skip to Main Content

About this episode

In this episode of The IoT Podcast, we are joined by Michael Gilfix – Chief Product and Engineering Officer at KX to dive into the secrets behind lightning-fast data analytics and processing in IoT, uncovering some mind-blowing use cases 🏎️.

Why are F1 racing car manufacturers are turning to KX to supercharge their performance? This episode explains all.

We begin the episode by discussing the connection between high-frequency trading and IoT, followed by case studies highlighting KX’s impact in energy (gamification), automotive (F1 cars!), and manufacturing industries. We dive into the technology behind KX, explaining how it processes and analyses data in real-time IoT environments.

Lastly, we explore the debate of data processing location – edge or cloud – and the trade-offs involved.

Chapters

  • 00:00 Introduction to Michael and Data Analytics in IoT
  • 08:07 The Value of High Speed Data processing in IoT
  • 13:22 Metering Case Study
  • 19:02 F1 Racing cars Case Study
  • 26:05 Handling Large Volumes of Data and Ensuring High Availability
  • 28:53 Edge vs Cloud – Most ideal data processing point?
  • 33:17 Democratising Analytics and Harnessing the Power of Data

Thank you to our season sponsor 5V Tech. Discover how 5V Tech can help you unlock your scaling potential in cutting-edge tech and IoT – Here

Tom White (00:01.267)
Michael, welcome to the IoT Podcast.

Michael Gilfix (00:05.378)
Thanks for having me. Excited.

Tom White (00:07.443)
I’m excited as well, actually, because we have a lot of people that come onto the podcast who are talking about IoT when it comes to IoT devices or a more of an industrial landscape when it actually comes to the physical sensors, right? But I think what we’re going to get into here is more the data analytics and the speed of those data analytics. And that’s such a crucial element to IoT, as we all know.

But some of the tech that you guys are doing in your business is phenomenal. But rather than me trying to explain it, I’ll ask you to kindly do that. So as a way of kicking off, could you explain who you are and what company you represent?

Michael Gilfix (00:47.886)
Sure, so my name is Michael Yelviks. One second, can we edit that out? I forgot that one minute. Let me just, can we restart?

Tom White (00:53.911)
We can, we can.

Tom White (00:59.211)
That’s all right, you go, you go. Do you need to go or?

Michael Gilfix (01:02.782)
Well, I’m just going to move him to a different room. That’s my elderly cat. So I’m going to move him to a different room. So we don’t listen to that during the podcast. Can you give me one second? I should have thought of that. Be right back. Sorry.

Tom White (01:05.12)
Okay.

Okay, no problem, that’s fine. Yeah, yeah, we’ll start again. That’s okay, don’t worry. All right.

Michael Gilfix (01:50.702)
Okay, let’s try that again. I put them in a closet, hopefully we’re not gonna hear them. Ha ha ha.

Tom White (01:51.471)
It’s all good. Ah, okay. All right, cool. Let’s just go from the top, I think. It’s probably the best, so I’ll just do that now. Welcome to the IoT Podcast, Michael.

Michael Gilfix (02:01.099)
Yes.

Michael Gilfix (02:08.958)
All right, well thank you so much for having me. Super excited.

Tom White (02:12.307)
I’m super excited as well because often we have people coming on to the podcast who are talking about the physical sensors in IoT, either from a smart meter in context or industrial, so on and so forth. But today we’re going to get into data analytics and more importantly, the speed of analyzing that data. So rather than me trying to attempt to explain it in my rather layman tone, I’ll ask you to do it, Michael. So as always, could you start by saying who you are and what company you represent?

Michael Gilfix (02:43.338)
All right. So I’m Michael Yelfix. I’m the chief product and engineering officer for a company called KX. And a little bit of what we do, uh, as you pointed out, we’re on the analytics side. So we have a technology that makes it really easy to take large volumes of data. And analyze them for business value. And one of the secret sauces of our technology is not only can we take these large volumes of data, but actually we can analyze them in time and help to gain insight from that behavior over time.

Turns out that’s super useful for IoT use cases.

Tom White (03:16.555)
Yeah, absolutely. I think it’s paramount, actually, I would say, in terms of the integrity and the analytics of that data done as fast as possible and accurately as possible. But it’s quite an interesting story, actually, about how the business came about and the origins of the company. So could you talk a little bit about that, Michael?

Michael Gilfix (03:36.462)
Sure, especially in the case of IoT, you might say, as I described the history, so how does this get to be an IoT business? But the origins of our business is, we were born out of one of the most demanding use cases, which is implementing capital markets for financial institutions. Capital markets is the buying, the selling of equity and managing those investments.

And if you think about those kinds of use cases, there are use cases that have a tremendous amount of volume. You get market data, stock tick data. You’re trying to make sense of the world around you and you’re using that to figure out, is this a good trade? Is it a good order? How can I do better? How do I build algorithms that can exploit that? And we built a data platform that actually allows you to perform incredibly well in those kinds of use cases.

So we can process very large volumes of data. We have sub-second response time. So you can figure out the right algorithm. We were able to validate models that run on the data super computationally intensive, but we still make all that stuff work. And you can run your rules in kind of in real time on that set of data. So you can make the right trade or analyze the best trade. And at the heart of that technology was also the ability to process time series data.

And time series data is basically looking at behaviors over time and pattern matching that data or finding ways to reason and predict what’s going to happen with that data over time. So fast forward, you know, many kind of years later, we’ve been bringing that technology to a variety of different industries that benefit from the power of this analytical platform and scale. And

It turns out that IOT use cases are just a tremendous opportunity for that technology because it has many of the same characteristics of that core use case, large data volumes, predicting how things are going to behave or looking at patterns of behavior over time, a sub-second response time, and the kind of scale and price performance that our technology brought.

Tom White (05:45.439)
Yeah, I think for me, it’s really fascinating because it’s not a world that I know incredibly well. So I’m a device guy. I came from embedded Linux. I know video, I know IoT and devices that kind of a low level embedded component system level, shall I say pretty well. And it wasn’t really until I listened to a couple of podcasts about Bloomberg, you know, and the Bloomberg terminal and the necessity for speed around this. So I can really see.

that the origins of the company are rooted in something absolutely fundamental, critical for that industry. As in literally seconds can make a difference and colossal that difference in terms of trades and what the trades look like. So it’s actually quite a unique for us segue into IoT that we’ve not really had many times in five seasons of the podcast, right? But when you talk about it, it seems so obvious, doesn’t it?

Michael Gilfix (06:43.85)
Well, you know, since you background the hardware person and it sounds like a lot of listeners are, I mean, everything these days is becoming virtualized. The hardware is only one component. It’s what you can do with the data that is incredibly valuable. So just to give a simple consumer example, and this is one from my own personal life. So I’m gonna hold up, see this, I gotta get some credit by the way, for advertising this. This is a whoop. If you’ve ever heard of a whoop, do you know what that is?

Tom White (07:09.195)
Ah, I’ve got one here.

Michael Gilfix (07:11.186)
Oh, you got one too. Okay, great. So you know exactly what I’m talking about. So for the listeners, a whoop is a device that, you know, monitors your, uh, your bio data effectively. And it helps you to track things like sleep, strain from workouts, general performance. So I use it for my athletic pursuits, but for those of you that actually own the device, one of the things you’d realize very quickly is that the hardware is actually like a really, really small portion of the device experience.

Tom White (07:14.165)
I do.

Michael Gilfix (07:38.962)
It’s actually recording this data. It’s uploading it to the cloud for analytics. The application is helping you understand baseline performance. They look at trends, like they survey you to say, what characteristics behavior did you have? And then they correlate that with your sleep performance. But all of that analytics is happening in the cloud and it’s being bulk uploaded. And so that’s maybe a really simple way to understand kind of the feeling of, you know, the hardware is really one portion. It’s the enabler, if you will, the on-ramp of the data.

But all of the power came from funneling that data into a platform that could make sense of it and then turn it into an application that someone could use. I think in the case of KX, we’re looking at a level of scale, probably that well exceeds, you know, whoop in our applications, but the real power again is taking that raw data and turning it into the kinds of analytics that open up just tremendous business possibilities with that data.

Tom White (08:32.843)
Yeah, that’s a great analogy, you know, and I’m addicted to mine. I mean, that’s a whole other topic about how it becomes a self-fulfilling…

Michael Gilfix (08:41.238)
Man, if you’re really goal-oriented, there’s nothing like having something critique your sleep performance to go, man, I gotta really, you know, gotta get on the ball on this thing. So it’s very useful.

Tom White (08:47.151)
Yeah, yeah, yeah. Oh, absolutely. Yeah, absolutely. I often worry and wonder sometimes if my recovery is below 70%, if I then think, oh, I’m going to have a bad day because of it. And sometimes I wish I’d never seen it, you know, but that’s a whole other topic. So going back to KX and the business in terms of the actual industries that are most prevalent now. So we spoke about your origins of where you came from.

Where do you see most of the activity coming from the data processing using your technology today?

Michael Gilfix (09:22.21)
So we have four industries that we see driving a fair amount of demand for us. One is manufacturing. We can talk about these a bit more detail, but I’ll just give you kind of an overview quickly. So manufacturing, so thinking about how you’re going to manage the equipment data that comes off of a production line, especially if it’s high tech manufacturing, like semiconductor fabrication, for example, highly, highly automated. The second is energy and utilities.

Tom White (09:33.345)
Yeah.

Michael Gilfix (09:50.102)
So being able to take information around metered electricity, related things that are driving energy consumption using that data in real time, tons of amazing applications emerging there. Healthcare life sciences, rife with deep data to understand about what’s going on with patients, what’s going on with procedures, how do you feed that back into analytics or even clinical data that can be monitored over time. And then automotive.

Uh, those are some really fun ones. It’s, uh, like I want to run simulation on my race cars. And while the race, so not only am I going to run simulations, let’s say in wind tunnels to provide performance, but I’m going to look at the race in real time and then run simulations to predict future performance and all that’s running in analytics in real time and each of these platforms have the characteristic of giving off, uh, effectively large amounts of data, uh, in a short period of time. So, you know, there’s a.

You need to be able to keep pace with it. You need to be able to make sense of it. And I think when you dig into each of these cases, there’s some unique requirements as well on how you handle data and be smart about its processing.

Tom White (11:00.039)
Yeah. So before we get into the individual industries in a bit more depth, is it primarily industries, so take automotive for instance, race cars, telematics, is it primarily where data, the decisions need to be made quickly? Is that primarily where

people are using and harnessing KX’s technology above and beyond other solutions that they might have out in the open market.

Michael Gilfix (11:30.846)
That’s certainly one of the drivers, but it isn’t the only driver. Um, one of the things that we’re really good at doing with our technology is a form of simulation. It’s when you run in parallel, lots of different scenarios to pick the optimal one, that’s not a true real time thing. Um, but obviously you want your simulation to end in a reasonable amount of time, but you’re running a large set of scenarios simultaneously. Our system’s really good at paralyzing that so that we can figure out an optimal

scenario. But I would think the two top ones are sort of either you’re uh, you’re either monitoring actually let’s do three. Sorry, you’re either monitoring making sure that everything’s functioning or you’re running a form of real-time analytic or you’re running a set of simulation to figure out let’s say optimal scenarios with key devices. I would put under monitoring by the way predictive maintenance for example for machinery is under monitoring as well.

When’s equipment going to fail? That one’s kind of obvious, but if you’re in say high tech manufacturing and you’re working on a fab production line, you obviously want to make sure that the process is going to succeed and downtime is extremely costly, right? For those kinds of highly automated production lines. And so your goal is to detect issues in the production line and try to resolve it as quickly as possible.

Tom White (12:55.367)
Yeah, yeah, no, absolutely. I think that’s really interesting how it’s, you know, that it’s not just a driver around speed, which, you know, could be forgiven for thinking about sometimes, right? Going into a bit of a deep dive into energy and utilities then. So can you talk to me about the real time meter monitoring? And also, we spoke about this in our discovery call around the game of occasion, which I thought was really interesting, actually.

Michael Gilfix (13:22.366)
Yeah, so let me give you kind of an example. We have a use case where literally our technology is monitoring all of the meter consumption for an entire country. And the meters report back somewhere between five minutes to an hour per meter, but there’s literally millions of data points that are being reported back simultaneously that need to be aggregated. And one of the things about meter monitoring is that

the order of the messages back really matter because otherwise you could draw the wrong conclusion around consumption trends. And consumption trends are typically what drives a grid. Like how do you value, if you think about like a lot of grids have things like surge pricing, right? Depending on the consumption, the pricing sort of scales, depending on the utilization of the grid. Likewise, you want to use that to forecast demand. And so that’s a running average, if you will, right? From taking the grid. And so it’s really important that

the data be ordered correctly in order for the forecasting to work. And it turns out when you have a ton of these different meters running, messages get lost. They come in out of order, right? And you imagine that at a massive scale, the analytics platform has to make sense of all this data. But once you get the data, there’s some really cool stuff you can do. So, uh, one of the things that you can start to do is plan your energy consumptions around peak periods. So, uh,

I mean, look, Tesla is trying to build a business model around this, right? You get the advertisement that says, Hey, would you like to sell your data back to the grid or would you like to store it during off hours where it’s cheaper and then use it in peak hours? And so you can be more energy efficient and green if you were just smarter and had better transparency into your data. One of the things I like, we were talking a bit about gamification. So I live in Austin, Texas.

For those of you that don’t know, Austin is really, really hot in the middle of the summer. And especially as temperatures have kind of gone up, I think Austin has gotten more prolonged days in extreme heat. I have to remember my Celsius conversion. So for those of you that know Fahrenheit, we have over a hundred degree days during the summer for an extended period of time. What is that? Is it like 40 Celsius? 42? Okay, yeah. Let’s call that the good enough conversion.

Tom White (15:41.031)
42 I think. Yeah, it’s hot. It’s hot, it’s hot. Yeah.

Michael Gilfix (15:46.462)
Needless to say, it’s pretty hot. We have really good air conditioning technology in Austin, Texas, but when it gets really hot like that, the energy demands on the grid go up substantially. Right. And there’s a lot of tension. So one of the things they implemented locally, once you get real time visibility into this consumption, one of the things they implemented locally is, uh, they would, uh, pitch you against your nearest neighbors. So I would get an email that saying, here’s how you performed versus the top hundred houses in your local neighborhood.

in a bid to encourage me to be more energy efficient and would rank me based on that week, let’s say, or that day, here’s how you did versus your neighbors. And that’s a great way to incent homeowners to be more energy efficient because you might look at that and say, first of all, my bill became astronomical during the summer because of surge pricing. And for those that live in Texas, they may know that by the way, Texas implements some things where the surge pricing can get pretty ridiculous past a clip level.

So not only do you get this expensive bill, but you’re dangled this great carrot of, and it doesn’t have to be that way. If only I found ways to make the thing more efficient. And that’s a pretty interesting way to use data. And if you think about the, how did we get there? Like step one was, you know, instrument all of these different sensors to pull back the data. Step two was process it so you had real time visibility.

Tom White (16:48.835)
Hmm.

Michael Gilfix (17:07.838)
And then step three opened up, well, once these new applications, you know, once I had the real time visibility, what could I do with it? Gamification, uh, being smarter about, let’s say buying and selling things on the grid for those that are, are green inclined, just tremendous opportunities, right? To be more efficient and the average homeowner has no idea because how would they know what the basis of comparison is? What incentive would you have?

Tom White (17:32.985)
Hmm.

Michael Gilfix (17:34.306)
For all you know, you’re the best at being green because you put some weather stripping in or something. But once you see how it compares to your peers, okay, well now you have a basis of comparison, you’re a little more motivated.

Tom White (17:45.875)
Yeah, yeah, oh absolutely. I think, you know, the gamification really does help because naturally people are competitive and naturally people want to do the right thing. So it serves two purposes by doing that, but I think it also works on scale. I mean, the start of that you said that you, I think, correct me if I’m wrong, but you received the data from a whole country’s

Michael Gilfix (18:14.746)
Yeah, in that particular case, it’s a European country where we’re actually running analytics in real time on the entire meter infrastructure.

Tom White (18:22.731)
I mean, that is absolutely colossal. And what a coup to be able to say that as well, right?

Michael Gilfix (18:29.802)
Yeah, it’s one of the Scandinavian countries. I think that’s as far as I’m allowed to go in advertising it, but, uh, it gives you a size of the scale. But it, for me, that’s someone who builds products. That’s a pretty exciting thing. Cause how often you get to talk about your product effectively powering an entire country. I mean, not everybody gets to say that, but, but think about that as. That’s a truly scalable. IOT use case. I mean, I think in the IOT market.

Tom White (18:32.875)
No, that’s alright. I’m not probing it. I don’t want to know who it is.

Tom White (18:47.509)
Yeah.

Tom White (18:53.058)
Hmm.

Michael Gilfix (18:54.282)
It’s been talked about smarter cities and stuff like that for a long time. This is actually in practice with regards to providing energy consumption and visibility.

Tom White (19:02.891)
Yeah, oh, absolutely. No, it’s fantastic, fantastic. So moving back to the automotive use case that you mentioned as well then, Michael. So we spoke about racing cars, the wind tunnel simulation, et cetera. So one of the things that I know from racing cars is, as I mentioned, the telematics, the need for low latency and for understanding things in incredible detail, because when you scale this up to…

your Indy cars or your F1s, it really does make a massive difference of course. So it’d be great to dig into KX’s involvement in automotive.

Michael Gilfix (19:40.958)
Yeah. So, uh, one of the most public companies we have in that space is Alpine, the French racing company and, uh, their use case has effectively two components to it. The first is wind tunnel simulation, meaning the goal there is to take the cars with their data, put them in the wind tunnel, simulate performance, and then effectively build a basis of data by which they can build a model of vehicle performance.

Tom White (19:46.384)
Okay, yeah.

Michael Gilfix (20:10.534)
And so you run a batch simulation, you get a bunch of data, you can offline batch process that data, and like I said, you can figure out these predictive characteristics for it. So think of this as that’s simulation time. That’s race prep time. Then what they do is during race time, as the vehicle’s going around doing laps, they’re taking the live telemetry data off the vehicle, they’re taking the predictive model that came from the batch simulation, and they’re making the call on.

You know, should I do pit stops? Um, what’s the likelihood of part failure? Like how long do I have to go before the tires give out? And they’re using that to inform strategies. And then what they do, there’s a third component is they keep the history then of how they did on race day. And that becomes now historical data. And they use that to further tweak for future races. So again, you have the simulation input. You have the, here’s what I ran on race day.

And then you have the historical, well, these are the strategies I ran on race day and the inputs and how did I do so it can inform what you might see in an individual scenario again on race day. It becomes sort of a, uh, I guess, uh, a circle that turns on itself, right? I can, uh, continuous, uh, improvement kind of circle. And so they’re, they’re feeding that data in to go and improve the, the racing component.

Tom White (21:25.26)
Yeah.

Michael Gilfix (21:31.082)
So that might give you a sense of sort of how the sensor data comes, because again, you have that batch kind of input from the simulation, but then you have tons of telemetry data that’s getting pumped off those vehicles while they’re in the race itself. And you got, that’s where real time really matters because you want to make sure that you’re calculating car performance and you’re coming up with the best racing strategy based on the data of the car.

Tom White (21:37.291)
Mm.

Tom White (21:54.175)
Yeah, yeah, I mean, it’s absolutely critical. And again, working with someone like Alpine, you probably learn a lot about how to perfect the software and how to perfect how it works in that environment as well. Right. And what is really important and what isn’t because it’s being done, you know, at a competitive enterprise, world-class level, right.

Michael Gilfix (22:17.438)
Yeah, and I think what’s critical for those kinds of things, you know, a lot of the data that comes in from those sensors is a form of signal processing. Like you’re looking for patterns in the signal over time because, you know, break temperature, measures of wear, handling, you’re looking at the behavior of those components over time and you’re looking in patterns in the signal effectively to figure out when do you act, not just predictive, but also looking for patterns on equipment

you know, failure and those kinds of things. So you need to be able to do a fair amount of time series processing on that data because the performance over time matters as much as it takes for predictive performance. And then those things need to be computed in real time, right, whatever the model is that you use, the rules, you’re running some pretty intensive time-based calculations when you go and do that.

Tom White (23:08.159)
Yeah, yeah, I can imagine. Yeah, crikey. Yeah, it’s really quite awe inspiring. I think that the sheer scale and size of teams’ budgets like Alpine to invest in these things and the differences can be made actually from harvesting the data, as we say. The other use case we spoke about.

was the manufacturing side. Now this was really interesting for me because traditionally speaking, we have a lot of people talking about digital twins on the podcast and digital twins are great because it’s really easy to replicate something without the costs of being able to do it in real life first and to look at stressors or issues that can be contained in a virtual environment. But what I really liked about KX’s solution is around that real time sensor.

around failures and so on. It would be great to dig into that as well.

Michael Gilfix (24:05.046)
So I mentioned earlier that we’ve been doing some work with some semiconductor manufacturing. And let me give you an idea of the two requirements that come out of working on those kinds of use cases. So the first is the sheer volume. So I thought I knew what scale was until I looked at some of those things, but we right now have tested for production line.

upwards of four to four and a half million sensors per second that gets thrown off these things that need to be processed. So that’s, that’s some pretty intensive data volume that’s being transmitted. And, and like a large number of data points that have to be handled. And then maybe the second thing is you have to have technology that’s extremely highly available because for, for our production lines don’t.

go down the way that regular software maintenance windows are. You’re not taking them down for a couple hours, right? Even if you think about things like cloud up times, cloud up times might have four nines, but you know, that’s still a couple hours, you know, a year. They, they might take two years to get to a 15 minute upgrade window. So the availability has to really be massive and because it’s, you know, the cost of taking down that production line is very intense.

Tom White (25:22.563)
Hmm.

Michael Gilfix (25:22.922)
So those are some pretty harsh characteristics that have to balance the data that comes out of that kind of an application. Just again, the sheer volume of it, the speed, the high availability. And all of that is basically used to drive automation, monitoring, it’s a key part of modern manufacturing being substantially more efficient, right? Especially high-tech manufacturing where it’s not like people are minting chips, right? Those things are pretty automated.

Tom White (25:50.639)
Hmm.

Michael Gilfix (25:53.218)
They’re highly small, they need to be done, so they’re going to be done in batch. Very sensitive processes. Those kinds of things require, they basically have a high cost if you get it wrong. So those are some pretty cool use cases. I mean that puts into perspective, I think, scale. What does it really mean to do a scalable IoT use case?

Tom White (26:05.035)
Mm-hmm.

Tom White (26:14.095)
Yeah, I think scale was a subjective term. And when someone talks about four and a half million cents as a second and so on, I mean, that really does separate, you know, scale from scale, right? So without meaning to go into the secret source, of course, but how does this all work then? So we spoke about some cool use cases, as you say. But how is it all possible?

Michael Gilfix (26:39.466)
Well, one of the things that our engine is really good at doing is paralyzing data processing. And that’s really the only way that you can scale this kind of technology is you have to be able to take the incoming data and partition it out to a whole variety of different computers that can process that pipeline of data in real time and then achieve

eventual consistency. Eventual consistency means, you know, I hand it out to these different places, it gets processed, and then it gets combined into a meaningful way for analytics. And that thing, when it gets combined, has to make sense. Those are eventual consistency systems. By the way, that’s different than, say, traditional databases, like transactional databases. A transactional database, most people use that in things like banking. Like, you gotta make sure that if I said that I paid you,

and something failed, then I don’t pay you twice, right? I got to make sure the data, that’s not an eventual consistency system. That’s a, I’m right above all else. So eventual consistency means it’s okay if the data is not perfect, but eventually it’s got to get there. And I can then scale those things out in order to go and process that volume of data. So that’s one key component to it. As I mentioned, having an innate understanding of time series analytical data is very important for a lot of these IoT use cases.

because you’re looking again at patterns over time, you’re looking at forward prediction, you have to be able to combine your data over context windows, you might have to be able to run simulations. And so you need a system that understands how to do that kind of analytics, but again, how to paralyze it. So that’s part of our secret sauce is we built this engine that will enable you to really paralyze your data processing like this. We have built in kind of concepts of how we divide up this time-based processing and…

Tom White (28:19.564)
Mm.

Michael Gilfix (28:32.906)
We also can store that data in memory, so it’s super fast, as well as on disk for historical. And we’re very clever about how we enable you to run analytics that span your real time in memory data with your historical on disk data to drive your conclusions. And our engine is highly tuned for that kind of stuff.

Tom White (28:49.943)
Got it.

Tom White (28:53.547)
Yeah, I mean, it sounds it, right? Which is important when it comes to, you know, understanding where, how, and to position this. Which leads me on quite nicely to my next question about, you know, ideal data processing points. You know, is it at the edge? Is it in the cloud? What are the trade-offs between the two? We’ve got a lot of talk around the edge at the moment and the abilities to speed things up at the edge, but.

Is there a con to that in your view?

Michael Gilfix (29:28.686)
I think it depends on what kind of analytics you’re doing that are in your model. Our technology is sort of set up so that it could work in maybe both cases. We have technology, we’re not gonna run, say, on an embedded device. So if you recall, in a typical edge architecture, you sort of have like three tiers, right? You have the sensor or the physical device itself, and then you have the analytical processing that’s close to the data.

And then maybe it gets aggregated or you send less data on to the central analytical platform. That’s a typical edge thing. Right. And if you think about that, it’s like a funnel of data volume. You know, the closer you are to the source, the more raw data you get. And then as you move sort of further away, you have less data. Um, and usually the data is summarized or it’s processed in some way so that you can get to the higher order analytic, like, uh, looking at, uh, I don’t know. Uh, uh,

say consumption, like how much do we charge for energy this month? That’s a higher order analytic, right? In the case of energy and utilities. So our technology can kind of run maybe on both of those tiers, the one that’s closer to the data. And a good example of that would be, you can’t do the fab use case unless the analytics are running pretty close to the fabrication floor like that, because you need to be able to get that data to the analytical source. I think the automotive one is in a way a good example of something.

that is sort of in between. It’s not quite edge, but it’s not higher order, you know, business analytics, but it’s something in the middle where, you know, they’re doing historical simulation, they’re running things live in the racing data, but there’s activities that go on in between races where they refine their model, they run historical analytics to figure out better strategies. So it’s a little bit in between, I think, of those things. So we see ourselves sort of playing in both places. I think…

It really just depends on the kind of analytics you’re running. I think even some of the examples I gave would still benefit from some of those architectures where, you know, you got to be closer to the data than the case of extreme data volumes, but it doesn’t take away from the value of running the higher order business analytics. Like in that energy and utilities example, someone still needs to say, here’s how much this neighborhood consumes in energy and they still need to have demand forecasting so that, you know, they can go negotiate contracts with their partners that doesn’t change.

Michael Gilfix (31:48.694)
those types of analytics, even if the meters are providing real time visibility into your daily energy consumption. You sort of have both things in that architecture.

Tom White (31:57.939)
Yeah, yeah, absolutely. I think, you know, the processing of the data and where you position it, you know, for me, it’s similar to, in a way, you know, good policies around data protection, and actually it needs to be relevant and right. And it’s the same as where you start that process. So actually, in one use case, it might be applicable to do it both at the edge or, you know, slightly further back.

And it really does depend on that individual circumstance and the benefit to be derived from that at that point. So yeah, I completely understand. So basically the answer was it depends.

Michael Gilfix (32:32.594)
Yeah. I mean, look, probably it’s a data volume and latency are some of the primary characteristics that drive your decision. If you have extreme volumes, if you need low latency, you’re probably going to have to be closer to the data source, right? The further you go up that pipeline, the more delay is there. But like I said, typically you’re mixing a higher order analytic in place.

Tom White (32:40.633)
Hmm.

Tom White (32:46.282)
Yeah.

Tom White (32:54.611)
Yeah, yeah, absolutely. So it’s a fascinating business and you’re about, is it 500 people now? Is that right as a company or more or less?

Michael Gilfix (33:03.338)
Yeah, we’re over 500 people now.

Tom White (33:05.043)
Over 500, yeah. And so what’s the future for the business? You know, developing more, you know, enhanced more accurate, you know, versions of the analytics, et cetera, et cetera.

Michael Gilfix (33:17.686)
Well, a few things. One is we want to make it easier for organizations, for people of all types in an organization to harness the power of the analytics themselves. And so we’ve been sort of on a mission to unlock different developer tooling, which opens up, for example, data science folks as an audience.

So just to contrast it, you know, historically you’d need sort of a developer who would sit there and write some code to write the analytics, but we see that as not being kind of the way we want. We want the data science crowd to do it. We want the power users to do it. We’re even working, for example, on a copilot plugin that will allow you to natural language go and query your data so that, you know, an end user can ask questions. And so that’s sort of the pursuit of democratization, if you will.

How do we enable more people in the organization to make sense of that data? I think that’s really exciting because if you can, you know, once you have the data there, the more people you can empower to make sense of the data, the more applications it opens up and you know, just this fortuitous cycle, right? Of, of driving additional requirements. I think the, the second thing is, um, you know, the goal I think of every enterprise is to harvest the entirety of their data sets.

And a lot of data is basically untapped in organizations. And so I think our mission is to give you the power of these analytics, but ensure that it can work across the entirety of your data estate. And so an area we’ve been investing in is actually looking at vector database technology, which allows you to process unstructured data, like documents and images. It’s very different than, I think, classic IoT data, but there is a pretty strong relationship.

between those. Like if you think about sometimes like asset management systems and how there’s going to be a live IoT component where you evaluate equipment in the field and then you mix that with here’s images of what might be broken, you know, on like think about like sensors on a bridge and then what’s broken on the bridge and then how do I know where it is. So there’s a lot of unstructured data as well that you could pull into the analytic and that’s a pretty interesting opportunity.

Michael Gilfix (35:36.17)
essentially allowing enterprises to use more of their data estate. So those are two exciting venues. I think we have, you know,

Tom White (35:41.283)
Yeah, that.

Yeah, yeah, absolutely. I think getting into the first one for me, I think is really poignant, you know, getting it into the hands of people to make it more accessible, make people be able to understand the data that they’re throwing off more accurately is something that’s been crying out for such a long time, actually, and people not necessarily knowing how to interpret this, how to use it, and how to make decisions based on it. So I think that’s such a

Michael Gilfix (36:10.946)
To me, that’s one of the great possibilities of this AI revolution we’re in, right? In the advances in natural language processing is we’re kind of at a point now where we’re about to open up the accessibility for such a much larger audience to make sense of their data. I think that’s gonna drive tremendous demands for

Tom White (36:15.39)
Mmm

Michael Gilfix (36:32.482)
things we haven’t even anticipated, because those people will have business use cases and user use cases and business process use cases and other things that we haven’t foreseen. And now we have sort of like a huge envelope of creativity compared to what we had before.

Tom White (36:48.159)
Yeah, yeah, absolutely. And what a way to wrap up the episode. I think that’s such a poignant phrase. You know, it almost is limitless what we can do with this moving forward. It’s been a pleasure having you on, Michael. Honestly, it’s been so interesting to take a kind of brief snapshot and deep dive into KX and some of the technology and the tools.

that you guys have been developing. And I’m sure our listeners and viewers have absolutely loved it. As we come to the end of the podcast, we always ask a series of questions, a little bit lighthearted. So I’ve got a couple for you, Michael, if that’s okay. What is a challenge that you’re facing in your everyday life, that there just isn’t a tech solution for at the moment, that you’d love to see change?

Michael Gilfix (37:21.498)
Uh oh. Hahaha.

Michael Gilfix (37:34.818)
that there isn’t a tech solution for, but I’d love to see change. Okay, this has nothing to do with work, but since I shared that I use Whoop, so I’ll tell you why I used Whoop. My main hobby outside of work is I compete in powerlifting. And so strength training sports, it’s really hard. If you do running and heavy cardio, it’s really easy to find equipment that can actually help you with that.

Tom White (37:53.013)
Oh wow.

Michael Gilfix (38:04.514)
But when it comes to strength training, it’s very difficult. And, uh, well, whoop still calculates your strain. I mean, let’s be honest, it’s kind of limited actually for strength training purposes. Uh, they actually acquired a company called push, which was designed for velocity based training where it looks at the speed.

of your movements, but it’s still not very accurate, I think, in terms of that. And it’s gotten very mixed reviews for velocity training. So there’s a whole opportunity for a type of sport for which I, I’m not sure that we have a good tech solution for it. It has yet to be properly digitized.

Tom White (38:37.259)
Hmm, that’s interesting. And it’s funny to say that as well, because I do notice that on mine. So I also like weightlifting and the strain levels are never actually that accurate. And you want, and you.

Michael Gilfix (38:50.87)
By the way, there’s two ways you can input it in the whoop thing, right? You can either do it as an activity, that’s the worst, or there’s like another thing, if you find the strength trainer, you can use that. That one’s a bit better, but it’s still very limited, I think, for what’s there. So if your audience, if anyone thinks they have a good idea, buy me on LinkedIn. I am happy to beta test your product because I really want to use it. So that would be great.

Tom White (38:54.175)
Okay. Yeah.

Tom White (38:59.907)
Okay, okay, interesting, interesting.

Tom White (39:14.069)
Well, someone’s going to be thinking of it. We had someone on the podcast about a year or two back, and I haven’t seen this in commercial gyms yet, but you had a profile and it basically had an RFID. When you walk closer to like a machine, it would know your routine, right? It would know that you want to be on whatever weight and it would move things around and it would adjust the seat much like a memory seat in a car.

to actually what lengths you needed to be to. And I thought that was really cool actually, but I’ve never seen it anywhere. But maybe in 10 years time, we’ll get something like that, right? And it kind of hits the mainstream. A couple of others. So a gadget that you can’t live without, and you cannot say your mobile phone.

Michael Gilfix (39:58.814)
Oh boy. My earpiece.

Tom White (40:04.839)
I was going to say that earpiece has been in and it was in on our discovery call, right?

Michael Gilfix (40:09.986)
Yeah, I use that all day versus the built-in speakers that are in my laptop. And I, I’m someone who travels pretty frequently for business. And so I’ve trained myself now that, uh, I just work off my laptop with my earpiece so that my work environment is always the same, regardless of my physical location, which is very counter, like a lot of people with monitors and stuff like that. But for me, then it’s not the same. So I don’t know why, but I’ve trained myself to do that. So I, I definitely live off of using Bluetooth stuff and I hate holding a phone to my ear.

Tom White (40:26.965)
Okay.

Tom White (40:40.191)
Do you know what? I can’t do it. It just feels such an unnatural thing to do now. And I actually get our make as well. I’m like, obviously just ridiculous.

Michael Gilfix (40:40.354)
Can’t live without that.

Michael Gilfix (40:49.7)
I don’t have the arm endurance anymore to hold the phone up. I’ve lost my…

Tom White (40:52.571)
No, it’s bizarre. Could you imagine, like 20 years ago, we used to have long phone conversations holding the phone, right? When even loudspeaker wasn’t a thing. It’s just like, how did we even do that? Well, you did being a power lifter, of course. And actually, I think I’m probably gonna know the answer to the last question, but the last question is something that you’re passionate about outside of work.

Michael Gilfix (41:03.634)
I know we just, we had very over-developed forearms and biceps back then. So.

Michael Gilfix (41:19.126)
Oh, well, yes, I already claimed my singular one, I think so.

Tom White (41:23.551)
Yeah, excellent. You’ve been doing it a long time.

Michael Gilfix (41:27.434)
Uh, I started competing about, uh, four or five years ago, actually, I think when I hired a coach. So, uh, and I’ve been competing in a national level and, uh, even though I’m getting older, I’m still getting better, so it’ll be exciting to see what happens this year, but it’s a lot of work, a lot of eating is required. Uh, a lot of resting and training. So it’s, uh, it’s the thing I do, I think outside of work, you got to mix up your techish stuff and sitting sedentary with, uh, something physical.

Tom White (41:30.911)
Nice.

Tom White (41:35.904)
Nice.

Tom White (41:40.64)
Nice.

Tom White (41:48.48)
Oh nice.

Tom White (41:53.103)
Do you know what, you’re absolutely have to, right? I’ve just started in the last couple of weeks doing mobility work every morning because I kept having to go to the physio and I basically said, it’s nice as it is coming in and seeing you, I wanna stop coming in here. And he said, do you do much stretching? Do you do much mobility? And I was like, no, not really anything. And that since I’ve started doing that, I’ve actually found myself feeling a lot less aches and pains.

been able to move things around. So I completely get it. What if it, when we’re in the tech world? Yeah.

Michael Gilfix (42:22.414)
To me, this is a great tech problem, right? I mean, we just don’t have great technology to aid with these things, and you have to go and see someone physically for it. I think the more you could self-service at home, I think increasingly, you know, the popularity of fitness has grown. There’s just, there’s a tremendous opportunity in the fitness industry, I think, and it’s almost untapped, because I think the quality of the data that we have today is relatively low.

Tom White (42:27.51)
Mm.

Tom White (42:40.568)
Hmm.

Tom White (42:46.295)
Hmm, there’s something there. Again, if you’re listening and you’ve got some ideas, connect with me or Michael on our LinkedIn. I’d love to hear as well. Yeah. Yeah, and if it’s really good, I might throw some money into it as well. So there we are. Great. Michael, thank you so much for coming on to the IoT Cosmetic. It’s been a blast.

Michael Gilfix (42:55.028)
Yeah. I’m happy to be your design partner, your beta customer zero. So yeah.

That’s right.

Michael Gilfix (43:09.046)
Thanks for having me.

Tom White (00:01.267)
Michael, welcome to the IoT Podcast.

Michael Gilfix (00:05.378)
Thanks for having me. Excited.

Tom White (00:07.443)
I’m excited as well, actually, because we have a lot of people that come onto the podcast who are talking about IoT when it comes to IoT devices or a more of an industrial landscape when it actually comes to the physical sensors, right? But I think what we’re going to get into here is more the data analytics and the speed of those data analytics. And that’s such a crucial element to IoT, as we all know.

But some of the tech that you guys are doing in your business is phenomenal. But rather than me trying to explain it, I’ll ask you to kindly do that. So as a way of kicking off, could you explain who you are and what company you represent?

Michael Gilfix (00:47.886)
Sure, so my name is Michael Yelviks. One second, can we edit that out? I forgot that one minute. Let me just, can we restart?

Tom White (00:53.911)
We can, we can.

Tom White (00:59.211)
That’s all right, you go, you go. Do you need to go or?

Michael Gilfix (01:02.782)
Well, I’m just going to move him to a different room. That’s my elderly cat. So I’m going to move him to a different room. So we don’t listen to that during the podcast. Can you give me one second? I should have thought of that. Be right back. Sorry.

Tom White (01:05.12)
Okay.

Okay, no problem, that’s fine. Yeah, yeah, we’ll start again. That’s okay, don’t worry. All right.

Michael Gilfix (01:50.702)
Okay, let’s try that again. I put them in a closet, hopefully we’re not gonna hear them. Ha ha ha.

Tom White (01:51.471)
It’s all good. Ah, okay. All right, cool. Let’s just go from the top, I think. It’s probably the best, so I’ll just do that now. Welcome to the IoT Podcast, Michael.

Michael Gilfix (02:01.099)
Yes.

Michael Gilfix (02:08.958)
All right, well thank you so much for having me. Super excited.

Tom White (02:12.307)
I’m super excited as well because often we have people coming on to the podcast who are talking about the physical sensors in IoT, either from a smart meter in context or industrial, so on and so forth. But today we’re going to get into data analytics and more importantly, the speed of analyzing that data. So rather than me trying to attempt to explain it in my rather layman tone, I’ll ask you to do it, Michael. So as always, could you start by saying who you are and what company you represent?

Michael Gilfix (02:43.338)
All right. So I’m Michael Yelfix. I’m the chief product and engineering officer for a company called KX. And a little bit of what we do, uh, as you pointed out, we’re on the analytics side. So we have a technology that makes it really easy to take large volumes of data. And analyze them for business value. And one of the secret sauces of our technology is not only can we take these large volumes of data, but actually we can analyze them in time and help to gain insight from that behavior over time.

Turns out that’s super useful for IoT use cases.

Tom White (03:16.555)
Yeah, absolutely. I think it’s paramount, actually, I would say, in terms of the integrity and the analytics of that data done as fast as possible and accurately as possible. But it’s quite an interesting story, actually, about how the business came about and the origins of the company. So could you talk a little bit about that, Michael?

Michael Gilfix (03:36.462)
Sure, especially in the case of IoT, you might say, as I described the history, so how does this get to be an IoT business? But the origins of our business is, we were born out of one of the most demanding use cases, which is implementing capital markets for financial institutions. Capital markets is the buying, the selling of equity and managing those investments.

And if you think about those kinds of use cases, there are use cases that have a tremendous amount of volume. You get market data, stock tick data. You’re trying to make sense of the world around you and you’re using that to figure out, is this a good trade? Is it a good order? How can I do better? How do I build algorithms that can exploit that? And we built a data platform that actually allows you to perform incredibly well in those kinds of use cases.

So we can process very large volumes of data. We have sub-second response time. So you can figure out the right algorithm. We were able to validate models that run on the data super computationally intensive, but we still make all that stuff work. And you can run your rules in kind of in real time on that set of data. So you can make the right trade or analyze the best trade. And at the heart of that technology was also the ability to process time series data.

And time series data is basically looking at behaviors over time and pattern matching that data or finding ways to reason and predict what’s going to happen with that data over time. So fast forward, you know, many kind of years later, we’ve been bringing that technology to a variety of different industries that benefit from the power of this analytical platform and scale. And

It turns out that IOT use cases are just a tremendous opportunity for that technology because it has many of the same characteristics of that core use case, large data volumes, predicting how things are going to behave or looking at patterns of behavior over time, a sub-second response time, and the kind of scale and price performance that our technology brought.

Tom White (05:45.439)
Yeah, I think for me, it’s really fascinating because it’s not a world that I know incredibly well. So I’m a device guy. I came from embedded Linux. I know video, I know IoT and devices that kind of a low level embedded component system level, shall I say pretty well. And it wasn’t really until I listened to a couple of podcasts about Bloomberg, you know, and the Bloomberg terminal and the necessity for speed around this. So I can really see.

that the origins of the company are rooted in something absolutely fundamental, critical for that industry. As in literally seconds can make a difference and colossal that difference in terms of trades and what the trades look like. So it’s actually quite a unique for us segue into IoT that we’ve not really had many times in five seasons of the podcast, right? But when you talk about it, it seems so obvious, doesn’t it?

Michael Gilfix (06:43.85)
Well, you know, since you background the hardware person and it sounds like a lot of listeners are, I mean, everything these days is becoming virtualized. The hardware is only one component. It’s what you can do with the data that is incredibly valuable. So just to give a simple consumer example, and this is one from my own personal life. So I’m gonna hold up, see this, I gotta get some credit by the way, for advertising this. This is a whoop. If you’ve ever heard of a whoop, do you know what that is?

Tom White (07:09.195)
Ah, I’ve got one here.

Michael Gilfix (07:11.186)
Oh, you got one too. Okay, great. So you know exactly what I’m talking about. So for the listeners, a whoop is a device that, you know, monitors your, uh, your bio data effectively. And it helps you to track things like sleep, strain from workouts, general performance. So I use it for my athletic pursuits, but for those of you that actually own the device, one of the things you’d realize very quickly is that the hardware is actually like a really, really small portion of the device experience.

Tom White (07:14.165)
I do.

Michael Gilfix (07:38.962)
It’s actually recording this data. It’s uploading it to the cloud for analytics. The application is helping you understand baseline performance. They look at trends, like they survey you to say, what characteristics behavior did you have? And then they correlate that with your sleep performance. But all of that analytics is happening in the cloud and it’s being bulk uploaded. And so that’s maybe a really simple way to understand kind of the feeling of, you know, the hardware is really one portion. It’s the enabler, if you will, the on-ramp of the data.

But all of the power came from funneling that data into a platform that could make sense of it and then turn it into an application that someone could use. I think in the case of KX, we’re looking at a level of scale, probably that well exceeds, you know, whoop in our applications, but the real power again is taking that raw data and turning it into the kinds of analytics that open up just tremendous business possibilities with that data.

Tom White (08:32.843)
Yeah, that’s a great analogy, you know, and I’m addicted to mine. I mean, that’s a whole other topic about how it becomes a self-fulfilling…

Michael Gilfix (08:41.238)
Man, if you’re really goal-oriented, there’s nothing like having something critique your sleep performance to go, man, I gotta really, you know, gotta get on the ball on this thing. So it’s very useful.

Tom White (08:47.151)
Yeah, yeah, yeah. Oh, absolutely. Yeah, absolutely. I often worry and wonder sometimes if my recovery is below 70%, if I then think, oh, I’m going to have a bad day because of it. And sometimes I wish I’d never seen it, you know, but that’s a whole other topic. So going back to KX and the business in terms of the actual industries that are most prevalent now. So we spoke about your origins of where you came from.

Where do you see most of the activity coming from the data processing using your technology today?

Michael Gilfix (09:22.21)
So we have four industries that we see driving a fair amount of demand for us. One is manufacturing. We can talk about these a bit more detail, but I’ll just give you kind of an overview quickly. So manufacturing, so thinking about how you’re going to manage the equipment data that comes off of a production line, especially if it’s high tech manufacturing, like semiconductor fabrication, for example, highly, highly automated. The second is energy and utilities.

Tom White (09:33.345)
Yeah.

Michael Gilfix (09:50.102)
So being able to take information around metered electricity, related things that are driving energy consumption using that data in real time, tons of amazing applications emerging there. Healthcare life sciences, rife with deep data to understand about what’s going on with patients, what’s going on with procedures, how do you feed that back into analytics or even clinical data that can be monitored over time. And then automotive.

Uh, those are some really fun ones. It’s, uh, like I want to run simulation on my race cars. And while the race, so not only am I going to run simulations, let’s say in wind tunnels to provide performance, but I’m going to look at the race in real time and then run simulations to predict future performance and all that’s running in analytics in real time and each of these platforms have the characteristic of giving off, uh, effectively large amounts of data, uh, in a short period of time. So, you know, there’s a.

You need to be able to keep pace with it. You need to be able to make sense of it. And I think when you dig into each of these cases, there’s some unique requirements as well on how you handle data and be smart about its processing.

Tom White (11:00.039)
Yeah. So before we get into the individual industries in a bit more depth, is it primarily industries, so take automotive for instance, race cars, telematics, is it primarily where data, the decisions need to be made quickly? Is that primarily where

people are using and harnessing KX’s technology above and beyond other solutions that they might have out in the open market.

Michael Gilfix (11:30.846)
That’s certainly one of the drivers, but it isn’t the only driver. Um, one of the things that we’re really good at doing with our technology is a form of simulation. It’s when you run in parallel, lots of different scenarios to pick the optimal one, that’s not a true real time thing. Um, but obviously you want your simulation to end in a reasonable amount of time, but you’re running a large set of scenarios simultaneously. Our system’s really good at paralyzing that so that we can figure out an optimal

scenario. But I would think the two top ones are sort of either you’re uh, you’re either monitoring actually let’s do three. Sorry, you’re either monitoring making sure that everything’s functioning or you’re running a form of real-time analytic or you’re running a set of simulation to figure out let’s say optimal scenarios with key devices. I would put under monitoring by the way predictive maintenance for example for machinery is under monitoring as well.

When’s equipment going to fail? That one’s kind of obvious, but if you’re in say high tech manufacturing and you’re working on a fab production line, you obviously want to make sure that the process is going to succeed and downtime is extremely costly, right? For those kinds of highly automated production lines. And so your goal is to detect issues in the production line and try to resolve it as quickly as possible.

Tom White (12:55.367)
Yeah, yeah, no, absolutely. I think that’s really interesting how it’s, you know, that it’s not just a driver around speed, which, you know, could be forgiven for thinking about sometimes, right? Going into a bit of a deep dive into energy and utilities then. So can you talk to me about the real time meter monitoring? And also, we spoke about this in our discovery call around the game of occasion, which I thought was really interesting, actually.

Michael Gilfix (13:22.366)
Yeah, so let me give you kind of an example. We have a use case where literally our technology is monitoring all of the meter consumption for an entire country. And the meters report back somewhere between five minutes to an hour per meter, but there’s literally millions of data points that are being reported back simultaneously that need to be aggregated. And one of the things about meter monitoring is that

the order of the messages back really matter because otherwise you could draw the wrong conclusion around consumption trends. And consumption trends are typically what drives a grid. Like how do you value, if you think about like a lot of grids have things like surge pricing, right? Depending on the consumption, the pricing sort of scales, depending on the utilization of the grid. Likewise, you want to use that to forecast demand. And so that’s a running average, if you will, right? From taking the grid. And so it’s really important that

the data be ordered correctly in order for the forecasting to work. And it turns out when you have a ton of these different meters running, messages get lost. They come in out of order, right? And you imagine that at a massive scale, the analytics platform has to make sense of all this data. But once you get the data, there’s some really cool stuff you can do. So, uh, one of the things that you can start to do is plan your energy consumptions around peak periods. So, uh,

I mean, look, Tesla is trying to build a business model around this, right? You get the advertisement that says, Hey, would you like to sell your data back to the grid or would you like to store it during off hours where it’s cheaper and then use it in peak hours? And so you can be more energy efficient and green if you were just smarter and had better transparency into your data. One of the things I like, we were talking a bit about gamification. So I live in Austin, Texas.

For those of you that don’t know, Austin is really, really hot in the middle of the summer. And especially as temperatures have kind of gone up, I think Austin has gotten more prolonged days in extreme heat. I have to remember my Celsius conversion. So for those of you that know Fahrenheit, we have over a hundred degree days during the summer for an extended period of time. What is that? Is it like 40 Celsius? 42? Okay, yeah. Let’s call that the good enough conversion.

Tom White (15:41.031)
42 I think. Yeah, it’s hot. It’s hot, it’s hot. Yeah.

Michael Gilfix (15:46.462)
Needless to say, it’s pretty hot. We have really good air conditioning technology in Austin, Texas, but when it gets really hot like that, the energy demands on the grid go up substantially. Right. And there’s a lot of tension. So one of the things they implemented locally, once you get real time visibility into this consumption, one of the things they implemented locally is, uh, they would, uh, pitch you against your nearest neighbors. So I would get an email that saying, here’s how you performed versus the top hundred houses in your local neighborhood.

in a bid to encourage me to be more energy efficient and would rank me based on that week, let’s say, or that day, here’s how you did versus your neighbors. And that’s a great way to incent homeowners to be more energy efficient because you might look at that and say, first of all, my bill became astronomical during the summer because of surge pricing. And for those that live in Texas, they may know that by the way, Texas implements some things where the surge pricing can get pretty ridiculous past a clip level.

So not only do you get this expensive bill, but you’re dangled this great carrot of, and it doesn’t have to be that way. If only I found ways to make the thing more efficient. And that’s a pretty interesting way to use data. And if you think about the, how did we get there? Like step one was, you know, instrument all of these different sensors to pull back the data. Step two was process it so you had real time visibility.

Tom White (16:48.835)
Hmm.

Michael Gilfix (17:07.838)
And then step three opened up, well, once these new applications, you know, once I had the real time visibility, what could I do with it? Gamification, uh, being smarter about, let’s say buying and selling things on the grid for those that are, are green inclined, just tremendous opportunities, right? To be more efficient and the average homeowner has no idea because how would they know what the basis of comparison is? What incentive would you have?

Tom White (17:32.985)
Hmm.

Michael Gilfix (17:34.306)
For all you know, you’re the best at being green because you put some weather stripping in or something. But once you see how it compares to your peers, okay, well now you have a basis of comparison, you’re a little more motivated.

Tom White (17:45.875)
Yeah, yeah, oh absolutely. I think, you know, the gamification really does help because naturally people are competitive and naturally people want to do the right thing. So it serves two purposes by doing that, but I think it also works on scale. I mean, the start of that you said that you, I think, correct me if I’m wrong, but you received the data from a whole country’s

Michael Gilfix (18:14.746)
Yeah, in that particular case, it’s a European country where we’re actually running analytics in real time on the entire meter infrastructure.

Tom White (18:22.731)
I mean, that is absolutely colossal. And what a coup to be able to say that as well, right?

Michael Gilfix (18:29.802)
Yeah, it’s one of the Scandinavian countries. I think that’s as far as I’m allowed to go in advertising it, but, uh, it gives you a size of the scale. But it, for me, that’s someone who builds products. That’s a pretty exciting thing. Cause how often you get to talk about your product effectively powering an entire country. I mean, not everybody gets to say that, but, but think about that as. That’s a truly scalable. IOT use case. I mean, I think in the IOT market.

Tom White (18:32.875)
No, that’s alright. I’m not probing it. I don’t want to know who it is.

Tom White (18:47.509)
Yeah.

Tom White (18:53.058)
Hmm.

Michael Gilfix (18:54.282)
It’s been talked about smarter cities and stuff like that for a long time. This is actually in practice with regards to providing energy consumption and visibility.

Tom White (19:02.891)
Yeah, oh, absolutely. No, it’s fantastic, fantastic. So moving back to the automotive use case that you mentioned as well then, Michael. So we spoke about racing cars, the wind tunnel simulation, et cetera. So one of the things that I know from racing cars is, as I mentioned, the telematics, the need for low latency and for understanding things in incredible detail, because when you scale this up to…

your Indy cars or your F1s, it really does make a massive difference of course. So it’d be great to dig into KX’s involvement in automotive.

Michael Gilfix (19:40.958)
Yeah. So, uh, one of the most public companies we have in that space is Alpine, the French racing company and, uh, their use case has effectively two components to it. The first is wind tunnel simulation, meaning the goal there is to take the cars with their data, put them in the wind tunnel, simulate performance, and then effectively build a basis of data by which they can build a model of vehicle performance.

Tom White (19:46.384)
Okay, yeah.

Michael Gilfix (20:10.534)
And so you run a batch simulation, you get a bunch of data, you can offline batch process that data, and like I said, you can figure out these predictive characteristics for it. So think of this as that’s simulation time. That’s race prep time. Then what they do is during race time, as the vehicle’s going around doing laps, they’re taking the live telemetry data off the vehicle, they’re taking the predictive model that came from the batch simulation, and they’re making the call on.

You know, should I do pit stops? Um, what’s the likelihood of part failure? Like how long do I have to go before the tires give out? And they’re using that to inform strategies. And then what they do, there’s a third component is they keep the history then of how they did on race day. And that becomes now historical data. And they use that to further tweak for future races. So again, you have the simulation input. You have the, here’s what I ran on race day.

And then you have the historical, well, these are the strategies I ran on race day and the inputs and how did I do so it can inform what you might see in an individual scenario again on race day. It becomes sort of a, uh, I guess, uh, a circle that turns on itself, right? I can, uh, continuous, uh, improvement kind of circle. And so they’re, they’re feeding that data in to go and improve the, the racing component.

Tom White (21:25.26)
Yeah.

Michael Gilfix (21:31.082)
So that might give you a sense of sort of how the sensor data comes, because again, you have that batch kind of input from the simulation, but then you have tons of telemetry data that’s getting pumped off those vehicles while they’re in the race itself. And you got, that’s where real time really matters because you want to make sure that you’re calculating car performance and you’re coming up with the best racing strategy based on the data of the car.

Tom White (21:37.291)
Mm.

Tom White (21:54.175)
Yeah, yeah, I mean, it’s absolutely critical. And again, working with someone like Alpine, you probably learn a lot about how to perfect the software and how to perfect how it works in that environment as well. Right. And what is really important and what isn’t because it’s being done, you know, at a competitive enterprise, world-class level, right.

Michael Gilfix (22:17.438)
Yeah, and I think what’s critical for those kinds of things, you know, a lot of the data that comes in from those sensors is a form of signal processing. Like you’re looking for patterns in the signal over time because, you know, break temperature, measures of wear, handling, you’re looking at the behavior of those components over time and you’re looking in patterns in the signal effectively to figure out when do you act, not just predictive, but also looking for patterns on equipment

you know, failure and those kinds of things. So you need to be able to do a fair amount of time series processing on that data because the performance over time matters as much as it takes for predictive performance. And then those things need to be computed in real time, right, whatever the model is that you use, the rules, you’re running some pretty intensive time-based calculations when you go and do that.

Tom White (23:08.159)
Yeah, yeah, I can imagine. Yeah, crikey. Yeah, it’s really quite awe inspiring. I think that the sheer scale and size of teams’ budgets like Alpine to invest in these things and the differences can be made actually from harvesting the data, as we say. The other use case we spoke about.

was the manufacturing side. Now this was really interesting for me because traditionally speaking, we have a lot of people talking about digital twins on the podcast and digital twins are great because it’s really easy to replicate something without the costs of being able to do it in real life first and to look at stressors or issues that can be contained in a virtual environment. But what I really liked about KX’s solution is around that real time sensor.

around failures and so on. It would be great to dig into that as well.

Michael Gilfix (24:05.046)
So I mentioned earlier that we’ve been doing some work with some semiconductor manufacturing. And let me give you an idea of the two requirements that come out of working on those kinds of use cases. So the first is the sheer volume. So I thought I knew what scale was until I looked at some of those things, but we right now have tested for production line.

upwards of four to four and a half million sensors per second that gets thrown off these things that need to be processed. So that’s, that’s some pretty intensive data volume that’s being transmitted. And, and like a large number of data points that have to be handled. And then maybe the second thing is you have to have technology that’s extremely highly available because for, for our production lines don’t.

go down the way that regular software maintenance windows are. You’re not taking them down for a couple hours, right? Even if you think about things like cloud up times, cloud up times might have four nines, but you know, that’s still a couple hours, you know, a year. They, they might take two years to get to a 15 minute upgrade window. So the availability has to really be massive and because it’s, you know, the cost of taking down that production line is very intense.

Tom White (25:22.563)
Hmm.

Michael Gilfix (25:22.922)
So those are some pretty harsh characteristics that have to balance the data that comes out of that kind of an application. Just again, the sheer volume of it, the speed, the high availability. And all of that is basically used to drive automation, monitoring, it’s a key part of modern manufacturing being substantially more efficient, right? Especially high-tech manufacturing where it’s not like people are minting chips, right? Those things are pretty automated.

Tom White (25:50.639)
Hmm.

Michael Gilfix (25:53.218)
They’re highly small, they need to be done, so they’re going to be done in batch. Very sensitive processes. Those kinds of things require, they basically have a high cost if you get it wrong. So those are some pretty cool use cases. I mean that puts into perspective, I think, scale. What does it really mean to do a scalable IoT use case?

Tom White (26:05.035)
Mm-hmm.

Tom White (26:14.095)
Yeah, I think scale was a subjective term. And when someone talks about four and a half million cents as a second and so on, I mean, that really does separate, you know, scale from scale, right? So without meaning to go into the secret source, of course, but how does this all work then? So we spoke about some cool use cases, as you say. But how is it all possible?

Michael Gilfix (26:39.466)
Well, one of the things that our engine is really good at doing is paralyzing data processing. And that’s really the only way that you can scale this kind of technology is you have to be able to take the incoming data and partition it out to a whole variety of different computers that can process that pipeline of data in real time and then achieve

eventual consistency. Eventual consistency means, you know, I hand it out to these different places, it gets processed, and then it gets combined into a meaningful way for analytics. And that thing, when it gets combined, has to make sense. Those are eventual consistency systems. By the way, that’s different than, say, traditional databases, like transactional databases. A transactional database, most people use that in things like banking. Like, you gotta make sure that if I said that I paid you,

and something failed, then I don’t pay you twice, right? I got to make sure the data, that’s not an eventual consistency system. That’s a, I’m right above all else. So eventual consistency means it’s okay if the data is not perfect, but eventually it’s got to get there. And I can then scale those things out in order to go and process that volume of data. So that’s one key component to it. As I mentioned, having an innate understanding of time series analytical data is very important for a lot of these IoT use cases.

because you’re looking again at patterns over time, you’re looking at forward prediction, you have to be able to combine your data over context windows, you might have to be able to run simulations. And so you need a system that understands how to do that kind of analytics, but again, how to paralyze it. So that’s part of our secret sauce is we built this engine that will enable you to really paralyze your data processing like this. We have built in kind of concepts of how we divide up this time-based processing and…

Tom White (28:19.564)
Mm.

Michael Gilfix (28:32.906)
We also can store that data in memory, so it’s super fast, as well as on disk for historical. And we’re very clever about how we enable you to run analytics that span your real time in memory data with your historical on disk data to drive your conclusions. And our engine is highly tuned for that kind of stuff.

Tom White (28:49.943)
Got it.

Tom White (28:53.547)
Yeah, I mean, it sounds it, right? Which is important when it comes to, you know, understanding where, how, and to position this. Which leads me on quite nicely to my next question about, you know, ideal data processing points. You know, is it at the edge? Is it in the cloud? What are the trade-offs between the two? We’ve got a lot of talk around the edge at the moment and the abilities to speed things up at the edge, but.

Is there a con to that in your view?

Michael Gilfix (29:28.686)
I think it depends on what kind of analytics you’re doing that are in your model. Our technology is sort of set up so that it could work in maybe both cases. We have technology, we’re not gonna run, say, on an embedded device. So if you recall, in a typical edge architecture, you sort of have like three tiers, right? You have the sensor or the physical device itself, and then you have the analytical processing that’s close to the data.

And then maybe it gets aggregated or you send less data on to the central analytical platform. That’s a typical edge thing. Right. And if you think about that, it’s like a funnel of data volume. You know, the closer you are to the source, the more raw data you get. And then as you move sort of further away, you have less data. Um, and usually the data is summarized or it’s processed in some way so that you can get to the higher order analytic, like, uh, looking at, uh, I don’t know. Uh, uh,

say consumption, like how much do we charge for energy this month? That’s a higher order analytic, right? In the case of energy and utilities. So our technology can kind of run maybe on both of those tiers, the one that’s closer to the data. And a good example of that would be, you can’t do the fab use case unless the analytics are running pretty close to the fabrication floor like that, because you need to be able to get that data to the analytical source. I think the automotive one is in a way a good example of something.

that is sort of in between. It’s not quite edge, but it’s not higher order, you know, business analytics, but it’s something in the middle where, you know, they’re doing historical simulation, they’re running things live in the racing data, but there’s activities that go on in between races where they refine their model, they run historical analytics to figure out better strategies. So it’s a little bit in between, I think, of those things. So we see ourselves sort of playing in both places. I think…

It really just depends on the kind of analytics you’re running. I think even some of the examples I gave would still benefit from some of those architectures where, you know, you got to be closer to the data than the case of extreme data volumes, but it doesn’t take away from the value of running the higher order business analytics. Like in that energy and utilities example, someone still needs to say, here’s how much this neighborhood consumes in energy and they still need to have demand forecasting so that, you know, they can go negotiate contracts with their partners that doesn’t change.

Michael Gilfix (31:48.694)
those types of analytics, even if the meters are providing real time visibility into your daily energy consumption. You sort of have both things in that architecture.

Tom White (31:57.939)
Yeah, yeah, absolutely. I think, you know, the processing of the data and where you position it, you know, for me, it’s similar to, in a way, you know, good policies around data protection, and actually it needs to be relevant and right. And it’s the same as where you start that process. So actually, in one use case, it might be applicable to do it both at the edge or, you know, slightly further back.

And it really does depend on that individual circumstance and the benefit to be derived from that at that point. So yeah, I completely understand. So basically the answer was it depends.

Michael Gilfix (32:32.594)
Yeah. I mean, look, probably it’s a data volume and latency are some of the primary characteristics that drive your decision. If you have extreme volumes, if you need low latency, you’re probably going to have to be closer to the data source, right? The further you go up that pipeline, the more delay is there. But like I said, typically you’re mixing a higher order analytic in place.

Tom White (32:40.633)
Hmm.

Tom White (32:46.282)
Yeah.

Tom White (32:54.611)
Yeah, yeah, absolutely. So it’s a fascinating business and you’re about, is it 500 people now? Is that right as a company or more or less?

Michael Gilfix (33:03.338)
Yeah, we’re over 500 people now.

Tom White (33:05.043)
Over 500, yeah. And so what’s the future for the business? You know, developing more, you know, enhanced more accurate, you know, versions of the analytics, et cetera, et cetera.

Michael Gilfix (33:17.686)
Well, a few things. One is we want to make it easier for organizations, for people of all types in an organization to harness the power of the analytics themselves. And so we’ve been sort of on a mission to unlock different developer tooling, which opens up, for example, data science folks as an audience.

So just to contrast it, you know, historically you’d need sort of a developer who would sit there and write some code to write the analytics, but we see that as not being kind of the way we want. We want the data science crowd to do it. We want the power users to do it. We’re even working, for example, on a copilot plugin that will allow you to natural language go and query your data so that, you know, an end user can ask questions. And so that’s sort of the pursuit of democratization, if you will.

How do we enable more people in the organization to make sense of that data? I think that’s really exciting because if you can, you know, once you have the data there, the more people you can empower to make sense of the data, the more applications it opens up and you know, just this fortuitous cycle, right? Of, of driving additional requirements. I think the, the second thing is, um, you know, the goal I think of every enterprise is to harvest the entirety of their data sets.

And a lot of data is basically untapped in organizations. And so I think our mission is to give you the power of these analytics, but ensure that it can work across the entirety of your data estate. And so an area we’ve been investing in is actually looking at vector database technology, which allows you to process unstructured data, like documents and images. It’s very different than, I think, classic IoT data, but there is a pretty strong relationship.

between those. Like if you think about sometimes like asset management systems and how there’s going to be a live IoT component where you evaluate equipment in the field and then you mix that with here’s images of what might be broken, you know, on like think about like sensors on a bridge and then what’s broken on the bridge and then how do I know where it is. So there’s a lot of unstructured data as well that you could pull into the analytic and that’s a pretty interesting opportunity.

Michael Gilfix (35:36.17)
essentially allowing enterprises to use more of their data estate. So those are two exciting venues. I think we have, you know,

Tom White (35:41.283)
Yeah, that.

Yeah, yeah, absolutely. I think getting into the first one for me, I think is really poignant, you know, getting it into the hands of people to make it more accessible, make people be able to understand the data that they’re throwing off more accurately is something that’s been crying out for such a long time, actually, and people not necessarily knowing how to interpret this, how to use it, and how to make decisions based on it. So I think that’s such a

Michael Gilfix (36:10.946)
To me, that’s one of the great possibilities of this AI revolution we’re in, right? In the advances in natural language processing is we’re kind of at a point now where we’re about to open up the accessibility for such a much larger audience to make sense of their data. I think that’s gonna drive tremendous demands for

Tom White (36:15.39)
Mmm

Michael Gilfix (36:32.482)
things we haven’t even anticipated, because those people will have business use cases and user use cases and business process use cases and other things that we haven’t foreseen. And now we have sort of like a huge envelope of creativity compared to what we had before.

Tom White (36:48.159)
Yeah, yeah, absolutely. And what a way to wrap up the episode. I think that’s such a poignant phrase. You know, it almost is limitless what we can do with this moving forward. It’s been a pleasure having you on, Michael. Honestly, it’s been so interesting to take a kind of brief snapshot and deep dive into KX and some of the technology and the tools.

that you guys have been developing. And I’m sure our listeners and viewers have absolutely loved it. As we come to the end of the podcast, we always ask a series of questions, a little bit lighthearted. So I’ve got a couple for you, Michael, if that’s okay. What is a challenge that you’re facing in your everyday life, that there just isn’t a tech solution for at the moment, that you’d love to see change?

Michael Gilfix (37:21.498)
Uh oh. Hahaha.

Michael Gilfix (37:34.818)
that there isn’t a tech solution for, but I’d love to see change. Okay, this has nothing to do with work, but since I shared that I use Whoop, so I’ll tell you why I used Whoop. My main hobby outside of work is I compete in powerlifting. And so strength training sports, it’s really hard. If you do running and heavy cardio, it’s really easy to find equipment that can actually help you with that.

Tom White (37:53.013)
Oh wow.

Michael Gilfix (38:04.514)
But when it comes to strength training, it’s very difficult. And, uh, well, whoop still calculates your strain. I mean, let’s be honest, it’s kind of limited actually for strength training purposes. Uh, they actually acquired a company called push, which was designed for velocity based training where it looks at the speed.

of your movements, but it’s still not very accurate, I think, in terms of that. And it’s gotten very mixed reviews for velocity training. So there’s a whole opportunity for a type of sport for which I, I’m not sure that we have a good tech solution for it. It has yet to be properly digitized.

Tom White (38:37.259)
Hmm, that’s interesting. And it’s funny to say that as well, because I do notice that on mine. So I also like weightlifting and the strain levels are never actually that accurate. And you want, and you.

Michael Gilfix (38:50.87)
By the way, there’s two ways you can input it in the whoop thing, right? You can either do it as an activity, that’s the worst, or there’s like another thing, if you find the strength trainer, you can use that. That one’s a bit better, but it’s still very limited, I think, for what’s there. So if your audience, if anyone thinks they have a good idea, buy me on LinkedIn. I am happy to beta test your product because I really want to use it. So that would be great.

Tom White (38:54.175)
Okay. Yeah.

Tom White (38:59.907)
Okay, okay, interesting, interesting.

Tom White (39:14.069)
Well, someone’s going to be thinking of it. We had someone on the podcast about a year or two back, and I haven’t seen this in commercial gyms yet, but you had a profile and it basically had an RFID. When you walk closer to like a machine, it would know your routine, right? It would know that you want to be on whatever weight and it would move things around and it would adjust the seat much like a memory seat in a car.

to actually what lengths you needed to be to. And I thought that was really cool actually, but I’ve never seen it anywhere. But maybe in 10 years time, we’ll get something like that, right? And it kind of hits the mainstream. A couple of others. So a gadget that you can’t live without, and you cannot say your mobile phone.

Michael Gilfix (39:58.814)
Oh boy. My earpiece.

Tom White (40:04.839)
I was going to say that earpiece has been in and it was in on our discovery call, right?

Michael Gilfix (40:09.986)
Yeah, I use that all day versus the built-in speakers that are in my laptop. And I, I’m someone who travels pretty frequently for business. And so I’ve trained myself now that, uh, I just work off my laptop with my earpiece so that my work environment is always the same, regardless of my physical location, which is very counter, like a lot of people with monitors and stuff like that. But for me, then it’s not the same. So I don’t know why, but I’ve trained myself to do that. So I, I definitely live off of using Bluetooth stuff and I hate holding a phone to my ear.

Tom White (40:26.965)
Okay.

Tom White (40:40.191)
Do you know what? I can’t do it. It just feels such an unnatural thing to do now. And I actually get our make as well. I’m like, obviously just ridiculous.

Michael Gilfix (40:40.354)
Can’t live without that.

Michael Gilfix (40:49.7)
I don’t have the arm endurance anymore to hold the phone up. I’ve lost my…

Tom White (40:52.571)
No, it’s bizarre. Could you imagine, like 20 years ago, we used to have long phone conversations holding the phone, right? When even loudspeaker wasn’t a thing. It’s just like, how did we even do that? Well, you did being a power lifter, of course. And actually, I think I’m probably gonna know the answer to the last question, but the last question is something that you’re passionate about outside of work.

Michael Gilfix (41:03.634)
I know we just, we had very over-developed forearms and biceps back then. So.

Michael Gilfix (41:19.126)
Oh, well, yes, I already claimed my singular one, I think so.

Tom White (41:23.551)
Yeah, excellent. You’ve been doing it a long time.

Michael Gilfix (41:27.434)
Uh, I started competing about, uh, four or five years ago, actually, I think when I hired a coach. So, uh, and I’ve been competing in a national level and, uh, even though I’m getting older, I’m still getting better, so it’ll be exciting to see what happens this year, but it’s a lot of work, a lot of eating is required. Uh, a lot of resting and training. So it’s, uh, it’s the thing I do, I think outside of work, you got to mix up your techish stuff and sitting sedentary with, uh, something physical.

Tom White (41:30.911)
Nice.

Tom White (41:35.904)
Nice.

Tom White (41:40.64)
Nice.

Tom White (41:48.48)
Oh nice.

Tom White (41:53.103)
Do you know what, you’re absolutely have to, right? I’ve just started in the last couple of weeks doing mobility work every morning because I kept having to go to the physio and I basically said, it’s nice as it is coming in and seeing you, I wanna stop coming in here. And he said, do you do much stretching? Do you do much mobility? And I was like, no, not really anything. And that since I’ve started doing that, I’ve actually found myself feeling a lot less aches and pains.

been able to move things around. So I completely get it. What if it, when we’re in the tech world? Yeah.

Michael Gilfix (42:22.414)
To me, this is a great tech problem, right? I mean, we just don’t have great technology to aid with these things, and you have to go and see someone physically for it. I think the more you could self-service at home, I think increasingly, you know, the popularity of fitness has grown. There’s just, there’s a tremendous opportunity in the fitness industry, I think, and it’s almost untapped, because I think the quality of the data that we have today is relatively low.

Tom White (42:27.51)
Mm.

Tom White (42:40.568)
Hmm.

Tom White (42:46.295)
Hmm, there’s something there. Again, if you’re listening and you’ve got some ideas, connect with me or Michael on our LinkedIn. I’d love to hear as well. Yeah. Yeah, and if it’s really good, I might throw some money into it as well. So there we are. Great. Michael, thank you so much for coming on to the IoT Cosmetic. It’s been a blast.

Michael Gilfix (42:55.028)
Yeah. I’m happy to be your design partner, your beta customer zero. So yeah.

That’s right.

Michael Gilfix (43:09.046)
Thanks for having me.

 

About our guest

Michael Gilfix is the Chief of Product and Engineering at KX. He is a seasoned software executive known for his expertise in driving growth and building scalable global software businesses. With a strong track record in product management, sales, M&A, and partnerships, he excels in creating high-performing software portfolios. Michael’s communication skills shine in public speaking and press interactions, making him a standout leader in the industry.

KX is a company that offers a suite of products for high-performance data analysis, including vector databases, time-series analysis, and real-time analytics. Their flagship product, kdb+, is a high-performance database engine that is claimed to be the world’s fastest. KX’s products are used in a variety of industries, including finance, healthcare, and manufacturing.

Find out more about KX – Here

Connect with our guest