Scaling AI Horizons: A Conversation with Victor Pereboom, CTO of UbiOps

0:00Welcome to the AI Chat Podcast. I'm your host, Jaden Schaefer. Today on the podcast we have the pleasure of being joined with Victor Peribu. who is the co-founder and CTO at UB Ops, which enables data scientists and engineers to very quickly turn algorithms into scalable, robust, and secure end-to-end applications without requiring knowledge to set up cloud infrastructure, microservices, automated scaling, or DevOps practices. So super excited to have you on. Welcome to the show today, Victor. Yeah, thanks. Great to be here. Super excited. Now, so a question I wanted to kind of kick this off with is I'm wondering if you can give us a little bit about your background.

0:40First off, my question would be like, did you always know you were interested in AI into this space? Was this something you kind of found throughout your career journey? Walk us a little bit through your kind of journey that brought you here. Sure, sure. So my interest in AI and data science started probably about like more than 10 years ago. I was still at university studying aerospace engineering at the time. I did a lot in autonomous control. And there is a lot of, it's the more like computer science, data-focused side of aerospace, I would say. And it was around the time that big data became kind of topic and more people started exploring what's possible with machine learning and those techniques.

1:22And I don't know, I've always been interested in more like software and computers and all that. So just out of curiosity, I started diving into the topic and just talk with me like this is really powerful technology. So yeah, I started also including kind of courses and things and just reading up on what's happening on data science and AI and big data back then and how it could be applied to what I was working on university. And later that, yeah, so that kind of handled my interest in the subject. and later I got the opportunity. I was more kind of well-versed on the topic then and had the opportunity with a friend from university to kind of start a company in the space.

2:13And we started actually working on a different range of topics, like solving machine learning problems for customers and especially around the topic of predictive maintenance, which is really cool because based on data, like IoT, sensor data, all that stuff, You can kind of predict if any device or component or something is going to break down at some point, if there are any failures and things. We did a lot of that work in railway infrastructure and tried to predict failures there. So it also was solving a big problem back then. It was really, really interesting to work on. And from there, we started building a company around that was more like professional services consultancy in the space and also trying to productize the things we were developing.

3:11So productizing the AI solutions we were building, like the different predictive maintenance models we were developing. And we just ran into the issue over and over again of like, hey, we need some kind of standardized infrastructure to kind of take the models, which were basically just like pieces of Python code, right? With statistics and everything in it. And how do we actually turn these into a working live application that we can sell and that we can sell to people and sell to the businesses that need them? Because that's eventually what they were going to use, right? Right. Kind of the type of applications that we're going to use in the field.

3:56We couldn't just ship the model and say, okay, good luck. And our work is done. we needed something yeah so uh let's let's say for predictive maintenance um let's say that that the end users of the application were more like the actual engineers who knew everything about the railway infrastructure but not much about data science and ai so they needed something they could you know like show them on an ipad when they were out there and just see what was going on But to get that whole system up and running and working, we needed to host our AI models, let the data flow through them and put that entire also IT side and cloud infrastructure in place.

4:43And for every application we were building and for every customer we were engaging, we had to do the same thing over and over again. So that's how we started kind of building that AI, machine learning, serving infrastructure that we now turn into a product and a company. That's incredible. That's super fascinating. That was a very long intro, by the way. Well, I love it. You gave the whole story. um so kind of diving into that a little bit and and kind of what that looked like at the beginning um you know what were what were some of your first steps when you decided you you know you wanted to make ubi ops was this something that you had you know kind of been working on some of this technology in the background and it was kind of a natural launch where you've seen some of these issues and you're like okay we need to like make something completely brand new um how did that partnership go with your you know your co-founder how did what did that kind of look like yeah so by By the time we turned UbiOps into a product, we were like a company of 15 people, probably.

5:51Most of them focused on data science machine learning. And we realized that this was a problem. So basically, the deployment and productionizing of AI machine learning, so turning the models we had into a working application was not only something that we were running into like that same challenge but we saw that like everywhere so also other data like analytics teams we we talked with and uh other companies we uh uh yeah we spoke to it was like yeah we all run into this problem of you know having this um um having good like data science data analytics capabilities and having the people in the house to do that but then at some point turning it into actually like a reliable product and service that's really hard because yeah data scientists are yeah those teams are not very at home in like cloud computing it software in general and to build a stable solution you need software engineers and so it started more as an internal product and an internal platform we were using to just deploy our own solutions.

7:01And then at some point we decided like, hey, this, like the market, the timing is right to also help other teams achieve the same goals. So we started actually talking to a lot of those teams at the companies we were working with saying like, hey, if you had a platform like this, would that be useful? And that's how we started kind of defining the scope of the UbiOps platform. And from there, we did internal development, started with a beta and started getting it out there into the market. Very cool. Exciting times. One thing I'd love to ask specifically about the platform and kind of what you guys are building is, how does UbiOps kind of simplify the process of turning algorithms into scalable applications for data scientists?

7:55Right. Right. So, great question. So it's a platform that sits between the data science code and like the cloud and IT infrastructure. So the main goal is making it as easy as possible for a data scientist to take his AI model, machine learning model, code, push it to our platform. and we do everything from, let's say, building a software container around it, scheduling that to doing like all the API, API management, traffic queuing and all that. So as soon as you push your code to our platform, you get an API endpoint in return. So it turns it into a scalable cloud-based service and you can just use the API endpoint to shoot data at your send data to your ai or machine learning model you get the response in in return so that's kind of then you suddenly have a connector to plug it into any application like a it could be it could be like an app could be some internal data system or rq or flow or whatever back end you're you're using so suddenly it becomes scalable your ai model becomes a scalable service that lives in the cloud, or at least in UbiOps, which can run in the cloud or can run on-premise too.

9:20But making that step suddenly very easy makes it suddenly that as a data science team, you can build and ship AI products and solutions much faster than you could before. Because before there was always this really, especially at larger organizations and companies, It was this really awkward handover between data team and IT teams. Like, hey, we developed those models. We need to run them. Can you help us out running this in the cloud? And then suddenly things start to become really kind of messy in a way. And now suddenly with our platform, you have, it guarantees for the IT team, it guarantees all this stability for like, yeah, you know, it can run on Kubernetes or plug into, you know, like your existing cloud tenant API and the interface to data scientist is very easy to use.

10:12So for them, it suddenly becomes very easy to, you know, create a model catalog, have all your like MLOps capabilities, have other features you need to do like model versioning, manage that, build really great workflows and pipelines out of the different models you have. So suddenly building and shipping AI applications and not just models becomes very easy. And that's our main goal. Very cool. Yeah. And I mean, it sounds like you're doing a fairly good job of that. And so I see a lot of people are quite excited about some of the things you guys are working on. One thing I'd be curious about, just from yourself personally, I've noticed that you've been on the Dev Network Advisory Board for AI and Machine Learning.

10:57and just you know off the top of your head i wonder you know how do you kind of see the role of advisory boards and shaping the future of ai obviously this is a really big new topic a big new field a lot of people are you know wondering how this is going to shape out so i'd love to get your take on that yeah i think it's i think it's great to um especially for like conferences and and events to have input from people from the industry and from the field to kind of know like what are the latest topics uh to you know including conference tracks what are nice things to um uh what are what are kind of emerging topics to maybe include next year and just i think it's great that they're getting that input from you know just like people working in the field like uh startup founders company founders but also just like people working at larger companies like like like meta or google or whatever and and yeah it's it's it's really nice to have all these different opinions to kind of shape what a uh yeah what a what a what are relevant uh items to include in a conference or events very cool yeah no i think you're i think you're spot on with that and i think getting all those different opinions uh especially from the field helps to to make some robust knowledge sharing and solutions and whatnot so very cool um i did notice ubiops offers zero devops solutions i'm wondering if you can explain how this kind of benefits companies, specifically maybe startups?

12:26Yeah, so we noticed that for startups and scale-ups, we are a powerful platform because usually they reach a point where they say like, hey, let's say I'm a startup or scale-up building a product based on AI, right? So let's say my offering is based on, I don't know, like image recognition and I'm building a mobile app for that. And at some point you can, you can build a beta fairly quickly. So you can, you know, like put quite a lot of stuff together and, and build a beta of your product. But it's a very different game to build a product. That's that like 10 ,000 or maybe a hundred thousand or a million people are going to use because then you sell any, the scalable infrastructure, not just for your app, but also for your, for all the AI stuff and all the logic.

13:18And that's what we can give you out of the box. So as soon as you start using UbiOps for deploying your models and integrating them in your application or the product that you're building, suddenly you're not just ready for your beta, but you're also ready for scaling for the next, for the future, basically. and you don't have to you know like like refactor your uh your whole infrastructure and backend it saves a lot of devops and and engineering resources but it also speeds up your time to market so suddenly you can ship products much faster and uh yeah it it it also saves you all the yeah devops overhead and uh that's a that's a big big deal in building scalable production grade systems right now.

14:12Very cool. That's awesome. Yeah. I'm sure that's a massive help to a lot of different people in the space looking for this kind of technology. Something I would, you know, speaking, speaking of something, I'd love to get your thoughts on. Is there any specific companies that you can, you know, use cases or people that have, you know, had success with your product that you could share? Yeah, sure. Yeah. We, we are actually, so we're serving quite a broad range of different companies and different use cases as some are more like enterprises like Bayer CropScience they're building computer vision applications for agriculture which is really cool we have a lot of very cool startups and scale-ups uh among our among our users uh some were working on like deep fake detection which is which is amazing uh like in healthcare and especially healthcare uh like image recognition and um um so that like like image-based applications in healthcare extremely powerful and uh personalized personalized medicine applications so there are tons of great uh yeah Great products being built on Ubiofs right now.

15:35Very cool. That's awesome. Yeah, exciting stuff. A lot of industries that I think you're going to see some big gains from this. I'm wondering, what are your thoughts on the future of AI infrastructure, specifically looking at the context of multi-cloud and on-premise solutions? Yeah, that's a big topic right now. And it's really interesting because there are a couple of drivers that kind of change the way we consume and look at the infrastructure. First, it was, okay, you choose a cloud, right? Like a few years ago, you choose a cloud, you stick with it, and that's kind of where you deploy everything.

16:16But suddenly, especially now with Gen AI and LLMs, now suddenly you can't get GPUs everywhere anymore all the time. So they just sell out in certain regions. and also compliance becomes a topic, especially for enterprises. So like, okay, I want to keep my data and my compute inside my own organization. So suddenly you have all these different kind of demands for like where stuff should run and how you deal with compute. So that really drives the adoption of hybrid cloud and multi-cloud infrastructures. so there are companies that say like yeah i want to you know like use my own infrastructure for uh because maybe you bought some like gpus and have them in your office or in the data center like yeah i want to use that for training but i want to do inference in the cloud because it scales better yeah so then but you want to control it from one uh um one control plane and one interface so we also started um we we recently released uh capabilities to also you know like deploy your models and workloads across like hybrid and multi-cloud multi-cloud architectures which is really cool because then you can slowly lift and shift models from your on-premise data center or like burst training workloads to the cloud or kind of pick what zone you have gpu availability so you can connect different compute environments to the same control plane and use that to optimize for cost, optimize for resource availability, for compliance.

18:00You suddenly have a lot more flexibility in what your compute landscape should look like. And that's a whole different thing than like a few years ago when you just like stick with one, let's say Cloud 10 and then build everything in there. Right. Because it's currently kind of limiting, especially if these are challenges you're facing. Right, right. Yeah, that makes a lot of sense. Something I'd love to hear your opinion on is, you know, where do you see the future of like AI in healthcare? Specifically, I know you have some experience in this space in the past. What are your thoughts on where this goes?

18:39I've seen some incredible innovations, right? AI being used for drug discovery and all sorts of things. Where do you think this all goes? Yeah, I think healthcare is an area where AI has tremendous potential. And you also already see a lot of very innovative applications being built in a healthcare space. I think the most tangible right now has been like in the computer vision image recognition space, you know, like detecting in like pathology or like medical imaging and doing automatic detection there of like, you know, like cancer cells and all those different applications. I think those are pretty, those are actually, yeah, they perform really well.

19:34And I think it's really a maturing field. The tricky thing in healthcare is the data, the sensitivity of the data. Right. And that's been a thing for a long time now. So data is usually, you know, it sits in different silos. It's hard to kind of break them down, especially when you, like for imaging, it's quite straightforward because there's a data stream from a machine and you pass it through an image recognition model which can also run on premise. Especially if you want to combine large bodies of data for more advanced machine learning tasks, let's say, you know, like patient files and things, then it suddenly becomes very tricky, but there's a lot of potential there.

20:15And so I think the challenging part of doing AI in healthcare is on the data side, but there's already, yeah, there's a lot happening right now. I think also, as you say, like drug discovery, drug discovery is a big thing where AI, and there's also a lot of data in those spaces. So I think for the future, I think also make a bridge from the hybrid multi-cloud part. This is also powerful. This can also help in the data or in the healthcare space because suddenly it becomes easier to bring uh your models to where the where the data is instead of pulling the data out of the silos where it's in and you know like migrating that to the cloud because then it becomes suddenly uh yeah kind of sensitive and and and people start kind of pushing back on that that movement because it's because of security and right you know compliance reasons yeah i think there's a but it's it's a it's a very exciting space for ai and i think we're going to see a lot of cool things yeah i mean personally it's one of my areas i'm like the most excited about just you know seeing the ways they speed up drug discovery or you know all sorts of interesting things you're looking at proteins and enzymes and this is crazy stuff and it's it's personalized medicine will be a big thing probably in the next 10 20 years as soon as yeah you can fine-tune these things and also get more information about uh you know like like individuals and have all that data and you know like totally find like what's the best uh what's the best treatment for this patient in this condition yeah and uh that's that's i think really exciting stuff totally i think it's gonna you know we're gonna look back on the time you know give maybe 5 10 15 20 years we'll look back on the time when everyone took the same pill or had the same treatment as like the stone ages right everything's going to be personalized.

22:13You're going to run biometric scans on your body and your DNA and all sorts of things and things that may be like custom to that. So it'd be super exciting. You know, I would love to get from you a piece of advice to aspiring AI experts, machine learning, people going into the field. What's a piece of advice you feel like you could give them? Oh, wow. I think it's changing so fast, right? The whole field. I think compared to five years ago, So I already hear people talking about like machine learning 1.0 and kind of the new era with more like foundation models and generative AI. I do think that a lot of the, let's say 99 % of the business cases and use cases still require like traditional machine learning and statistics.

22:59well i think right now when um that becomes we have like more powerful techniques more powerful hardware uh see a lot of i think there's a lot of potential there so i think it's definitely worthwhile to still you know like understand the the basics of how machine learning works before you i mean to jump into the uh jump into the space uh llms and gen ai and transformer models foundation models it's really it's really exciting it's really exciting stuff and but it's still like it's just very advanced statistics in a way where it's just like you you you increase the amount of parameters in a model and the complexity of the relationships between them but still it's based on the same foundational statistical principles as like the older machine learning stuff so it depends a bit on what you want to what you want to do i still think there's a lot of um right now everyone is looking at uh gen ai and llms and looking like okay you know let's see what we can do with this and how it can be applied but um double kind of that that that's like we're on the top of the hype curve right now so that will that will that will also yeah become a bit more like quiet in a way and then i think it also helps kind of boost the potential for more traditional machine learning and approaches where there's still a lot of ground to cover and a lot of cool business cases to solve.

24:28And actually turn into applications and products to make this world a little better. so um i think as aspiring data scientist um it will be uh yeah it's good to have an idea of like the foundations and principles of how all this stuff works but um yeah eventually when you work at a company it's about solving uh solving a business case in the in the in the right way so So whatever technique works, it's usually the best. But I've seen the past that sometimes just like linear regression can get you quite far. But right now, I think there are tons of exciting opportunities with Transformers and LLMs.

25:27And we haven't really figured out all the applications yet. So there's a lot of exciting stuff to do there. Very cool. So yeah, there's so much to come. You know, so much has already happened. So much more to come. Really exciting. Victor, thank you so much for coming on the AI Chat Podcast, sharing your insights, perspectives, background. It's been a phenomenal time. If people want to, you know, find out a little bit more about what you guys are building at UbiOps, maybe start looking at using some of your tools, implementing them into their tech stack and what they're working on. What's the best way for them to find you?

25:59Yeah, so you can go to ubiops.com and actually create a free account there feel free to feel free to uh to reach out or connect to connect me on linkedin if you have any have any questions or want to chat uh definitely uh definitely open to that so uh yeah just uh just look at our look at our website and connect one that one from our team and uh yeah exciting to get in touch very cool and to the listeners i'll leave a link to ubiops in the show notes so you can go check it out there. But once again, thank you so much for coming on, Victor. To the listener, thank you so much for tuning in to the AI Chat Podcast.

26:38Make sure to rate us wherever you get your podcasts.