Beyond the Blue Dot: How AI Gives Location Data Meaning

Transcript

Nick Earle: Hi, this is Nick Earle, host of IoT and AI Leaders, and today we’re talking location, location, location.

Actually, that’s not quite true. We’re gonna be talking about location, context awareness, abstracting semantic meaning, and business cases that you can then create, and new opportunities. We’re gonna be talking about the technology, we’re gonna be talking about use cases, including one that’s to do with, getting cows pregnant, so, get ready for that one.

And, the main opportunity being the factory floor, and then the, future of manufacturing, in an IoT and AI world. My guest is, Samuel Van de Velde, who is the CTO and founder of Pozyx. He’s based in Belgium, and he is a real visionary actually in terms of where [00:01:00] all of this is going. And towards the end, he talks about the future of software and software being created on demand, almost real time by users based on what they need to know right now, and how that he believes that will happen and how we’re getting closer to, doing that.

And the huge opportunity there is for factory optimization, starting off from a basis of indoor location tracking. so there’s a lot in this one, and he explains it, really, really well. I think you’re really gonna enjoy it. So, without further ado, I will hand over to my, pod, with, Samuel Van de Velde, the CTO and the founder of Pozyx.

 Samuel, welcome to the IoT and AI Leaders Podcast Good to see you. why don’t you, start things off by just explaining to our listeners, and indeed our viewers, some people watch on YouTube, a little bit about your background [00:02:00] and how you came to found Pozyx?

Samuel Van de Velde: Yes. Uh, first of all, thank you for having me.

it’s a pleasure. so my name is Samuel Van de Velde. I am the founder and the CTO of Pozyx. And the company, was founded about 10 to 11 years ago as a spinoff of, Ghent University. So I myself, I’m an engineer as a background. I did my PhD, as an electrical engineer, studying things, related to indoor positioning, all different types of technologies.

And based on that, ultimately I founded the company, and the reason for that was that, in the academic world, it seemed as if positioning was a solved problem, indoor positioning. But then when I looked outside, I didn’t really see a lot of, or even any application of it. so why was that?

Why wasn’t anybody using it yet? And that’s why I founded the company, because I really wanted to have indoor positioning, yeah, be used in, in many different, [00:03:00] environments.

Nick Earle: You know, that’s interesting that, that you say that. Um, you know, as obviously I’ve been running an IoT company for about eight, nine years, and you’re absolutely right in what you say.

when I first came to SI, I assumed positioning, whether it be indoor or outdoor, but just in general, positioning would be 60, 70% of the use cases. Actually, it’s probably less than 10%, or maybe even 5%, and it, it doesn’t seem right. And it… And it’s mainly people tracking vehicles moving and things like that.

But indoor positioning in particular, and I know the technology, and we’ll talk about, about the technology. The technology’s been developing, particularly with tags and sensors and, and, and, and whatever. But yeah, it, it is not, it has not taken off, and I think that’s what we’re going to be talking about, is, is how you guys do a lot more than just the dot on the map, right?[00:04:00]

I mean, that’s the way I, I look at you. It’s not just about the, the dot on the map, like the Find My app on your iPhone or something.

Samuel Van de Velde: Exactly. And I think the, the Find My app really helped, indoor positioning a lot. But still, up to today, I think there is much more that positioning can offer and bring to the table than just, finding something back.

Nick Earle: Okay.

Samuel Van de Velde: I- And ho- hopefully I can give some examples of that

Nick Earle: Yeah.we’re gonna get into them. I know you got a couple of, of, of, of great ones we’re gonna come to. But before, let’s get into the technology. We have a technical and business, audience. So can you just talk about, just in, in general, what technology you use, and, why, why accuracy, matters so much.

So why does accuracy matter so much? And I know you use ultra-wideband, so just talk a little bit about why you took that approach.

Samuel Van de Velde: Yeah. So, well, as a company, we provide, different technologies, and it’s also true that not every technology is always the right technology for every use case. that being [00:05:00]said, the, the main technology where we started from was ultra-wideband technology, and that’s considered a little bit the golden standard for indoor positioning because it provides very high accuracy, about 10 to 30 centimeters, and also because it is very, reliable and it, can handle very challenging environments with a lot of metal and obstructions and reflections.

if you compare that to, for example, Bluetooth, which is also a, a popular, positioning technology which we also provide, there the accuracy is more around five meters, Yeah … or even 10 meters, and it can suffer a bit more from, from more challenging environments in industrial environments.

Nick Earle: Yes.

Samuel Van de Velde: each technology has its merit.

Nick Earle: Yeah …

Samuel Van de Velde: There’s not a single technology that, that wins in, in every single environment. So that’s why also from our company, we approach it based from a use case, and then we see, okay, what technology do you require? And also we, we try to bring these technologies a little [00:06:00] bit together, that you’re not isolated into one technology and then you’re stuck with that,

For example, you have tags, the trackers themselves that maybe have multiple technologies inside such that they can seamlessly move from different areas where you have potentially different, requirements in terms of accuracy and update rate and things like that.

Nick Earle: and before we get onto your first, use case, which is a great one, and a spoiler alert, we’re gonna be talking about cows.

but, um, back to the point about, you know, the dot on the, the dot on the graph is not, is not enough. if you can, get more context and more rich information, what does become, possible when you get that accuracy down to the… You know, you talked about 10 to 30 centimeters of accuracy. So, so what…

If, if you can do that using whatever technology is the right one, as you said, for the use case, what then becomes possible other than just pure location?

Samuel Van de Velde: Yes. So, well, [00:07:00] outdoor with GP- GPS, well, uh, the outdoor world is much bigger. houses are, are much further from apart, from, from each other. Indoor, everything is, well, smaller.

A room might be five meters, by five meters, or even an industrial hall might be big, but still, the level of detail is, is, so much more bigger. So with, other technologies like Wi-Fi, you might simply know that you’re on one side of the building, which doesn’t really bring a whole lot of value, especially if you’re looking for something.

ultra-wideband is, on the other hand, there it’s extremely accurate, but that, allows you to also do automations. Because now, because it’s very accurate, you can detect if, an asset has moved inside a certain area, even maybe inside a bin area where, where you’re, putting area, where you’re putting your material.

we can also better detect movements, and that can allow you to see, okay, am I actually using, this asset, this material? Imagine it’s [00:08:00] tools or forklifts. what we often see is they’re just standing still or they’re, they’re not moving all that much. and that is something that you can only do if you have that level of accuracy.

If you just know it’s somewhere in that corner, then you can also not know if it’s moving or not, or if it’s… how long it’s been there, if somebody just picked it up and things like that.

Nick Earle: So, so let’s talk about that first case study, because y- you talked there about assets, assets that move, and, into different parts of a, let’s say, an indoor, indoor building.

And, and, and actually, that is certainly true when it comes to cows. They wander, they wander all over the place, and they go to certain places inside the, the sheds, or barns. I know you’ve got, you’ve got a lot of manufacturing case studies, and we’re going to come to that. But let’s start off with the, what I would call the fun one, the, the, cow one.

So can you just talk our listeners through, that, that one case study of how you can bring context? It’s not just about [00:09:00] knowing where the cow is, right? It’s… Your value is more than that, right?

Samuel Van de Velde: Yes, exactly. And so several years ago, a large dairy company came to us and said, “Okay, we, we want to track cows.”

Uh, we didn’t know anything about the dairy industry, so we said, “Oh, sure, we can track cows.” Yeah,

Nick Earle: easy.

Samuel Van de Velde: so we, we worked together with them to build an ear tag that you can attach to a cow’s ear. And so, well, what you might imagine first is, okay, why would you need to find a cow? and, and it does happen sometimes, but it is a smaller use case, and it’s a, a smaller reason why people would invest in such a system.

for dairy farms with more than 150 animals- There, at that point, there’s a tipping point where the farmer doesn’t know every cow by name anymore, can’t pick it out from the herd. So defining the cow is still a valuable thing there. But the real, the absolute value [00:10:00] is in the analysis that you can do, with the cow.

So a little bit naive, but when, when we started with, with this, this application, I didn’t know that dairy cows need to become pregnant in order to give milk.

Nick Earle: Yeah.

Samuel Van de Velde: there’s a whole thing, a whole science around getting the cows pregnant because you want to have that, as quickly as possible,

so that the production of the milk is, is, You maximize the

Nick Earle: milk yield.

Samuel Van de Velde: And so finding the right moment that a cow, can get pregnant, pregnant is not so, not so easy. And so because you have to send a veterinarian, he has to inseminate the cow, it costs a lot of money. And so if you’re wrong about that timing, you’ve lost a lot of money, a lot of time.

And so it turns out that location data is extremely well, to determine if a cow is ready to be pregnant or not. And, and how can it do that? So, one of the things, is that, well, maybe a cow just, gets a bit more frivolous, and starts moving.

Nick Earle: we’re drifting into [00:11:00] a very bizarre conversation.

Samuel Van de Velde: Yeah.

Nick Earle: But we’re all learning a lot.

Samuel Van de Velde: Yeah. But it’s, it’s, it’s primarily about, activities. And so imagine a barn, and the cow is moving around, and you have an, an eating area, you have a cubicle where it’s resting. You have a drinking area, a milking area. All those areas, well, we, we draw geo fences around that, and by having accurate location, we can accurately know, is the cow in that area or not?

And for example, related to eating, it turns out that, well, if a cow is ready to, to be inseminated, it might eat more frequent but for a shorter duration of time. And that’s something that you can very accurately see with this location data, not by the raw X and Y coordinates, but more about, okay, how much time is it spending in this area which has some kind of a semantic meaning.

And so bringing that all together, that allows you to very accurately determine, yeah, the, not [00:12:00] only is the cow ready to be inseminated, but also a lot about the cow’s health. of course, this is done by our partner that, that, that then, runs AI models on top of the data that is coming out of, out of the location.

But basically, it’s about, yeah, knowing how much time it’s spending in certain areas. And I think it’s, it’s a, a super powerful and, and interesting, use case.

Nick Earle: When we did our, prep call, you’ll know that, but for the listeners, when I heard that, case study, I said, “Oh, well, that’s, interesting,” because one of the previous, guests we’d had on the pod, is a couple of veterinarians, over in the US who

have got a similar but different use case for, cattle. and just to, there is a part for those of you who are interested in this, and who knows, we may well have, people who are in the, cattle or meat or dairy, business. But this is to do with the biggest killer of, cows, which is bovine respiratory disease.

[00:13:00] it kills something like 13% of all cows. And it’s, by the time you detect it, it’s pretty much too late. the cow has got… It’s basically cow pneumonia. and it’s the same thing, is that when you talk to vets, you know, the technologist says, “Well, there’s no way IoT can spot a disease in a cow.” But actually it’s because of their behavior.

It’s the context not the dot, as we talked about. And apparently when cows get it, they, the first thing they do is they go to the middle of the herd for protection because they don’t feel very well And then when they really, really don’t feel well and they’re, they’ve probably got a few days or, or, or weeks to live, they start to lose weight a lot, so you lose the meat yield.

But then they die. They actually then pull away from the herd, and of course, they eat less, they drink less, and there are actually some very predictable graphs that show where they are. And the point about it was that these guys came on, and we did the same thing. We created tags. we were using [00:14:00] Bluetooth tags, and then we used, cellular to backhaul the data from remote farms in the US, which are huge.

But, they actually were able to, The bottom line on their trials, literally field trials, the bottom line is they were able to identify disease seven days earlier through pattern matching, and those seven days would allow the veterinarian to be called out and, and inject the cow and save the cow.

They, they calculated it something like a, globally, an eight billion dollar a year opportunity in terms of just keeping those cows alive and getting more meat out of it. But a very difficult problem because, as I’m sure you found out, i-it’s one thing doing trials on humans and…

or people walking around hospitals and, and we’re gonna talk about the factory floor. It’s another thing trying to get five hundred cattle, to behave properly and stand still while they fit the tags. I mean, it’s the little stuff like that that makes it [00:15:00] difficult. But they are actually in full field trials now, with using pattern recognition to, same thing as your case study but outdoors, to actually reduce BRD, the number one killer of, of cows in the world.

And so it’s fascinating that we both have independently got a very, very similar, case study for, cows, not things or people

Samuel Van de Velde: But to me, it’s, uh, the example or the use case is, is exemplic because, uh, I, I think you could apply the same, uh, approach to other use cases. Yeah, exactly. Exactly. For, for example, and we’re not active in that, in that industry, but in the, in the healthcare or the elderly care, you could do exactly the same thing to track, the elderly people in, in, a retirement house to see, okay, are they still moving?

Are they still cooking their own, dish? Exactly. Are they going to the toilet, sufficiently? Did

Nick Earle: they come back? Did they go to the toilet in the middle of the night but not come back? Yeah. Maybe they’re on the bathroom floor. Ex- exactly.

Samuel Van de Velde: Yeah.

Nick Earle: that’s where the use cases are with tracking.

it’s not [00:16:00] they’re trying to find out where the car is in the parking lot.

Samuel Van de Velde: And, and I often compare this also to the big tech players like the Googles, because Google, they have Google Maps, they already track you, and there’s just this immense amount of data in doing so. They do this on a, a higher level, so more, for example, they track on which, website you go or they could even track, in which, shop you go.

But it’s, it’s a bit the same. You have your, your digital location, which is a website, or your physical location in which store. But then you… They give semantic meaning to this because they know it’s a sor- store about this type of, goods, and then maybe, well, you like these type of goods, and then we can sell this more to you.

So, but I’m convinced that, that with location, you can actually, in the future, may do a lot more. we don’t do this, and it’s, it’s at the border of, of, of ethics, but I’m convinced that with accurate location data, you could estimate a person’s gender, maybe the age. I’m, I’m convinced that that’s [00:17:00]possible.

We don’t do that. we, we focus on industrial applications. But there’s just so much, value- Yeah … and power in, location data, especially-

Nick Earle: There, there’s a whole different podcast there. You mentioned it’s on the border of ethics. and there have been examples of people interpreting.

There was a very famous example of, I think it was a mother getting an email, or somebody got an email which was to do with, birth control, when, they hadn’t told their family. And I mean, there are, there are some ethical things here, but, but, it, it, it, it… You know, one of the reasons why w- our pod became, AI and IoT was that we talked about there’s 50 times more data from things, and a thing could be a cow a thing could be a wrist for a Fitbit.

A thing could be factory floor, where we’re going next. But there’s 50 times more data from things than there is today, data in all of the LLM models worldwide, ’cause that’s done by scraping the internet, for d- data, voice, and video, content. And the point that you’re [00:18:00] making, which, which I also believe is true, it’s not just that you can get 50 or 100 times more data.

We had a guy on from Volvo who says, “You know, I, uh, I’ve got 500 million things on the factory floor, and my job is to get the data from all of them and stop the factory, uh, the, the 100 and f- any of the 150 production lines that Volvo have globally from stopping at any time, 100%, you know, production 24/7, 365 days a year.”

It’s the context, as you say, the semantic meaning of what it means i-in, in terms of adding intelligence to the tracking. And it goes back to that first point that you made, is that tracking so far, which, you know, it’s one of the biggest verticals, certainly one of the top three, but it’s nowhere near as big as it should be because it’s simply been the blue dot.

Well, before we go to the manufacturing one, let’s talk a little bit about Pozyx. Because people listening to this could say, “Okay, well, what type of company are you?” ‘Cause you’ve talked about [00:19:00] hardware I know you’re an RTLS, provider, real-time location services, provider, but you’ve also talked about software, and you’ve talked about end-to-end solutions.

So if I was, talking to you and, and, and thinking about using you for a use case, what, what’s the best way for me to look at you? what… How would you describe where you sit in the value chain on this- Yeah … on this sector?

Samuel Van de Velde: So we have, developed our own location system. It’s an, an ultra-wideband, and Bluetooth RTLS system, a real-time location system.

that’s where we started from. And with that, we can partner with, with all types of, companies that want to use it for different kind of, industries or applications. For example, with the cows, but also in, in, entertainment, in museums, and in industrial environments. but what we’ve seen over time was that, a lot of the use cases were in the, industrial realm.

And that’s why about five years ago, we decided to [00:20:00] also, build more of a, an application layer specifically dedicated for, manufacturing and logistics, that already solves a lot of problems, that we can provide more than just, a stream of location data, but that we can already, solve use cases and, and, and bring value from the start without requiring a full, application development or, or, or other applications, to do so.

So we, we focus on that industry because, well, we cannot, focus on all the, applications or all the industries- but yeah, manufacturing and, and warehousing and logistic, that’s really where we, we have a lot of experience by now, and a lot of, uh,

Nick Earle: solutions. And, and it’s probably the big opportunity.

And we’re going to be talking about AI, in the second half of this recording. We’ll be there soon. But, but let’s, let’s talk about… Because I, you know, back to the Volvo example, it is the huge, [00:21:00]o- opportunity, and an area where historically there’s been automation at the equipment layer, but there hasn’t really been, any of this context, meaning capabilities, in, in, at the factory floor.

So can you give, our listeners an example of, a manufacturing example, one of your clients, and what sort of value you, add to them?

Samuel Van de Velde: Yes. So for example, in, in discrete manufacturing, so goods are being produced. They go through different production steps in the manufacturing, process. Typically what’s happening today is that there are barcode scanners everywhere to scan if something arrived at a certain workstation.

This could be for the order, but this could also be for the materials. So scan, scan, scan all the time. If scans, are missed, then, yeah, the data in your ERP or MES system, yeah, gets, stale, is outdated. And so we can, by, by tracking the orders, [00:22:00] that go through the manufacturing, by tracking those directly and maybe also the material, you don’t require the scanning anymore And what this also brings to the table is that you can do all kinds of analytics, you can do process mining on this data.

If you also track the operators, you can actually, compute the time that operators have spent on a certain, work item. Because everybody can, easily compute the, the bill of material of a certain order, but what’s much more difficult is the bill of labor. How much time was a person or a group of people spending on a particular product?

And especially if you have this high mix of customized products, it could be that certain customizations there require much more time, than others, and that’s something you, you just wouldn’t, wouldn’t, know very well, without this, this system. Also, something that you can do is, do an inventory, an automated inventory.

We see very often [00:23:00] that, manufacturing companies, they shut down production when they need to do an inventory count, which, is this huge manual process where they have to go through and count everything. Well, if everything has a, a tracker, you don’t have to do that anymore. You, you can just continue your operation.

You don’t have to shut down, uh, your operation.

Nick Earle: My interpretation of, uh, what you’re describing there is, um, I, I always relate it back to other stories and historical, lessons in business. I mean, obviously Henry Ford, was the first person to actually define the manufacturing process in terms of the Model T Fords and the conveyor belts, but it was really at the high level.

What you’re saying is at a much lower level, which is that you may have well-defined your manufacturing process, but you’re certainly not getting that data from how long does an individual person spend doing it and where the inefficiencies are.

So my take, and I just wanted to check this, and I think it’s really important in the context [00:24:00]of the AI discussion we’re about to have, is that what you’re actually then able to do if you do that is produce a high level, what I call a, a, a real workflow definition of how work gets done at the factory floor as opposed to the what’s written in the manual, right?

There’s what’s written in the manual, this is the way it works, and then stuff goes wrong and you don’t know why. But if you’re actually tracking it at the real time, the component and the operator and the location, then I think you’re getting a very rich semantic definition, of real workflows on, on the factory floor, which is the gold dust.

Because once you’ve got it defined, you can then say, “How can I improve it?” Or, what would give me a 10% improvement?” have I got that right?

Samuel Van de Velde: And, I think it’s, it’s, also I think, up to today, a lot of the, location use cases are on the, the dot on the map, but then also a little bit on analytics, which is, in this case, you, you, [00:25:00] you try to get insights on, on this location data.

But I think what’s also the type of use cases that are relevant are the automation use cases. So going back to this manufacturing environment, so imagine that, that, something arrives at a certain, workstation. Well, in manufacturing you have something like just-in-time, production, and you, you want actually that the materials for that work order, they arrive just in time, and you don’t have a huge pile of stock at- Yeah

the workstation to do the manufacturing. And, and that’s also something that you can automate, and that you can reliably automate, I would say, by having the location because you can know that, well, a new order is arriving. It will be there in a certain, amount of time. You can then launch a logistics, request- Yeah

based on that location, and even better, you can track it if that logistic request- Yeah … is actually coming. oftentimes, they… Sometimes they have to shut [00:26:00] down a line, because, Yeah.

Nick Earle: They, they know what to do, but they don’t have the, the materials to do it.

Samuel Van de Velde: Yeah. And they don’t have the visibility.

Like, okay, is it, is it arriving? Is it not arriving? Is it even in this building? Yeah. Uh, I’ve heard crazy stories that sometimes they have to, yeah, get a helicopter to expedite certain goods- … to bring them in, just to not stop the line, because stopping the line is extremely expensive.

Nick Earle: Extremely unbelievable how bad the re-

Samuel Van de Velde: that.

Nick Earle: the return on investment of IoT and stopping the line is 10,000 times more expensive- Yeah … than, than, the cost of implementing it. And what you’re… this is advanced IoT that you’re now using, but the world is changing- Again, I mean, most people aren’t doing what you’re saying anyway.

But when it took– comes to AI, then the opportunities to, to add value are gonna be, exponentially bigger, and [00:27:00] efficiency and value. So let’s try and unpack that. So at a high level, the, the interpretation of the data in particular, it, it, you know… I guess the question is what happens when IoT and AI come together, in this, factory?

Let’s pick- keep on the factory floor environment. does the, does the ability to interpret and, and, and get meaning, semantic meaning from the data grow exponentially with AI? And if so, can you explain how that could happen?

Samuel Van de Velde: I absolutely believe that. I think, with AI, it’s much more easy to formulate what you actually want and then maybe get graphs or dashboards that answer those questions.

And with IoT, you can actually answer those questions with actual data from, the factory floor, and not, maybe stale data that is living in your ERP system. it obviously, all this data can come together. [00:28:00] Now, one thing I do, want to warn for is that while AI is built on large language models, so they can handle language very well.

handling, large streams of, bytes and, and locations and numbers, it’s not as good in that. And, and also- Yes … quite frankly, if you have this huge amount of data, just, loading all of this data in, just to feed it to a, to an AI model, it will take a long time. Also, I think all of your credits will be- Right

will be spent.

Nick Earle: Expensive.

Samuel Van de Velde: So I, I do believe that in order to do that, you already have to reprocess the data, add layers of semantics to it. So for example, with the location, just giving raw location data doesn’t make sense. You have to say, “Okay, maybe it was in certain geo-fences.” So, giving contextual meaning, maybe saying how long it has been standing still, maybe giving, input about, okay, how many [00:29:00] times has it moved today in terms of trips.

And, these things are something that an AI can better reason about and very quickly then, catch, anomalies and, and, and show you what you actually want to see, so.

Nick Earle: So let’s try and make, put that into context so people can sort of create a picture in their mind of it. So let’s say we’ve got a big factory, and I- in my mind, I can see hundreds of forklift trucks zooming around the factory.

I think what you’re saying is that just simply collecting data from the forklift trucks, what they’re doing and where they are and the timestamp and whatever, and feeding it into AI doesn’t, doesn’t give you the value that you need But you do want to interpret and, and, and see whether the forklift trucks are in the right place and, and whether they are doing the right thing.

So are you talking about an abstraction? Are you talking about, creating an abstraction layer that, that, that sits above, all of that to, to give, to, to enrich the data and give [00:30:00] it context and meaning before you- AI does the interpretation?

Samuel Van de Velde: So w- with ultra-wideband positioning, you can get the location of a forklift at 10 to 30 centimeters every second or a fraction of a second. So these are huge amounts of, of data. If you have hundreds of forklifts, just querying that data might take minutes. and, and then if you ask a question, “Okay, do, am I using all my forklifts efficiently?

do I actually need as many forklifts, lifts, as I have?” That will be very difficult. But of course, if you pre-process it, then, you can already give, an indication of, okay, how many trips are they doing and when are they doing those trips, and at which time? can, you can see, okay, when is this activity happening?

And maybe in certain time periods or maybe the entire day you have an overcapacity, and that’s something that, AI would easily spot. Then once you have this pre-processed data, the AI will be able to tell you, “Well, actually, you [00:31:00] don’t need as many forklifts,” or you, you maybe do, but only on these days when you have certain peak moments.

Nick Earle: And I guess the more you do it, the more you abstract, the more it can recognize the patterns. And so over time it learn… You, you’re training the model with the data, but you’re not trying to, get intelligence from the data. Meaning, is that you’re putting it into this abstract layer, and I use the phrase workflow, but you’re…

It’s almost like your intelligence layer. I remember we had the research director from IDC on in the US, talking about the enterpr- the need for large enterprises to develop the enterprise brain. and, and some people talk about, well, that’s what the digital twin model is gonna be. But essentially this idea that the…

It’s not just the raw data, it’s the context and the learnings and the patterns, that actually allows you to then go back down into the, raw data and say, “Is that normal? Is that optimal? Is that inefficient? How do I improve [00:32:00] it?”

Samuel Van de Velde: And you still would want to maybe have this, this, highest level of detailed data still available that you can dig into it when needed.

But I think, yeah, that would only be when you’re digging into it, and not when you’re doing, yeah, your analysis and, and, and asking, uh, your AI, high-level questions.

Nick Earle: There’s a lot of people right now, the most common reach-out is people talking about this area, which is in manufacturing, which historically just, it’s been automated.

I’m old enough to say, back to SCADA systems and just getting basic data, if any at all, from the factory floor. it’s an area that started off very early on with electronics and components, but arguably has had the least progress. Most factory floors still look very similar.

Most, not all, but, but a lot of them look very similar to the way they did 20, 30 years ago. So it’s an area where this abstraction layer, the, what I call the workflows or the [00:33:00] intelligence layer or semantic brain, or lots of people using different, different phrases is, is, is a lot of people are trying to get that, and because they think it’s their future competitive edge But where I wanted to go to, and as we get towards the end of the pod here, is that when we first chatted, you actually were saying, “Well, yeah, that’s where we’re going and that’s the big thing right now, but there’s something else.”

I found that really fascinating. I was like, what do you mean something else? It’ll take years to implement that.” And you were actually talking about a totally new class of applications that almost get created on the fly, Not by the IT department, but by the user requesting something that they need, maybe even by talking, to something or saying, “This is what I want,” and an application is dynamically created to even further optimize it, not at the abstraction layer, but right at the sharp end.[00:34:00]

And I, I… First time I heard that, I was like, “What?” and then I thought, “Wow, that’s a very interesting idea.” So I know you’ve got a strong view on this. could you give a better explanation than I’ve just done on what you think- is the next step?

Samuel Van de Velde: Absolutely, and it’s something that I’m very enthusiastic about because I think using AI to interpret the data, that to me is, is not new. I think now with the, the large language models, it can happen easier and faster, but I think trying to interpret the data using AI machine learning, that has happened for quite some time.

What I’m really excited about, with this new AI is that all of a sudden it becomes much more easy to, develop small applications, small or large, but, small applications that you could run on, the shop floor, in, on the factory floor potentially. and why does this matter is because in my view, a lot of the IoT [00:35:00] systems today, the data goes up, and the data goes up into some database and, then you can do some analytics.

But very rarely the data goes back down again to the shop floor where people can actually see things, interact with it, and, and also automate it more. And my view is that, well, the reason why that’s not happening is because, well, you need custom applications, you need terminals to do that. You need to have buttons that, that are still relevant for that type of business, and that is very, costly to do so.

But now with AI, it is very easy to make these very simple terminals where you can get the data again from the cloud, get it back to the shop floor, and allow you to automate or, or just, yeah, use it much more. And yeah, I think especially for an RTLS, a, a location system, the challenge that a location system has always had was [00:36:00] that you have to install some infrastructure, you have to tag, your assets.

but there is not- Or, or ve-very few times a one killer application. it’s more that you have many applications, many use cases, but they need to be implemented, all of them. And so now it becomes very easy to implement these smaller use cases that bring still a lot of value, and combined a massive amount of value, where before it, it would just cost a lot of money and take a lot of time to do so.

One example, for example, that, that I can give was a company that, rolled out the RTLS but they said, “Well, we actually also want to track the visitors that are going through, through the factory. We, we, we, we don’t want them to walk alone through there. They have to be accompanied, and we have to register, where they are.”

Well, all of a sudden, with AI, we could develop this very simple [00:37:00] app, this kiosk, application that you often see when you enter a building where you enter your name. Well, it was like that. You enter your name, you grab a tag, you scan it, you, you put it around your neck, and all of a sudden that visitor is being tracked by the system.

And at the end of the, the trip, it takes the, the, the, the tracker off, and everything, was registered and, and this application by itself costed almost nothing to develop, whereas before, it would be a completely different solution provider maybe providing a solution for visitor tracking.

Nick Earle: Yeah, You have an idea, and these things- Yeah … can be developed in… Well, with AI, they, I mean, it’s like a, an agent can be, spun up and, and you could develop dozens of them very quickly on demand.

Samuel Van de Velde: For example, having a terminal that just shows you on a workstation, okay, which ones are my assets that are laying around here for the longest time, or maybe ones that have a warning that I have to attend [00:38:00] to.

Or maybe on a forklift you have an app that, that shows you your location and shows you where you have to go to pick something up, and that automatically, detects

Nick Earle: Excuse me, I was thinking of that, to bri-bring it back, to the forklift. If you had the 200 forklift trucks r- in a, in a big factory, as you say, the equipment to communicate, they probably have to dock and they have to plug in, and then there’s this thick screen that’s been there for, for, for years and it’s very expensive, as you say.

But the fact is you could have a very cheap little terminal, on the forklift truck. I mean, basically, not, not the guy’s iPhone, but something fixed to the truck which actually gives a lot of dynamic context to where they, where… Not just where they are, but what they should be doing, what they should be doing next.

“Oh, there’s been a change. G-go to this point, not, not that point.” And so what you’re describing is- it’s, it’s, it’s almost like a new frontier. It, it, it’s this… It’s right at the sharp end where the user meets the… where the user is working on the factory floor, the worker [00:39:00] is working on the factory floor.

Their, the actual applications that they have are being defined by somebody else. They’re being defined by somebody else in an IT department a long, long way away, and, and the data is just coming down, and, Sorry, the data is going up, and there’s not much intelligence coming down.

And what you’re describing is dynamic micro apps or agents or whatever you want to call them. They’re context-specific applications that can be developed almost, not quite real time, but pretty, pretty damn quick, that actually are, are, customized to the contextual data that you’re collecting and the, and the semantic interpretation.

And so you are actually, your IT system is creating apps and pinging them out based on what’s actually happening, which is a world away from the application development process-

Samuel Van de Velde: Exactly …

Nick Earle: But now you’re talking about apps being just created, created based on what’s happening. And they’re [00:40:00] micro apps, and they may be for simple use cases like tracking visitors or, or this forklift truck is not being used, or there’s a forklift truck around the corner, or change, change mind, go here, not there.

But the efficiency from all of that is something that’s not been available because all that happens is the data’s been going northbound, and you’re talking about context and intelligence, coming southbound to individual people through very low-cost devices- That’s it … so that, from this abstraction, this intelligence layer, the enterprise brain, as IDC call it, and that’s basically the fu- that’s a whole new opportunity for entrepreneurs and software developers to create

a new set of applications that would get mass adoption on a pay-per-use… I mean, it’s almost like an API. You know, each time you use the app, maybe there’s a fee, or you create it as your IT department, and there is no fee because it’s what you’re- Yeah … you’re creating.

Samuel Van de Velde: Yeah. And so, as a company, we’ve been [00:41:00] thinking about this.

How can we support that? And, what we understood is an application requires very strong APIs, and we’ve actually made it possible that with these APIs and also with the security required to, to do these things, we can offer our customers to, to build these, applications themselves.

But then on the other hand, yeah, we also provide software. So I think towards the future, we’ve also been thinking, “Okay, what is the future of our software if everybody can develop software of, of their own?” And w- what my belief is, and we’re not there yet, it’s… I think it’s, it’s, it’s a bit even more challenging, is that if, if we provide software, what always happens is that, well, we put something on a screen, and it might not be exactly what a customer wants to see on that screen, and, they may require a customization.

In the past, this was through elaborate [00:42:00] maybe dashboards and with widgets and whatnots. But they

Nick Earle: could do it themselves.

Samuel Van de Velde: Exactly. but I believe that in the future, maybe software providers would be providing a software where it’s possible for the user to change up the certain screen- Yeah … just with AI, talking to it or, or just inputting it and saying, “Well, I actually, when I click on an asset, I want to see the asset, but I actually also want to see this and this information,” which doesn’t even live in your application.

It lives in another application.

Nick Earle: And,

Samuel Van de Velde: And I want to bring those together because I don’t want to build a, a completely from the, from the ground up new asset tracking application or, or i- i- in other business areas, a, an, a completely new ERP system or a CRM system.

Nobody wants to build this from the ground up. But you definitely want to customize it to your own processes, and without, too much pain, and in a secure way, and in a scalable

Nick Earle: And that’s the trick, isn’t it? Secure, [00:43:00] auditable, knowing that the data is correct, it’s been checked. But it’s a completely different model for software where you just describe what you want And in theory, it, it appears, even if it has to go outside to get it.

And, wow, that’s a, a lot of change, and it’s probably a, a lot for our listeners to wrap their brains around- … because we’ve covered, we’ve, we’ve covered a pretty, pretty wide spectrum here. you’re clearly, Samuel, you’re clearly a visionary. I know you’re the CTO and, a founder. you’re based in…

You talked about Ghent University, your PhD. So you’re based in Belgium. I, I omitted to, say that, Yeah … and I… And you founded this- after your PhD. Yeah.

Samuel Van de Velde: Yeah, we also have a, an office in the US, so. We’re mainly active in Europe and, and the US.

Nick Earle: So if people wanna find out a little bit more about your company, maybe you can just spell it out for people if they want to just check you out on the internet.

Samuel Van de Velde: Yes, it’s, P-O-Z-Y-X. So it comes [00:44:00] from positioning and then XYZ coordinates but then flipped around,

Nick Earle: That’s such an engineer’s description of a word. There we go.

Listen, Samuel, this has been a great, great episode.

I’m sure our listeners will, um, enjoy it. And there’s an awful lot to wrap their brain around, but there’s a lot of people in the manufacturing space who are looking at this. As I said, I get a lot of, requests to talk about manufacturing and the opportunity for IoT and AI to come together. It’s probably, the, the, the first area where th- these two will really come together and add totally new forms of value.

Samuel Van de Velde: So thanks for, uh, being my guest on the Podcast. It was a, a very interesting podcast, I have to say.

Nick Earle: Yeah. Great. Thanks for being my guest.

Samuel Van de Velde: All right. Thank you very much. Bye-bye.