migrate and come to a solution where data and compute works together, right? Moving data is still a problem and it will actually get worse, right? I mean, the data problem is much, much, much bigger than the speed at which technology innovates, right? And so even though some companies tried to convince you otherwise, but the hardware is actually lagging behind the pace of innovation.

Rob Strechay

>> Yeah. And I think that's where... I mean, I'm a software guy, right? I did hardware because I worked at a particular place and had to put my software on something because funny enough, it doesn't run in air. So when you start to look at it, you dive in. I think a lot of these storage devices are also becoming very large software defined in many ways internally and externally to them. Do you see that being tied up with, because we're big fans of things like DBT and other stuff, the modeling technology where it's, "Hey, I need to be able to understand lineage and indexes and things like that." Where do you see all of these new features that seem to be kind of coming down towards that storage layer?

Frederic Van Haren

>> Right. If you look at the let's call them the file system vendors, I mean, just supplying an API to create, modify and delete files or move directories, those days are over, right? They have to supply a lot more. And the reason why they have to supply a lot more, it's a little bit like the databases where the file system traditionally were relying on metadata. Those metadatas are really databases, which the same problem. So you will see the file system vendors kind of adding a lot of functionality. I wouldn't compare with Iceberg, but similar concept where you want to try to reduce the blocking, and then you have the DBTs of the world and others where you take access to data and then you try to get as much value as you can, right? And it's not a single step process, right? So a lot of your viewers probably are familiar with the silver, gold-

Rob Strechay

>> Bronze.

Frederic Van Haren

>> Bronze, yeah, kind of strategy where it's not one solution, right? You go from one to another to another, and then you might have multiple at the end, multiple goals in order to deliver your data in a usable form. So I do see that as really key. And it's revolutionizing, right? It's like every day when I read a little bit more about some other vendor, there are some features that come up. But the name of the game really is, in short, is you have data sitting on a pile and you want strategic value out of it and you need to automate it through a workflow. And since I'm coming from the AI world, agentic AI is now really being pushed as, "Well,, maybe agentic AI can help with those workflows. Maybe agentic AI can do a lot of those automations." But in reality, you still need software to glue everything.

Rob Strechay

>> Right. And I think that's what we were talking about is when as you go from the storage layer into that data services layer, which, again, gets you to technical metadata that we were talking about where, hey, maybe this gets unblocked some of the file systems, some of the object storage, allows me to access data faster and brings that up into that data management layer, which is where sort of the DBTs and others and Informaticas and others live in that kind of data management. And then up above that you have the GRC layer and real true business and technical metadata coming together. Where do you see, as we look into the future here, how far does this consolidate? Because right now most organizations in our estimation are using three to four different products in that stack per silo and then trying to glue it together. How do you see that going forward?

Frederic Van Haren

>> Well, I probably do agree it's three or four. I would even say it's more for the simple reason that those products leapfrog each other. And since a lot of them have a buy-in trial or even open source, depending on the organization, they might always have the attitude of always looking for the next tool. And so that's one big issue I see. It's not three or four and stick with it forever, right?

Rob Strechay

>> Right.

Frederic Van Haren

>> It's not like Microsoft Office where people have been buying that for the last 20 years. You will see there's a lot of competition there. And I think from an enterprise standpoint, it's very difficult to say, "Am I going to bet the farm on this product? Maybe they get acquired, maybe they will disappear. Maybe even worse, there's another one that's better and our competitors are going to use them." So I think there's a lot of strategic effort that's going to happen there. And since we're talking about the future, I think there will be many more organizations or startup vendors, if you want, that will provide a better way than the current way. And one of the reasons for that is that the biggest challenge still today is you can have a pile of data. And certainly from my speech recognition days, it took us six months to rebuild all of our models, right? And then now we're probably at two or three days.

Rob Strechay

>> Wow.

Frederic Van Haren

>> That's being as a factual. The reality is people want this to be almost real time, and the future is... And I read something about Amazon a few years ago, not sure if it's true or not, but where Amazon was kind of saying, "We're going to start sending packages to consumers before they decide to order it." And I think that's the holy grail. If anybody can come up with a solution or a workflow where you can analyze the incoming data much, much faster than today, because that's one caveat I see with the problem today is people understand that data is valuable, but then they collect so much data and throw it on a pile, that everything that follows is inefficient, right?

Rob Strechay

>> Yeah.

Frederic Van Haren

>> And so I hope that in the future, there's a lot of analysis that can happen when the data comes in such that they can figure out which data to keep or to delete as early as possible. I think this is still an area where people are saying too easily, "Let's throw it in Databricks or let's throw it in Snowflake." And then realizing that the business value of that data is not that high. I think, maybe I'm hoping that in the future we can do a better job there.

Rob Strechay

>> Yeah. And that's where you guys at HighFens really spend a lot of time is helping them understand what's valuable and not valuable based on the use case they're trying to solve for.

Frederic Van Haren

>> Exactly. So our more mature customers are always asking about, "How can we improve?" and asking for the impossible, right? It's like, "Hey, look at a piece of data. Can you tell me what the value of that data has?" And so we do have some products on the side that we have built in order to help and guide customers.

Rob Strechay

>> Some software that you've built, yeah.

Frederic Van Haren

>> Definitely software, yes, to help them kind of guide. And surprisingly, it's not only helping identifying business cases or where the value sits, but it also helps a little bit kind of eliminating the needs to throw more hardware at it, right? I think that's a whole different topic, but it's much easier for an organization to say, "Let's throw some more hardware at it," because that's something they understand as opposed to, "Let's use some software to fix the problem."

Rob Strechay

>> Totally. It definitely does. So as we bring things back around and wrap things up here, again, looking out into the future, do you see that AI, and you talked about some of the AI tooling and things like that, but you still need the software to knit it together, do you see that as AI and more traditional cloud-native or even non cloud-native applications start to become all entangled there, do you see that these data silos just continues to grow and become more complicated?

Frederic Van Haren

>> I think it will get more complicated because the solutions will be more complicated. But I think data silos, it's a cultural thing. And so some organizations have so many different groups and they don't necessarily know from each other what they're doing. I think data silos, very, very difficult to eliminate. It's also very difficult for an organization to force one system on all the groups. I do think it will still be a problem and it will be a bigger problem as it will take more effort to fix the silos.

Rob Strechay

>> Yeah. And we're doing some deep research in this area, and we're going to definitely try to pick at that as well. So hey Frederic, thanks for coming on board today. This has been great. Always, we could sit here and probably talk for an hour easily, but there is time, so next time.

Frederic Van Haren

>> Great. Well, thanks for having me.

Rob Strechay

>> Yep. And thank you for watching this episode of the Future of Data Platforms. We'll be right back with more.