Preventing Disasters & Paving the Future of Computing with Roland Groeneveld & Said Ouissal

(Intro Narrator at 00:00:01) Today, we're talking to Roland from OnLogic and Said from Zededa about the ways in which edge computing is pivotal to the advancement of modern technology and Roland and Said's most impactful leadership lessons. Thanks to OnLogic for sponsoring this episode. Go to onlogic.com to learn what they can do for you at the edge. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:29) Hello, guys. How are you?

(Said at 00:00:31) Hey. Hey, Joel.

(Joel Beasley at 00:00:31) Where are you guys calling in from today?

(Said at 00:00:33) Silicon Valley.

(Joel Beasley at 00:00:34) Silicon Valley. And what about you, Roland?

(Roland at 00:00:36) That sounds much more fancy than where I'm at. We're in Vermont.

(Joel Beasley at 00:00:40) Nice. But you're both from the Netherlands, right?

(Roland at 00:00:42) Correct.

(Said at 00:00:43) Yeah.

(Joel Beasley at 00:00:43) Were you friends over in the Netherlands first, or did you meet later?

(Roland at 00:00:48) No, we met really only a couple of years ago. Said actually reached out to me through LinkedIn. And we have a number of common connections and things like that, but we hadn't met each other before. So it was a new connection, and he said, "Hey, we're working on this software, and is there an opportunity to work together here?" So I said, "Let's talk."

(Joel Beasley at 00:01:09) So you guys met through LinkedIn. What was the reason for doing business together? How do you interface with each other?

(Said at 00:01:17) So we've been building an operating system for edge computing, and some of our early customers basically told us about OnLogic. We run on any hardware, but they were telling us how happy they were with the OnLogic hardware and they were using our software on the OnLogic hardware. So then, based on that, I started doing a little bit of research on OnLogic, realized the company color was orange too. So then realized Roland was Dutch, so I just reached out and, typical Dutch, said, "Hey, I think we have some joint customers. I think we could do more together." And happy that Roland responded to that.

(Joel Beasley at 00:01:53) So tell me more about what it is you actually do together.

(Roland at 00:01:57) It really, if you think about it, we make hardware and Zededa makes cloud orchestration software and does a lot more than that. But we really want to focus on the hardware, but our customers tend to want to focus on buying a whole solution. And this is really where partnerships really come in, right? So it gives us the ability to offer the full solution without having to do it all ourselves, and that's really what this is.

(Joel Beasley at 00:02:23) Very cool. So you're the hardware component, and then, Said, you're the software component.

(Said at 00:02:27) Yep. Yeah, we always talk about it as a peanut butter and jelly. Yeah, each of them are separately really good. When you put them together, they're extra good.

(Joel Beasley at 00:02:34) Yeah. And that's what I want. I want to buy the sandwich. I don't want to buy everything and piece it together, right?

(Said at 00:02:39) Yes, that's right. Yeah. And that's one of the reasons we decided to work together. We obviously have been focusing on this new emerging market called edge computing from both a hardware and a software point of view. And as we talk to a lot of these customers, in particular large enterprises, they definitely would love for vendors to work together and simplify the experience. And that led to the partnership between the two companies.

(Joel Beasley at 00:03:02) And I want to get into that. I want to get into exactly what you guys are doing with edge deployments and nerd out a little bit. But before we jump into those details, I was hoping, a lot of the people that listen to the show, they'll listen for tips and advice on how to become a better leader. And so what I like to ask for each of you is, what are you currently learning right now as a leader?

(Roland at 00:03:26) Wow. Deep question. I think you're always learning as a leader. I've actually recently gotten more involved in the day-to-day operations of our engineering team. And what I found is that setting expectations is extremely important, right? And people can't, it's a very basic situation, but people can't live up to expectations if those expectations aren't voiced. And I think that's really, really critically important because once you do, you get a much better rapport with people and everybody's on the same page. It's so basic, but it's something that I'm finding that not all managers do. And then you get into a team, and you're seeing that that really has kind of caused some morale issues and things like that. So that's kind of the flavor of the day for me right now, but it's actually a very important piece of what I'm doing right now.

(Joel Beasley at 00:04:18) Do you have some tactics, like how it actually expresses itself? Like, how do you actually do that?

(Roland at 00:04:25) Yeah. That's a good question. So we have, from the start of the business, we've always really focused on company culture and really early on set our core values and how we operate, right? And one of those core values, well, couple of those core values are open and fair and those really go hand in hand. But open is all really about open feedback and giving people constant, constant feedback. And both ways too, right? Super important. So that's a piece where if you have that sort of baseline in place, it becomes so much easier to give people the feedback, set those expectations, and really give that ongoing feedback cycle.

(Joel Beasley at 00:05:04) Yeah. I found myself with setting expectations, the thing I end up doing the most is just being simple and repetitive with what the expectation is. Because you'll say it once, and as leader, you're like, "It's been said. Move on. It's been said." And then it's like taking a shower. You have to do it all the time.

(Roland at 00:05:24) So true, right? We have a management coach that we work with, and he always tells everyone in the management team, "You've got to repeat something seven times for it to really stick." And it's important to really keep thinking about that because you do think, "I've said this to someone. Why don't they pick this up?" Right? But it's all about repeat, repeat, repeat. And not necessarily in the exact same way, but in different ways and use examples and situations to talk about these things.

(Joel Beasley at 00:05:48) That's why I tell everybody that the fastest path to become a better leader is to have kids. Yeah. You'll repeat yourself a lot. You'll learn a lot. I've got three kids under the age of seven, and man, has that helped my patience, right?

(Said at 00:06:02) Especially patience, yeah, right?

(Roland at 00:06:05) My kids are 18 and 20, and at this point, I don't think they take much feedback anymore.

(Joel Beasley at 00:06:12) That's a good thing, though. You don't want them taking too much feedback at that age. You should have trained your AI model correctly, and it should be operating in the wild.

(Roland at 00:06:19) Exactly. They're out there doing their own thing. Yep. Yeah.

(Joel Beasley at 00:06:23) Said, what are you learning right now as a leader?

(Said at 00:06:25) Well, I would say definitely, to your point on communications, you can always improve and get better in communications, like setting expectations, communicating, following up. Zededa itself, we've gone through a lot of growth last year. About 40% of the company has been on board now less than twelve months. So we're dealing with an influx of a lot of new people. Also a lot of different locations in the world. We're quite an international organization. We just opened up an office in Abu Dhabi. We have offices in India and Europe and other places in the world. So for us, it's also been like not only communications, but how do we make sure the new employees get on board as soon as possible, feel part of the team? How do you balance all the good new things they can bring to the culture, but balancing it also with staying close to some of the core values and what is important for us as a company? And I think it's interesting around the communication because sometimes you have to remind the people that have been on board for a little longer that they can't expect the new people to know it all right away, and it's not always clear. That patience is, I think, important. We have to, especially with the people that have been on board a little longer, explain to them there's more responsibility on them to help their new colleagues to be successful and get them up to speed and answer questions, however basic they may be, and then figuring out how to scale that as we continue to grow the company.

(Joel Beasley at 00:07:40) Okay. So help me with this. You said you've grown like 40% in the past year or so. So let's say you could go back, rewind a year from everything that you've learned with that growth and the mistakes of integrating people into your culture. What's the one piece of advice, if you could have an all-hands meeting, go back in time to one year before you started this rapid growth, what would you tell your people?

(Said at 00:08:03) Well, I told them that we were going to grow, and we're going to add a lot of people, and that it would put stress and everything else in the organization. But probably I should have said it seven times.

(Joel Beasley at 00:08:14) Yes.

(Said at 00:08:15) That's probably it. Just keep repeating it. Because sometimes I go like, "I'm not surprised, and that's part of it." But it's been, it's not been as rocky. It's just something you have to keep in mind. And sometimes, unfortunately, people don't get their honeymoon or their period to get up to speed. You have to just throw them in the deep and say, "Listen, I need you to go and solve this or work with this customer or accelerate that." So that is all part, I think, also of the startup experience.

(Joel Beasley at 00:08:40) Very cool. Good answers. You guys are some strong leaders. I like that. Thank you so much. Yeah. Let's talk about the reality of edge deployments, right? So let's just start real simple. What does it look like in 2025 to get it right at the edge?

(Said at 00:08:57) So we started focusing on edge computing before the industry was called that. So we've been, the company started back in 2016, 2017. And we saw sort of two trends happening that would create this new category of edge computing. One was that the cloud computing model was going to win, not because of the utility of cloud, the rent of resources and all that, but actually it was changing the way how we were building applications in cloud native software development, everything else. So our view was pretty clear then already. It's like anybody that will be building an application moving forward, that's not going to run on a phone or PC, will be building it in cloud native ways. The other trend we saw was internet of things, IoT, more and more things getting connected to the network. These things generate data. You need software to analyze and process that data. And so for us, it was kind of like, okay, there's going to be two ways how this is going to pan out. Either you upload all the data to the cloud where an application makes sense of the data, with more and more AI, by the way, infused in it. Or do we take this application and push it to the edge of the network and we'll be running these cloud native applications in high numbers of remote distributed locations and things like that? So we said that's what's going to happen. You can't upload all the data. It's just impossible, too much data. Networks are not built for uploading data. So instead, people will start putting edge compute hardware, like Roland's OnLogic systems, in more and more discrete locations, and then they're going to want to run these cloud native apps on that. And we basically built an orchestration solution with an operating system to give customers the exact same experience they get from the cloud guys. So APIs in the cloud and a kind of self-service instant experience, but instead of these apps running in a data center, they run on all of these distributed edge environments. I'd say in the early days, edge computing was more a hype than reality. There were a lot of companies talking about edge computing. Frankly, if you were not a cloud company, you were calling yourself an edge company in those days. But I think what we've seen now is that large enterprises have started to adopt it more and more at scale. And it's a couple of things that I think matter a lot for them. I think number one, what matters a lot is the flexibility because every edge compute deployment is unique. You're solving a specific business problem. It's not all SAP on VMware everywhere. It's like you're running on a vessel in the middle of the ocean. We have customers doing that, and they're connecting to containers on the vessel, and they're trying to get data out of the cargo. Or you're deploying in oil and gas sites and you're part of drilling, or you're deploying in solar farms and you're running unmanned solar farms all around the world. So they're all very unique deployments. The second thing we saw was security being a big deal, like zero trust security, things like that, which is one of the things actually that with OnLogic, we've built a really nice integrated solution there around zero trust security. And the last thing is just being able to scale. I mean, a lot of these deployments start small, a few hundred nodes, but we have one customer that has deployed over 15,000 edge compute nodes in dealers and service centers that are running an application stack that's used to diagnose cars when they get dropped off for service. So that's a massive deployment, and every one of them is running Kubernetes and everything else. So how do you scale with the enterprise has been a big part of it.

(Joel Beasley at 00:11:56) Wow. 15,000 units they have out there?

(Said at 00:11:59) Yeah. Pretty much around the world in dealers and service centers for a very large automotive brand.

(Joel Beasley at 00:12:05) And what is edge washing?

(Said at 00:12:08) Well, edge washing is, I think, using edge computing to take an old, maybe more mature market. Let's talk about data centers as an example, and trying to call that edge computing, right? And to me, people that run applications in data centers, they're thinking about how to move them to the cloud. What we do with edge computing is about how we take applications that are in the cloud and push them out to the edge. So I think we're kind of more after instead of before cloud.

(Joel Beasley at 00:12:35) Got it.

(Roland at 00:12:36) We had a not too long ago meeting with a company in our space. And there was a whole discussion around near edge and far edge. And I go like, "I don't even, what, that really doesn't matter, right?" It's not really, why they may try to make a distinction there. But I think instead of, they were going after, what they were looking at is this near edge would be kind of like servers that you have on-premise and far edge being the devices throughout. But it's all a matter of definition. It's like, Said was saying, years ago it was embedded computing, then it became IoT. Intel then coined the word fog computing, right? It was cloud computing and fog computing and that was like a big thing, you know, like the next big thing, fog computing. Now that went nowhere, clearly. Then it became edge computing. So it all kind of is newer names for the same thing over time. Although, having said that, the technology does progress and the software really ties it all back together. And I think that's where what's really interesting is that the combination of cloud and edge combined is where you really get the value out of. And Said always says, "You can't have edge without the cloud," right? And that makes a lot of sense.

(Joel Beasley at 00:13:52) So other than edge, is there any other terms that they're doing that with right now?

(Roland at 00:13:56) I mean, AI, right? You know, it's like AI is everywhere. Yeah. Edge AI.

(Joel Beasley at 00:14:02) Edge AI. There we go. Cloud and edge AI. That should be the most popular episode we ever do.

(Roland at 00:14:08) Yes. That's right. Combine it all.

(Saeed at 00:14:11) I would say that at least the customers that we work with, they don't really call it edge computing. I don't think they wake up in the morning and say, hey, I'm gonna go and do some edge computing. I think it starts really with actual business problems. They're trying to really improve their operations, or they're trying to introduce new capabilities in their business that makes them more competitive or generate new revenues. I mean, those are usually the drivers.

(Saeed at 00:14:36) And then as part of it, a piece of software needs to run outside of the cloud. It just can't run all in the cloud, right? And then that sort of drives, okay, well, we're gonna need hardware, we're gonna need orchestration servers, we're gonna need all these other things in addition. That's typically what we see with our customers, and I think that's why over-rotating on marketing for edge computing, I think it's fine. It explains what you do, but I think it's more important to talk actually about what customers are doing with edge computing and how it's solving their issues. We see that being a lot more valuable.

(Joel Beasley at 00:15:07) Yeah, let's talk about that. I know you guys sent over a few examples. One of them I thought was really interesting is AI computer vision in rail operations. Rail as in trains?

(Saeed at 00:15:18) Yeah, so there's obviously, a couple years ago, as you may know, there were a couple of really large disasters, like East Palestine and others, where trains derailed. They were often caused by mechanical defects on the train. Some part of the train had a mechanical defect they didn't know about. I don't know if you know nowadays, but a freight train can be two to three miles long, which means the engineer in the front has no idea what's happening two, three miles behind them.

(Saeed at 00:15:42) So actually, there's been a lot of pressure also from legislation to improve the things called defect detection systems. They're actually called wayside defect detection systems, which basically are stations on the railway track, where as the train passes, they use cameras and other sensors to see if the train is mechanically okay. So for instance, if a wheel disintegrates, it starts generating lots of sparks before it completely falls apart. So a camera can detect that, say, hey, I'm seeing this wheel not turning normally, but it's kind of being slid over the track and sparks coming off, and can signal to the engineer through software and communications to stop the train because it may have an imminent failure, right?

(Saeed at 00:16:25) So that's an example. And we've been working together, OnLogic and ZEDEDA, with a couple of the largest tier one railway companies in the US to improve their railway defect detection systems. And that uses edge computing.

(Joel Beasley at 00:16:39) Okay, so let's talk more about that. So Roland, I think it's pretty obvious that wherever they have these stations, that's where your computers are gonna sit, your hardware is gonna sit. Is that correct?

(Roland at 00:16:50) Right, right. And I think, you know, then you combine it with some sensors, cameras, those kinds of things, right? Usually these locations also need 5G or 4G connectivity because there's frequently no internet. There's, you know, typically barely power in those locations, right? And power can also be an issue in some of those. But really where ZEDEDA comes in in this situation is, you know, it's very similar to how you would cloud provision an application in the cloud. Now, instead of having to manage each and every one of these computers individually, what happens is as soon as you turn the unit on, it connects to this ZEDEDA cloud orchestration platform, or edge orchestration platform.

(Roland at 00:17:33) And from there on, you can then manage each and every one of those computers. You can double-check the security is exactly the way it's supposed to be. You can do remote updates. You can manage all the data that's on there, things like that. You can add and deinstall applications as needed, those kinds of things.

(Joel Beasley at 00:17:52) And then where does Saeed's technology sit?

(Saeed at 00:17:55) So basically, first of all, building hardware that sits in the middle of high temperature, low temperature for prolonged periods of time and is reliable, it's not easy. So let's start with that. That's what we love also, what OnLogic has been doing in terms of the way the systems are built and the practice behind that. So you need hardware that meets the requirements first, then our software runs on top of it.

(Saeed at 00:18:16) And what we basically built is we took Linux, but we adapted it for edge computing, just like Android took Linux and adapted it for mobile computing. So think about it that way. We made it very easy to run applications on OnLogic hardware, things like security and everything else. And we basically built an API-driven operating system. And then we sell, in addition, a SaaS service that gives customers that cloud experience where they log into a SaaS interface, and then through that, they can deploy, update, monitor the entire stack that runs on all of these locations.

(Saeed at 00:18:48) Because it's not about only the first install. The first install is obviously the hardest, get that system up and running. But then you want to continue to improve the software stack on these systems. This is the difference, I think, between embedded computing, where you kind of build software and you deploy it and you only change the software when you change the hardware. But what we're doing with edge computing is we actually unlock more and more new capabilities and new features and innovations on top of that software stack on the same hardware over a period of time. So for instance, in a railway, maybe you add another sensor which improves the ability to detect certain defects, or you add a camera and you can now monitor a railway crossing if this station happens to be next to a railway crossing. So you can add that in addition to monitoring the train, you can monitor also those types of things.

(Saeed at 00:19:30) So more and more sensory information requires more and more software, and you want to be able to update that dynamically just like you're updating software in the cloud.

(Joel Beasley at 00:19:39) And then as far as the actual technology, so my background is software engineering. So are you doing the infrastructure and orchestration? Are you actually writing the software that's processing the images and data?

(Saeed at 00:19:52) No, we basically do the infrastructure layers. So think of us, the best example I can give is think about what VMware did in a data center, where they gave you the software-defined data center with hypervisor and the management and everything else. But we did that for edge. And the big difference is it runs on a high variety of different hardware because every deployment is unique because of the environmentals, the amount of data you're trying to process, the type of sensors that are connected.

(Saeed at 00:20:16) It all drives very diverse hardware. Once again, the partnership with OnLogic has been great because OnLogic has been able to really build a modular architecture that can rapidly adapt to all of these different deployment targets. And then unlike VMware, which I think was more an IT tool, we're thinking of ourselves more as a cloud-like tool. So giving more things like APIs in the cloud, CI/CD, use Terraform, all the modern cloud tools that people use to manage infrastructure in the cloud. We extend that now to the edge.

(Joel Beasley at 00:20:45) Oh, that's really cool. The most frustrating thing for me, Josh, is when I call support and I can't understand them.

(Intro Narrator at 00:20:51) Yeah, man. I hate that.

(Joel Beasley at 00:20:52) That's why I like US Cloud. Not only is it better, faster support, but all the engineers are US-based engineers, and it's also a lot cheaper. 94% of US Cloud's clients report saving a third or more when switching from Microsoft Unified Support to US Cloud. So now you'll just have to figure out what to do with all that extra money. If it were me, I'm responsible, so I'd reallocate the money to improve my team. Josh, what would you do?

(Intro Narrator at 00:21:17) I'd probably just buy more guitars.

(Joel Beasley at 00:21:19) More guitars. Visit uscloud.com to book a call and figure out how much your team can save. I've gotta ask. I almost cringe at asking, but are you using AI or LLMs or GPTs inside your orchestration?

(Saeed at 00:21:34) We are introducing more and more AI, but we're not trying to do it because it's a buzzword. We're trying to really do it as part of improving the customer experience. One of our models is customer first. So we really want to help customers with that. So one of the things we're doing, for instance, is we've built an analytics engine where if we see a lot of events happening at the same time, we can correlate those.

(Saeed at 00:21:54) Like, let's say you have a piece of hardware and it's rebooting our system, and it's rebooting three times within three hours, and it never normally does that. We then identify it as an anomaly and then can bring that up to our customers and say, hey, we've had these reboots. We're seeing this happening more widespread since the software update or whatever. There may be a problem in the software or something else we have to investigate further. So that's how we think about AI more from our product, improving the experience. We put all documentation in AI on an agent, so you don't have to go through the table of contents and find what you want. You just tell us what you want. I want to install the OS on a hardware. Here's the description how to do that. So we're trying to use AI in an intelligent way.

(Saeed at 00:22:34) Now many of our customers use us to deploy AI at the edge. And so Roland and myself, we've been really talking a lot about how do we see this evolution and how do we deliver more and more accelerated compute that makes it easier to run more complex models with vision and other things, including gen AI, with customers that are trying to replace the entire HMI, the human-machine interface of a machine, with a voice gen AI interface. They can talk to the machine just like Star Trek. Those are pretty cool projects. They're early, but I think they're starting to push the boundaries of the possibilities here.

(Joel Beasley at 00:23:07) And then at the same time, that technology is just advancing at an incredible speed. I can't even keep up with all the stuff on X. Every week there's some massive, there's this new way of doing things. And I think one just came out, and Josh can correct me, but it was called MCM or something. But basically the idea was that it was this framework for connecting datasets. Have you looked at that, Saeed?

(Saeed at 00:23:33) Yeah, it's the MCP. It's a way for agents to talk to each other, which is sort of pushing this agentic AI thinking. It's really exciting. And I think, again, we will see more and more AI agents being deployed all across the edge and the cloud. It won't all run in the cloud. It won't all run at the edge. But yeah, we're in a very interesting time where there's so much great innovation happening and people building off each other's ideas. It's really a great time in tech, that's for sure.

(Joel Beasley at 00:23:58) And then I want to know what you're doing with the predictive maintenance and solar tracking.

(Roland at 00:24:04) So a lot of the time, right, in our field, customers need a device that's in a very remote application. Solar tracking is one of those examples, but same for any sort of energy environment, whether it's oil and gas or wind or solar, all those areas. You're talking about very remote deployments, but also very difficult environments.

(Roland at 00:24:27) So for us, that's where we have these systems that can withstand extreme temperatures, shock, vibration, those kinds of things. And more and more are we adding an AI accelerator or GPU or an NPU to those systems to then, you know, because more and more software, AI software requires more and more powerful hardware really at this point. So that's kind of what we're working on. And from there on, it's through the AI software that's loaded on there, you can actually then do all these predictive maintenance kinds of things. So you can, you know, for solar farms, it's a great example, right? You want to make sure that every panel is functioning as expected. And if one panel is out, you can actually detect that based on performance, right? You can compare the performance of one string of panels to another string of panels, and all of a sudden you see a difference. You see a delta happening, and this is where AI automatically picks out, like, here's where a problem is, and pinpoints it and sends data back to a central location, and a technician can come out and come fix it.

(Joel Beasley at 00:25:33) That's pretty cool. There's a lot of really interesting use cases.

(Saeed at 00:25:37) Yeah, especially when the cost of sending somebody on-site is higher than the actual cost of the repairs or replacement.

(Joel Beasley at 00:25:45) Mm-hmm.

(Saeed at 00:25:46) So with predictive, you can actually start using failure patterns that happen before whatever panel fails. We had a customer in wind farms, same problem. Wind farms are even harder because you need a crane to go all the way up there. So they were really trying to say, okay, if we can predict failures, if we know that a solar panel or wind farm or whatever the asset is starts exhibiting these types of things that we can detect with a sensor, and we can go in and fix three or four panels at the same time, or three or four wind turbines at the same time, we significantly reduce our repair costs.

(Saeed at 00:26:18) And number two is reduce our downtime, right? And those are the two things that improve ultimately your bottom line, right? So that's kind of what we see a lot of people trying to use predictive analytics for and predictive AI.

(Joel Beasley at 00:26:31) Yeah, I can believe that with the cost. I talked with somebody a while back, and they were saying that they would spend like a hundred K to have a part for their mining vehicle. It was in Australia, and they mined ore or something out of the earth. But this one part broke, and it was like a 10 or 20 gram part, but they would have to spend like a hundred K to helicopter it in within 24 hours because the amount of money they would lose from that thing not mining the earth's ore was just so staggering. And I was like, wow, that's crazy. I never think about that.

(Roland at 00:27:05) Again, mining is an example where our computers are used a lot too, right? You end up in these really, really tricky environments. A lot of dust, a lot of shock, vibration, those kinds of things. And a lot of predictive maintenance on the machinery, absolutely. It's that whole piece.

(Joel Beasley at 00:27:22) Fusing the edge and the cloud, right? So where do they exactly come together? Is it something we want to do? Is it something that's happening?

(Saeed at 00:27:35) Yeah, I mean, I don't think you can do one without the other. And frankly, you know, if people want to do edge without cloud or not doing cloud, it usually tells me they want to kind of stay in the old world. They're not really ready for the new world. And the reason is because, you know, you still need to store data somewhere. And the edge is not the best place to store your data. You don't want to have a lot of little storage places for your data. You want to have, ideally, a central place where all your data gets stored, your historical data. You also need that data if you're doing things like training, you know, training models. You need lots of amounts of data, so ideally centralized. So I think of it as a pipeline.

(Saeed at 00:28:09) I think of it like, okay, you got a sensor. It generates, let's say, a video feed. You could upload that video feed to the cloud and analyze that video feed to detect what's happening, detect objects or whatever, which is quite costly, especially if you have a lot of video feeds. Or you put edge servers, edge computers locally, and you process the video feed there. And then what the AI model detects in the video feed, you send still that to the cloud. So you're part of a data pipeline.

(Saeed at 00:28:34) And those you store in a nice database. You say, I saw object one at this time and this camera, object two at that time and the other camera. And that becomes sort of your place where you can visualize that information. You can share that information within your organization. You can use that to train on that information. That makes a lot of sense, but you still need edge. So I think of it as a data reduction kind of pipeline, where the majority of the data gets processed at the edge, and then the useful bits get sent back to the cloud and stored there.

(Joel Beasley at 00:29:02) And I've gotta ask, what does the name of your company mean? Zededa. Like, what does that mean?

(Saeed at 00:29:09) Yeah, it's a great question. So, you know, as you kind of alluded earlier, both Roland and I are from the Netherlands. I was born and raised in the Netherlands, but my family is actually Moroccan. So my roots are Moroccan.

(Saeed at 00:29:19) And Zededa is a Moroccan word. It means new or innovative. And then we chose the orange color to reflect the Dutch part of the heritage too.

(Joel Beasley at 00:29:27) All right.

(Saeed at 00:29:28) Yeah, there we go. There's actually a city in Morocco called Zededa. So, you know—

(Joel Beasley at 00:29:32) Is there?

(Saeed at 00:29:32) Yeah. It's spelled with a J, not a Z, but it's pronounced the same way.

(Joel Beasley at 00:29:36) You've got a city called OnLogic in Vermont there?

(Roland at 00:29:38) We know, we know.

(Saeed at 00:29:39) But I'm—

(Roland at 00:29:40) I think it's about time.

(Joel Beasley at 00:29:41) It's about time. Yeah. So what's the future look like with edge? I know it's a broad question, but I'm just curious. Like, we've been talking about edge this whole time. What's the future look like?

(Roland at 00:29:53) Yeah, I think more and more AI. Right? I mean, that's really what you're gonna see. More automated systems that are just more effective. A lot of the time, it's fault detection, you know, just quality control, automated predictive maintenance, those kinds of things. And there's just so many areas where you can deploy that. Right? So if you think about it, it really is not so much about—there's really no limits if you think about it. Right? So if you think about the factory floor where the company is producing their widgets, whatever they are, it's still in most factories very, very, very manual. Right? And there's so many things where you can start automating those things and do all this quality control and predictive maintenance as part of that as well. I think that's where there's a large amount of growth. We see in our customer base factory automation, warehouse automation is big. You know, solar, wind farms, those kinds of things. A lot of those areas, energy in general, lots of opportunities there as well.

(Joel Beasley at 00:31:00) Saeed, what do you think the future is?

(Saeed at 00:31:02) Yeah, no, I think just building on what Roland just said. I mean, we think it's AI, and we kind of call it also physical AI. So today, it's a lot about the predictive analytics and processing data, but we're gonna move more into control as well. Right? So controlling drones, robot arms, machines. And, you know, if you think about what's happening right now with the humanoid robots that Tesla and Figure and all these companies are building, that's actually a form of edge computing at the end, and there's going to be a lot more needed there. So I think, just to build on Roland, it's AI in many different shapes, regardless of generative, physical, predictive, vision AI. And I think more and more organizations are going to try to figure out, like, how can we build an architecture that supports all of these if we're an enterprise and we manufacture, we do transportation, we have logistics, we have all these other pieces. How can we create a horizontal edge platform, both hardware, software, to support the business needs and continue this autonomy and automation everywhere?

(Roland at 00:31:59) And what also comes in there is security. Right? So historically, there's this concept of OT and IT. Right? Operations technology and information technology. Our systems are Greengrass certified, so they are similar kind to Zededa. They're ready to go. And as a matter of fact, it works with Zededa as well. It's an ingredient into the whole chain as well. AWS is a partner of ours. We've been working with AWS for a long time, so you can probably explain a lot more about that.

(Saeed at 00:32:26) Yeah, I mean, the cloud guys, obviously, they're trying to make edge happen, and they're providing tools to customers to run software on the edge to take data back to the cloud because, ultimately, they care about more and more cloud consumption. So Greengrass is a great example. We integrate with it. We integrate with all alternatives of that from Azure and other companies, clouds. Yeah. And I think AWS has been a great partner. We work with AWS, work with Azure, work with all the cloud companies really, really well. And it's great to see sort of how the cloud guys are embracing edge.

(Roland at 00:32:54) So what we've seen is that historically in a lot of factories and warehousing and stuff like that, you had the operations team kind of run their own little IT space. Right? It's separate from the IT team who's running the office space and the cloud and things like that. That is no longer acceptable for a lot of CTOs. Right? They need control over all these devices because you don't know anymore what's in your network, and that's a security risk. So what's happening now more and more is a real focus on security. Right? Security from the start. Right? Security from literally the supply chain, what components are in those computers, the firmware, the operating system, the application level, the connections between the cloud and the edge. All those things are super important as well. And I think that's, you know, as much as there's the opportunities, there's also the risk side of things to keep in mind here.

(Joel Beasley at 00:33:47) Yeah. You don't want your coffee maker having a web server.

(Roland at 00:33:51) And guess what? A lot of them do, actually.

(Joel Beasley at 00:33:53) I know, I know.

(Saeed at 00:33:55) Yeah. Yeah. Well, or worse, if you start deploying in environments where it's mission critical, like you're controlling, you know, physical things, you could really create harm and damage. Like, I mean, one of the reasons I talked a little bit about this automotive customer. Their biggest concern was software updates to the car getting compromised from the developer to building a new version of some software that runs in the car and then getting that update into the car. There's many ways I can deliver an update to a car, and they want to make sure that entire chain was extremely secure. Because imagine somebody introducing some kind of a backdoor in a software update, and millions of cars suddenly start driving around with, and somebody executes the backdoor, it could cause a lot of harm. So I think people are really worried that as you put more software everywhere and all those things are connected to the network, it just opens up a whole new slew of challenges that we didn't have before.

(Joel Beasley at 00:34:47) I know. I feel like such a dangerous person when I'm using the full self-driving Tesla. It's like at any moment, we could get hacked.

(Saeed at 00:34:55) Yes. Yeah. Those are the challenges. And this is gonna become more and more as we build more and more heavy machinery or big machines that are running AI and run lots of software. It could be in a factory. It could be in a chemical plant. It could be, you know, self-driving buses, trains. I mean, there's so much more happening right now where more and more is getting assisted and more and more is getting supported with AI. The security is, you know, top of mind for everybody.

(Joel Beasley at 00:35:19) Roland, have you guys done any on-site hardware that's specifically designed for inference?

(Roland at 00:35:27) Yeah, we do. Actually, we just released a new edge server that is specifically for inference that we call the Karbon 300. Karbon is our server line, our edge server line. And it can hold up to—well, depending on the type of graphics card, but it can hold up to seven different graphics cards in there. And so you end up with this GPU acceleration that is pretty amazing. That combined with a Xeon server chip, and at that point you can really do some pretty great stuff. So we're doing more and more on that side, but it's really to kind of support our customers that do both sort of both sides of that. Right? They're really looking at the AI in the field and sort of the opportunity to really have the training and inferencing in sort of really high volume and high capacity going as well.

(Joel Beasley at 00:36:22) When you're—let's take the GPU specifically that you're just talking about. Do the GPU manufacturers, do they make hardened versions of their GPUs so that you as a manufacturer can take and put that into your existing casing, or do they just sell the same thing everywhere? They don't make a hardened version, then you have to figure out how to make the casing so good it protects it.

(Roland at 00:36:44) Yeah, it's a combination of hardened and life cycle. A lot—so the big problem in our industry and GPUs in general. Right? Life cycles are pretty short. You might be able to buy a product, a certain GPU, for a year or a year and a half, maybe two years if you're lucky. But frequently, our customers are looking at a deployment over a number of years. So you want a product that's available for a large period of time. So you end up with—we have specific long life GPUs that are available, but also GPUs that are meant for sort of more extreme temperatures and shock or vibration, those kinds of things. You know, and then from there on we can also do much more custom solutions that focus on, yeah, more integrated. So it's not necessarily a PCIe card at that point. It becomes a GPU right mounted onto a board, like MXM or other platforms.

(Joel Beasley at 00:37:36) And because the environments are so different, I mean, that's why you exist in the marketplace. There's not just one solution you can just buy. Everyone has to go, here's the specific environment, the length of time I can maintain. Like, you have to give all the details, and then you, with your experience and your teams, you figure out how to get them a solution. Is that right?

(Roland at 00:37:55) Yeah. Every one of our systems is effectively custom. Right? You can go on a website right now and order a system, and those are highly configurable to begin with. And a lot of customers do. But usually in the large deployments, it's even more specific to that particular customer, whether it be the firmware that we load on there, the software we preimage, but also, you know, frequently it's a highly specific add-on cards or completely custom designed that we designed from scratch for that customer.

(Joel Beasley at 00:38:27) So you can put Saeed's software on there if it's for that customer.

(Roland at 00:38:30) Saeed's software will run on anything. And but it is really all about—the beauty of what we're doing is we preload that software together, but also Saeed's team, Zededa, will actually go and validate the hardware with their software and make sure it all works really well. So when you buy an OnLogic computer that is validated with Zededa, it just is certified. It just works. And I mean, that's kind of the beauty of that. It's just—you don't have to worry about it. It's ready to go.

(Joel Beasley at 00:39:03) And then do you handle, like, I'll call it last mile. Do you handle the actual install, like, plugging it in up at the windmill, or is that all on the client?

(Roland at 00:39:12) We don't. We leave that to clients. And usually, a lot of our clients have either a large install team themselves. They have their own IT or people that actually manage those installs. Or we work with, you know, systems integrators that do these kinds of things. We really focus more on the design and the, you know, the hard stuff, the engineering. The install is relatively—it should be relatively straightforward. I think that's also the other side of this. Right? So the install is so easy because there's nothing to configure. You plug in the box and you turn it on. And then from there on, Zededa takes it from there, basically. And you don't—nobody needs to—you don't need a technician to go and set it up. It's—all of a sudden, you know, you plug it in and it shows up in the Zededa platform. And from there on, whoever's managing that can actually install additional packages and updates and things like that.

(Joel Beasley at 00:40:09) So now no guys climbing the tower to enter in a Windows 95 key.

(Roland at 00:40:13) Correct. That's exactly what we're trying to do. Right? We're trying to make it easy and very predictable and very manageable. So it's also from a, you know—let's just say for a second that—oh, great example. We just worked with a customer who has our computers in power plants, and they were doing a test with a shock conversion to make sure we can withstand earthquakes. Yeah. The computer, everything is flying around. The OnLogic computer was actually fine, but the mechanical pieces failed and it could be—the fluid on the ground got physically—the computer was still working but physically damaged on the outside. At that point, all you need to do is send a new computer, turn it back on, and it works. You don't need to actually provision it anymore at that point. That's the beauty of that.

(Joel Beasley at 00:41:00) How did they test that?

(Roland at 00:41:01) There are very specific companies that do these tests.

(Joel Beasley at 00:41:05) Jumping on the bed.

(Roland at 00:41:06) Yeah. We—yeah. As part of, you know, our systems, you gotta go through all these certifications. Right? You go to the CE and FCC and UL and those kinds of things, but then you go beyond that for, you know, there's these earthquake tests. So you—and this is really meant for, think about it. If you have a data center and you mount, you know, a server in a data center, you wanna make sure that that server can withstand and the data center can withstand certain amounts of earthquakes in a situation like that. And so all those data centers actually are designed to withstand typically nine and a half Gs or so because that's kind of the highest earthquake that's ever been recorded. And so apparently our customer was testing this at something like 40 Gs, which is just a little more extreme than the unit. So it looks like—

(Joel Beasley at 00:41:51) They're testing at 40 and the highest recorded is eight or nine?

(Roland at 00:41:54) Eight or, yeah, nine and a half is the highest recorded. Yeah. All right.

(Joel Beasley at 00:41:57) I mean, look. I'm a pretty conservative person. I'd go 15 if I was being extra crazy. 40.

(Roland at 00:42:03) They did a—they explained to us. They did a little bit of a calculation issue, and they went and went a little overboard in the test.

(Joel Beasley at 00:42:11) What I'm saying, Roland, is I wanna strap a hand grenade to it, and I want it to still work.

(Roland at 00:42:16) But here's the crazy part. In this case, right, at this very high—I think it was 40 Gs. It fell on the ground, you know, and then there was a display and a number of other parts that actually broke. There was no other part, but the computer just worked. It was not—did not fail.

(Joel Beasley at 00:42:33) There you go.

(Roland at 00:42:34) I love—

(Joel Beasley at 00:42:34) They should just make the spaceship out of that.

(Roland at 00:42:36) They might just. Let's put it that way.

(Joel Beasley at 00:42:38) So closing thoughts, a couple rapid fire type questions. If you could give one tip and one tip only to the tech leaders listening for them to nail their edge deployment strategy, what would that tip be?

(Roland at 00:42:52) I would say get going. It's that simple. Right? For one, I, you know, I think it's important not to completely overthink the whole strategy. Start small and build it out over time.

(Roland at 00:43:05) But if you don't get going, you're going to be left behind because it's everywhere.

(Saeed at 00:43:10) Yeah, totally agree with Roland. And I think the other thing is, you know, surround yourself with partners that have experience in this. Leverage the collective wisdom out there. We actually created, along with other folks, a community called Edge Monsters, which is an edge architect's community where people can get together and exchange learnings from different verticals, different industries.

(Saeed at 00:43:30) And there's many more like these, but I think there's already a lot of edge happening. Learn from those folks that have already been a little bit longer working on it. Avoid making the same mistakes and surround yourself with partners that can help connect you to other people.

(Joel Beasley at 00:43:44) And last question here. I'm going to go Roland, then Saeed. Roland, if I gave you a time machine and you could go back in time to that first moment where you and Lisa spent that 80k on that vendor you weren't sure about to start this company, what's one piece of advice you'd give your past self?

(Roland at 00:44:02) Stick with it. You know, that's it, right? There's so many moments where you kind of question like, hey, is this really going anywhere? Is this the right approach? Are we doing the right thing? And you're going to make mistakes. It is what it is. But as long as you make more good decisions than bad decisions, you move ahead, right? That's how I feel it.

(Joel Beasley at 00:44:23) Saeed, same thing. Time machine, you get to go back to when you founded the company. What advice? And you're the founder, correct? I'm just assuming.

(Saeed at 00:44:30) Yeah, with a few other folks, we started it. I mean, I would say similar to Roland, worry less. Accept mistakes will happen. And I think to Roland's point, you just got to—there's no good or bad mistakes. There's making decisions and then making them work, right? And I think if you can operate with that, you get really far.

(Joel Beasley at 00:44:47) No hesitation. Just take the action. Get it done. If it doesn't—once you make a decision, if it doesn't work, you can just make another decision.

(Saeed at 00:44:53) That's right. Yeah, I think Jeff Bezos has this thing called one-door versus two-door decisions, and we try to apply that a lot in Zededa too, because you can overthink. An earlier point about edge computing: get going. You know, it's not a one-way door. I think similarly, you just kind of have to understand, and especially one-way door decisions, just go. You know? Just make them and get going.

(Joel Beasley at 00:45:15) Well, this has been an absolute pleasure.

(Saeed at 00:45:17) Great to meet you, Joel, and I really love your podcast. So I'm really looking forward to this episode coming out.

(Joel Beasley at 00:45:22) Yeah, and then tell Lisa I say hi, Roland.

(Roland at 00:45:24) I will absolutely do that. Yeah, see you right around the corner, I'm sure.

(Joel Beasley at 00:45:29) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.