John Harrington of HighByte: Stop Making Data Swamps, Start Shipping Chocolate

We talk with John Harrington, co-founder of HighByte, about why factory-floor data stays invisible to the teams who need it most and how Industrial DataOps closes that gap. We explore contextualized data pipelines, the post-IoT architecture shift toward cloud data platforms, and why AI agents will force a new level of data quality and governance. 
• Moving beyond “throw it over the wall” design and giving engineers real manufacturing feedback loops 
• Defining Industrial DataOps and why context makes raw OT data usable 
• Handling messy realities across MES, ERP, historians, inspection systems, files, and streaming telemetry 
• Avoiding data swamps by standardizing, governing, and observing data pipelines at scale 
• Using no-code tooling to build and maintain pipelines without relying on programmers 
• Filtering and sampling data based on use case, frequency needs, and event triggers 
• Preparing for AI agents as massive new consumers of shop floor data 
• Realistic talk on AI and jobs, focusing on better work through better signal detection 

Transcript

Welcome And Guest Background

Roopinder 0:07

Hello and welcome to FoDES, the future of design and engineering software podcast. My name is Roopinder Tara. On the show, we will have guests that will discuss tools and technology that engineers will find interesting and useful. Welcome to the show. John, I'm surprised if we haven't crossed paths already. You were with Kepware and then joined PTC, right?

John Harrington 0:31

Yes. I actually uh worked at PTC back in the mid to late 90s. Actually, I guess it was 94 to 97. Yeah, spent some time with PTC. Fairly familiar with design software, as well as you know, the last um 15 years of my life have been in uh operational manufacturing data and manufacturing software related.

Roopinder 0:56

Yeah, I looked a little bit into your history, besides your starting out in uh with Kepware. I don't think you started out in Kepware, but that was where I first would have noticed you. Are you in Maine? Is that where you started?

John Harrington 1:10

Yeah, I'm in Portland, Maine, and that's where Kepware was based, and this is where HighByte uh is based.

Roopinder 1:16

And you're a co-founder, correct?

John Harrington 1:18

Yeah, of HighByte, yep.

Roopinder 1:20

Okay, all right. HighByte. Yeah, I'm just wondering about that because you know, we're I'm in the Bay Area in San Francisco. We think we're the center of the tech universe. The first person I've talked to at Maine. Is that a lot of tech out there?

John Harrington 1:35

Well, it's growing. It's uh it's exciting. You know, I think there's been a fair amount over the years, but smaller companies. Portland's only about two hours from Boston, so you can always get down there if you need to interact with companies down there and interact with uh investors and whatnot, and press and PR and everything down there. But uh yeah, Portland's has a small tech scene and uh it's growing and it's exciting.

Roopinder 2:04

I forgot about that. Uh proximity to to Boston, and of course, that's where PTC is headquartered, and so that probably helped that relationship. Has DC divested itself of Kepware?

John Harrington 2:16

Yes, it has that just happened in the last couple of weeks.

Roopinder 2:21

And of IoT. I think they took a turn back to say the maybe their core business, maybe, and did that. Okay, but John, you're you are a mechanical engineer, correct? So we have that in common. Yeah. Okay. But you've gone over to the, I'll call it the other side, manufacturing data.

John Harrington 2:41

And that's still very relevant today. And now, you know, I think one of the relevance to this podcast is the ability to provide data to the design engineering teams so that they can better understand what is actually where the time is being lost, where the quality defects are being created, to then loop that back into improving the design, improving the manufacturing, the manufacturing processes to so that you're more efficient, right? You're getting more product out the door, faster, cheaper, less waste.

Why Engineers Need Shop Floor Data

Roopinder 3:20

Yeah. Yeah, I've always felt like the engineers haven't been given enough information for design for to design for manufacturability at all. They treat that as, you know, throwing the drawings or designs over the wall, and then somebody's going to take care of it. And uh consequently, there's a lot of improvements that can be made in manufacturing. We're very curious about what happens now and how we can change that.

What Industrial DataOps Really Means

Context And The Factory Data Puzzle

John Harrington 3:44

We talk obviously in today's day and age, a lot of discussion around uh around AI, but um, highlight is an industrial data ops supplier. So we're a software company, and I'll define what industrial data ops is in a minute, but we're a software company, we provide software to manufacturing companies really so that they can leverage data throughout their business. Whereas 10 years ago or 20 years ago, the focus was how do I automate my factory floor and how do I run that? Today, what we're realizing is if we could extract the data from that, then there's a lot more business functions that could leverage that data. So whether it's the quality team, the manufacturing team, who are who are off the shop floor, um, the supply chain team, the design engineering team, all of them, if they could have rich knowledge of what's happening on the shop floor, when do problems occur, what was happening, when that problem occurred, what are the set points, how are we manufacturing these products, what part of the line or what part of the work cells are we seeing defects? They could all execute their jobs much better. So HighByte software company, it was formed in 2018, really recognizing that we're trying to leverage this data. This was kind of the advent of how do we leverage data for IoT platforms and cloud. And the data wasn't ever designed for that, so we need to make it usable. Uh, to date, we have employees across the United States. We work with customers who are global. We have customers across 24 countries, a really wide range of manufacturing companies, as you can see here. These are just a small subset of our customer base, but really everything from automotive uh assembly companies, manufacturing, to discrete products, food and beverage, pharmaceutical, uh, oil and gas, building products, you name it, really wide range of customers, all of them focused on how do I leverage data. One of the key challenges in today's day and age with industrial data is that you've got all this equipment on the factory floor. It's all on premise. It communicate it's able to create tons and tons of data, just like our phones or our smart watches and everything else create lots of data. So does the equipment on the factory floor. The challenge is how do we get it where we need it? And how do we do that efficiently and how do we manage the flows of data? We've got different applications. We talk about the manufacturing and execution system, the MES system. We talk about ERP Enterprise Resource Planning System that has some data in it. We've got uh inspection equipment on the factory floor, CMM, coordinate measurement machines. We've got um lots of different data here, there, and everywhere. We've got AGVs, automated guided vehicles that are creating data. And how do we use all the information from them to drive our factories faster, to reduce the number of defects, to reduce the number of line shutdowns? But the challenge not only is in the systems, but it's also in the different types of data. Some of it is what we call telemetry, so just streaming data coming up, machinery. You have streaming data coming off, it's just constantly changing. Other data, we've got uh transactions. You know, think of a typical database, and we've got these transactions that are coming off uh, you know, out of the MES system, the RP system, the CMMS, the the um the maintenance systems. Then we've got historians, which are constantly just saving all the storing all this data. How do we get access to that? And then we've got files. We've got cameras, we may have these quarter measurement files, we may have CAD files that we want to pull out of the engineering team's uh PLM system and move that down to the factory floor or G-code files that we want to, you know, to run our manufacturing. So so lots of data and files and whatnot moving around, but it all needs context in order to use it. And you hear about context now with AI as you know, context is really just describing what this data is. So you have a pump. Well, we need to know what is that number? Is it the pressure, is it the temperature, is it um, you know, the state of whether the valve is open or closed, or let's actually bring all that together. Let's grab the pressure, the temperature, the state of the machine, let's also know where it is and whatnot, pull it all together. That's context. So, how do we do this at scale? Um, with of course, we have limited resources. So that's really the biggest challenge. That those these four items are the challenges that we see. So there's been this birth of a new type of or a new category of software called DataOps. Um, data ops really comes out of IT, and data ops is about how do you curate data for the target system. So think it's the next generation of integration platforms. Okay, how do I move data efficiently across all these different systems and recognize that in order to send it to whether it's a LLM or an MES system or a you know an analytic running in the cloud, I need to provide it with a data set that is knowledge that the system is looking for and that has the context so that I know this is pump A versus on work cell one versus pump D on work cell 15. And and I can discriminate those two, and I'm not mixing that data up. I need to keep that separate. So DataOps is all about curating that data, delivering it, but creating an enterprise solution to be able to do that where you're not only orchestrating the data, but you're also making providing observability, you're providing high-quality data, and you're providing governance so that it's controlled. And one of the things you may have noticed in our our customers that I showed you in the last slide is that um, you know, these are global companies, they've got many, many sites. In fact, we have we have two of our customers who've deployed us at over a hundred sites already. And so, you know, how do they orchestrate and govern this across all those sites? So that's you know another key challenge. And so HighByte as a software company has built a product, it's called the HighByte Intelligence Hub, as a software solution to manage this flow of data. And and uh it's focused on being able to connect to the systems that it needs to connect to. So both on the OT side, which is commonly referred to as all of the technology on the factory floor, we've referred to as OT operations technology, and then the IT side, all the technology that's not on the factory floor that could be running in the cloud, it could be running on people's desks or whatnot. The software is really designed to kind of bring those two disciplines together, OT and IT, so that we can easily move data, contextualize data, standardize data, unify data systems across your ecosystem. And you know, highbuddy is really focused on the ecosystem. We partner with a lot of the different vendors in these different applications, as well as cloud vendors like Amazon and Microsoft and Databricks and Snowflake. And you know, we so we it's about moving data across them, it's not about replacing them. So we don't actually analyze or visualize the data, we just make the data usable. So high bike provides this software solution to be able to manage all these interactions with these different systems, bring the data in, add the ability to standardize and contextualize it, and then we create what are called data pipelines to move the data from you know, we're gonna collect this data and this data, and we're gonna transform it in this way, and we're gonna send it out to this target system. And that's that's really what it is. It's a platform that enables companies in a no-code solution to pull data from different systems, transform it however they need to, and then deliver it, and then to be able to manage that. Because you know, the first step is being able to do it, but that's kind of the easy step. Step two is how do I manage and maintain that over time, and how do I do it at the scale that uh that our customer base is dealing with, where they've got hundreds of thousands of data points and they've got many tens or twenty or you know, hundreds of different systems, and I just need to efficiently move and manage that. And of course, that's all getting even more challenging with AI. Greater opportunities, but also greater challenges.

Roopinder 12:13

AI makes it makes all this data usable from what I've heard from other people that have been on the show, is that we could use AI to manage that structured data, but that is in effect what iByte is doing, correct?

John Harrington 12:28

Well, so so AI is is really interesting. Now, we actually have AI built into our solution um so that it can help you build and manage these data pipelines. Um, yeah, it can be very powerful. Um, AI is great at analyzing data. AI is great at um trying to find uh common um objects. So think I identify one pump and I've got this massive namespace of data coming from the system. Uh, can you find the other 500 pumps in in a million data points? And it can go out and it can do that sort of thing. It's very powerful to uh support the configuration and scale that out.

unknown 13:12

Okay.

John Harrington 13:13

AI is also going to be the biggest consumer of data. And and in fact, AI agents, um, AI agents are best when they're uh really focused. So, you know, the future is that at every every machine cell in a factory, we're gonna have an AI agent for maintenance, an AI agent for monitoring quality, an AI agent for monitoring the supply, the raw materials in, um, or even you know, the quantity of goods that are being produced out, or any sort of metrics. But what that's gonna do is that's gonna take the number of consumers of industrial data and move it from maybe uh 10 to 50 to hundreds or thousands of consumers because we're gonna have thousands of agents. Um it creates though it creates a lot of opportunity, it creates a whole new set of challenges. But you know, having a software solution to manage that flow of data actually uh makes those enables a company to implement AI agents without without any control of that data, it makes it very hard. So we see AI agents as a huge opportunity as well as um an enabler of the software.

AI Agents As Data Super Consumers

Roopinder 14:32

I want to go back a few years, probably around five years ago, when IoT became a term and it became a almost a fashion. Like everything's gonna be instrumented, there's gonna be lots of data, every machine would be streaming data. Uh the effect was enormous. Uh, I think, I think uh just the data that was gathered and delivered was so good. They talked about data oceans, right? That uh a manufacturing operation can generate anywhere from, let's say, a terabyte of data a day to 50 terabytes of data a day, right? Yes. And my question, ongoing question was what the hell are you gonna do with all this, right? Nobody knows that's too much data. That's like we're gonna be drowning in that ocean of data. And most of it, if you don't act on it soon, it's just gonna you know that last scene of Raiders of the Lost Ark where everything is in a warehouse? Yes, and it's just buried. That whatever you want, the ark in that case, it's just gonna be buried. You're not gonna find that ark of the covenant. Exactly. Yeah, you know, is that the is that the problem that's a good idea?

Why Industrial IoT Platforms Stalled

John Harrington 15:46

Well, so so yeah, let's build on that. So there are two two challenges with I'll call it the IoT initiatives of, you know, IoT really hit its stride in 2015 and kind of started to die out in 2020. And I think two things happened. Number one, they struggled to get access to the data and the data in a way that was usable. Okay. Okay. So there's one problem which was usability of data. And the other problem is that IoT platforms came in, or and I'm gonna differentiate between industrial IoT and IoT. So IoT platforms are still very prevalent with specific centers for specific problems. For instance, you know, you may want to control the light bulbs in your home, or you may want to uh you may have a smartwatch or whatnot, and you can go to a cloud system and you can interact with that a much greater than dealing with the watch directly. However, in the industrial space, there's lots of different users of that data, lots of different people within the company, engineering, supply chain, quality, and they all have their own existing systems already. So if you're not going to replace those systems, the real value is when you can land the data in those systems or when you can get it to those people where they're at. And they could be, you know, it could be a custom application running in the cloud, but IoT, industrial IoT platforms tended to really be focused on either on one use case, even though they kind of sold themselves as all use cases. So they may be predictive maintenance or maybe quality or whatnot, but but really they would do the whole connect, collect, store, analyze, visualize, but it was really one use case. And then companies realized that well, that's not going to scale. I need multiple platforms for my company. I'm gonna need one for maintenance and one for quality. And so the architecture has changed since then to let's just do the contextualization, standardization of data into the cloud, into say AWS or Azure, and then we'll buy multiple applications to go on top. And actually, I'm gonna I'm gonna show you a slide that kind of talks to this. And let me grab this and say share. So, what we're seeing is a motion from kind of that single IoT platform into more of an industrial data strategy where you've got all of your systems of record and you want to get the data from them and you want to move it up, and then you store it in these massively scalable platforms like Amazon or Azure or whatnot. And then you build applications on top of that. So then you may have ignition as a as an IoT platform that uses this storage, or you may have Grafana dashboards or Seek or Medix. These are all different application vendors that could access it. And you may also have a lot of custom, like, well, you got Power BI or Tableau or other applications. And so what we've seen is a transition from having kind of a full stack vendor, because then I need 20 full stack vendors, and they all need to get access to the same sort of data, and it doesn't make a lot of sense, it's hard to manage, into more of a layered approach where, and this is really where Industrial DataOps fits, where it's able to move data across these systems of record as well as up into the cloud so that the applications can get access to it. This is what we've seen evolve from the uh the concept of the IoT platform. And you know, one of the key challenges, like I said, it was uh connectivity, another one was standardization of data, and then the third one was just scale. The sheer, to your point, the sheer volume of data, we need to put it in something that's really, really scalable, like a data lake, and then build applications on top of that. And if we don't contextualize the data, you started to hear stories about people talking about what they call data swamps. Because if all you do is just push all the data and you don't contextualize it or do anything, it's just not usable. Right.

Roopinder 20:06

Um it's too much to handle, isn't it? I mean, without without like curating it, like you said, it's like you who can deal with that. I'm thinking of uh in the old days when they had camera systems mounted on every machine, and they had 50, an office would have like 50 screens in front of it, but there's not 50 pairs of eyes or 50 brains behind them. So all the time.

John Harrington 20:29

Well, I will say, agents, agents are getting better at that. So agents are kind of providing the ability to AI agents provide the ability to have a dedicated viewer of the data for a specific task. And then, like a quality engineer could essentially put an agent on every single work cell, it could be looking for certain problems or defects or whatnot, and then report up when they have a problem so that the engineer doesn't have to be standing next to every single work cell to know whether the work cells are running properly or not. Um and that that's really the future of AI. But yeah, I think hopefully that gives you some exposure to kind of what happened to IoT. It's it's just evolved, it's transitioned.

Roopinder 21:13

Everyone's talking about agents now. So is I can totally get that. You need an agent because there's not enough people, intelligent people, to monitor all these feeds. So an agent would be perfect for that. Agent does it never sleeps, never wants a raise, you know, never calls and sick, that kind of thing. You know, and doesn't and pays attention. I mean, I can't I can't believe anybody would want to look at a camera feed for a machine for more than an hour, right? So exactly, exactly.

John Harrington 21:43

In fact, we um we have a customer, Lint Chocolate. And so Lint Chocolate makes chocolate lender balls, and they are manufactured in a tray that I'm not sure the exact dimensions, but it's something like you know, 15 by 15. 15 balls. So you've got somewhere multi-hundred uh linder balls in one tray, and they have an operator who would look at the tray to try to identify a defect. Now these trays aren't moving really fast, but even every 15 seconds, having to look at a tray of over a hundred uh items, in this case chocolates, and to identify small defects is very hard for a human, minute after minute, hour after hour, day after day. But it's very easy for an AI. An AI, you know, can do that sort of thing, can analyze, we can just take a picture, we can move that to the cloud, they can analyze it, and then they tell the human, it looks like there are some defects in this sector and in this. And so the person just goes and pulls those out or decides not to because the defects aren't bad enough. So, you know, we often talk about AI, just it turns humans into superhuman. It makes us much more effective and much more efficient. And, you know, you still have a person on the on the line who's who's pulling out those defects, but they don't have to be tasked with scanning it because it's just too much.

Roopinder 23:10

I love those lead chocolates. I always have a bag of bag of them. Hopefully they give you some samples, John, if they're customers.

John Harrington 23:19

Yes, actually they uh they produce them down in uh New Hampshire, so not far from us.

Roopinder 23:24

Oh, not bad.

John Harrington 23:25

And we've been down to the plant. They actually produce them globally. Obviously, um they're using our software both in the US as well as in Europe.

Roopinder 23:32

But uh yeah, you you are certainly well located. So there's enough smarts to uh analyze a video image.

John Harrington 23:40

Yeah, in this case, if they're taking snapshots. Um, and in this case, they're moving it up into the cloud, but we have other customers who will analyze that at the edge. Uh we call it at the edge in the factory um on a computer right beside the camera. And then, you know, there's lots of different uh approaches for that. But you know, either way, we you know, we're often involved in moving the data, whether it's an image or whether it's just the results of uh the calculations and analytics.

Roopinder 24:11

I was at a conference two years ago, and in there they mentioned a system, and it made the you made me think of this, that a system that would, I think at that time they were aspiring to a system like this, but it sounds like something like this may already be in place. Tell me how this sounds. They would they had a system that would listen to a machine, and if it would detect an inappropriate sound, meaning that it was going to probably uh malfunction in they could tell, somehow they could tell how long it was gonna last. They would alert the operator, they would there's zero in on the part that was malfunctioning, they would alert the purchasing to start ordering the part, they would go into the supply chain and figure out that hey, this part is a let's just say China and it's gonna be difficult to get, right? Here's what you could do and and present the whole thing, like from and even schedule downtime or whatever what was needed. It was like, wow, this is really cool. I thought that was a fairy tale at the time, but is that something that's not possible?

John Harrington 25:18

No, um, that's something that is very actively being being worked on, and uh you know, certainly monitoring supply, how much supply do we have, how much is it being consumed, and getting much more granular with you know how much is being taken out of the actual bins beside the operators, and and the accuracy of all that has gone up dramatically, and then being able to send out uh supply requests, yeah.

No Code Pipelines And The Intelligence Hub

Roopinder 25:45

Yep, it's it's exciting things. Is Hype then it does it have a product or does it come in it uh that actually it can put in place, or is it gonna be uh uh should I say something that you guys would help a company add and it's a process? How does it work? And and then also when I ask about how how does it fit into what already exists because the US has its solutions, you know, there's also MES solutions and ERP solutions, and they all a lot of them are claiming that they do these sort of things, right? So are you guys like the middle layer here between the factory and and the higher end exactly, exactly.

John Harrington 26:28

Um you can see my dashboard now, is that correct? Okay, so this is just a dashboard, it's showing me what I'm connected to, how things are going, if I have any alerts that I have to react to, if how much data I'm sending around, and and it's how I manage the system. But what's really important is I'm gonna jump into this broker view where you can see my data. So so first off, I want to start here with a raw data set that I'm just pulling off of my machinery. Okay. Now, some of this data has some context in it, but it's it's coming at me constantly, and it's really hard to know what everything is, and it's hard to organize it. Okay. So this is how it gets started. Well, what would be really nice actually is if it looks like a data set like this. So here I've got you can see the site, you can see the area, you can see you know where this, what cell this is in, you can see certain key data about this work cell, you can see the state of it, you can see maybe the weather. So a very verbose contextualized data set that if I put this in an analytic or if I put this in a dashboard, it would tell me exactly what I'm looking at. And that's that's the key thing is that we take that raw data and we turn it into something that's usable. Let me jump back into here. The way that we do that is we organize the data in what's called a namespace, and then and we basically are just dragging and dropping data sets. And here is a motor uh data set, and you can see here I've got my instance, and if I do a quick test of my instance, that data set for that motor is being defined right here in my system. So you define what data coming off of what system gets defined for these different attributes, and then you build up the data set, and then you can send that out at some frequency. So the key is if I jump back here to to build up these data sets in an organized way that you know, if I hand them off to my maintenance engineer or I hand them off to my quality engineer, they can work with that data as opposed to at the beginning, I just had that you know, raw kind of data spewing at me.

Roopinder 28:42

So you're very adept at using this, but I'm I'm looking at it like a foreign language, right? Yeah.

John Harrington 28:51

It's all a no-code interface. Um here it's it, you know, you identify the different stages that you want to do to the data. You drag and drop these stages in here. I'm pulling data out of my namespace and then I'm just writing it out to a target. And um, you know, if if I want, I can I can actually see what data is being written out. I can let me just do a quick view of this. See the data payload. Exactly. It's drag and drop like Vizio. You may need to do a little bit of configuration here, yeah, but there's no code writing. Um, this is what we call a data pipeline.

Roopinder 29:31

No code. This isn't even low code, this is truly like no code, drag and drop.

John Harrington 29:36

Yep. Okay, um, you know, you can have much more complex data pipelines doing lots of different things, but the key is that you know you're you're building up this visual view of what I want to do with this data. You can build up data sets. That's what here I've got, uh I've defined a standard structure, and then um I apply it and I can just view what that looks like. So instead of having it all kind of hidden, it's very easy to troubleshoot, very easy to define drag and drop. Again, this is very much drag and drop where you drag from your list of inputs and just drag it into your uh references.

Roopinder 30:19

Okay. That's that I'm uh delighted to see this. Still, I have to I I'd expect to spend some time with it before I've actually it's useful, right? What is the what is the burden or training period for for something like this? And who does it? Is it the IT staff or is it the process engineers who actually work?

John Harrington 30:38

That's a great question. So a system like this really spans both the IT and the OT teams. And so a lot of times the IT teams will define this is my standard model for, in this case, a thermostat. These are the attributes that I need when we're sending it up to, say, Amazon, into S3. I want this data set every time I send a thermostat. And then when we're defining the instances, that would be often like a controls engineer or process engineer. Because the key with defining the instances is you need to know the source systems. Where am I going to get this data from? How do I need to transform it in order to deliver it? And so it certainly it's often an engineer, but it's not a software engineer. It's not a programmer, as much as it's an engineer who knows the systems, knows how the data is structured, how they can get access to it, and is able to, as you've seen with a low-code to no-code solution, just drag and drop and build up your data, your instances and then your uh your data pipelines to deliver the data.

Roopinder 31:48

So it's better all the time. Now tell me this. Okay, have you heard this phrase or something like this? So I'll probably butcher this. War is long periods of total boredom punctuated by sheer terror. And uh, I liken that to machines on the factory floor. Like you can be watching it, monitoring it for forever. And then there's one time where you need to be there. And so what I'm going with this is like all that 99% of that data is useless. So does hypite act to filter out all that unnecessary thing right at the edge and only send in critical parts? Like all I need to know is a timing of that analogy, the severity, and that it exists to recognize it, the severity, that kind of thing. I I don't need all that other data, right? So is that filtering?

Filtering Data By Use Case

John Harrington 32:40

So we can do that sort of filtering. Yeah. And and the key with filtering the data is actually thinking about what's the use case that I'm trying to solve for. For instance, we have customers who are collecting extremely high speed motor quality data, power quality data going feeding a motor. Think of a uh paper machine. In order to detect that there are issues with the motor, you can look at the power quality that's being fed into it and the the um the volume that it's consuming to identify issues with the motor. However, in that data, you want to collect it at like one millisecond. But you don't need that forever, and you don't even need that continuous. You may need a one-second burst, so a thousand records of one millisecond data, and then you want to grab that every 10 minutes or every hour or every day just to analyze to make sure that the motors are operating correctly. Because at the frequency, at the frequency that those motors are running, you need really high resolution, but you don't need it continuous. So the key is that first you can define this data I want to collect at an extremely high rate and move it, you know, but not continuous. Other data you may need every 10 minutes because it's for a management dashboard. Other data, when a batch is complete or when a part is complete being produced in a work cell, then you may want to grab all the data that was collected while it was in that work cell and then move that up into the target system. So the key is that it's not that you're putting crude filters, it's that for different use cases and different targets and different consumers of the data, you're able to filter it differently.

Roopinder 34:30

So it's very adaptable. You could it's not so much as like a software product, like a Pro E or something, or a Creo, or that's like you you could you would adapt it to the specific function or task that it needs to.

John Harrington 34:44

Yes. Yep. Yep. And that's what's really important about data is that it's curated for the consumer. And that's one thing that's really critical is that people think about how are we using this data, and therefore what data do we need, what frequency do we need it? You know, is it sort of event-based? Say when when a part is complete in a work cell, then we collect the data, or is it continuous and at what frequency?

Roopinder 35:11

Did you have more uh that you wanted to cover?

John Harrington 35:14

We added the ability to uh have an agent. So I can actually have an AI agent right here, right beside me. So I could say to this agent, I could ask it, what does this pipeline do? Let's say that I was coming in here as an engineer who maybe set this up but hasn't hasn't touched it, as you said, for weeks or months, and all of a sudden it's starting to throw errors. I can easily ask it uh questions and it'll tell me, for instance, what that agent's doing or what that pipeline's doing.

Roopinder 35:43

I can say, you know, add so a natural language interface is built in.

John Harrington 35:48

Yes, add a buffer stage after.

Roopinder 35:52

So it'll accept commands that it knows as well as just natural language, English.

John Harrington 35:58

Exactly. Well, it takes natural language English, and and so I can say, you know, add a stage, or I can ask it what's going on with this pipeline. It has access to the latest runs. And so what we're doing is we're integrating AI into the product so that you know it can improve the experience of the person who's configuring it, and then I can either accept that stage or I can reject that stage and discard it or whatnot. So I can make additional changes. So we really see AI helping the user. We call that AI for data ops, as well as enabling the solution enabling AI agents, like I was telling you earlier for the engineers.

AI Assistance Inside DataOps Tools

Roopinder 36:42

I'm very jealous. I wish some of these design software would have the natural language interface like you have just shown me. This is this every every software, you know, they they have 300, 400 to 400, 500 commands, and I have to learn them every time I learn a program. And I don't want to do that. I don't want to, I want this. I want I want to be able to ask at things intelligently.

John Harrington 37:05

It's extremely useful for people who, you know, maybe come into the software once a week, need to understand what's happening, need to uh make some slight modifications, or need to add a new data pipeline. It's very useful.

Roopinder 37:19

You mentioned Lin. Who else is using your your software?

John Harrington 37:22

Um yes, they tend to spin all of manufacturing and tend to focus on large large companies.

Roopinder 37:30

Okay. Yep. Uh services, too. I see FedEx. Now they're that's more or less a process-driven company, right?

John Harrington 37:39

So well, it you know, it turns out FedEx has a lot of uh distribution centers. And a distribution center is very similar to a manufacturing plant, except that they're taking trucks and they're taking all the packages off the truck, they route them to other locations and lay, you know, load them back on other trucks and the high, very high-speed equipment. Um, but it's very similar to manufacturing. And of course, you know, no one wants a FedEx plant to go down uh the beginning of December. Uh, that would be a major problem if the distribution center stopped. So the key for them is um predictive maintenance and being able to maintain their equipment when the when the distribution center is not running, so that when it's running, it runs at full speed.

Roopinder 38:27

So just looking at this slide and and looking, and then uh it makes me think of all the variety of machines and processes you have to deal with and industries.

John Harrington 38:40

Yes.

Roopinder 38:41

You have to be like a Swiss Army knife then for all of these, so so many different things, right? This is not like a one machine to another integration. This is uh or you're not standardized in any one industry or specializing in one. You have to everything to everyone. That's that's a huge challenge, right?

John Harrington 39:01

Yeah, I mean, you know, there's some commonality across uh manufacturing, the equipment on the factory floor has some commonality to it, and they're different commonality and applications, but every plant is unique, even within one customer. You go into one plant versus another, they're all very unique. And so uh the the intelligence hub software was really designed to be able to work in that environment. You know, that was the environment that uh myself and the other co-founders of High Byte knew and really designed the software to to meet.

Roopinder 39:35

Yeah, I have to ask, I ask everybody about this AI thing, but don't so don't feel single now. It's like I always hear about AI helping humans with their jobs, and from the humans, I always hear about AI is going to take my job. So you're obviously on the side of it's helping humans, right? But do you have to deal with places that are doing the reverse, that are actually automating to the uh damage, should I say, of uh a workforce? Is that something you deal with at all?

John Harrington 40:10

Obviously, there's there's a lot of talk around that. I've been in the technology environment my entire career, technology companies and and technology companies just like um CAD software got rid of a lot of drafters back in in the day, back in the 80s uh and 90s. But you know what we see is a lot of the data usages weren't ever utilized in the past. So data was available for the factory floor operators and for the management team, maybe, who's running the factory floor, but the quality team didn't get access to any of that data. The engineering team didn't get access to any of that data. And if they did, they would have this massive volume of data and not be able to make sense of it. What AI is doing is it's enabling us to process all that data. And so AI is able to process all that data and provide data. So you still will have the same number of quality engineers, but they'll be so much better at their job. You'll still have the same number of design engineers, but they'll be so much better at their job because now they've got access to what's happening in real time. How do I design for better manufacturability if I have no idea where my bottlenecks are and I have no idea where my defects are being created? And now we're able to deliver that data to them in real time and to systems and to AI that can actually process that. And so it can what AI is very good at is finding the signal in the noise, if you will. Okay. And what humans are not good at is finding the signal in the noise because we just get overrun. Yeah, as you said at the beginning, there's just so much data available. But with the AI, we can filter that down, and then we can enable the humans to do their jobs better. So, you know, will some people lose their job? You know, I'm sure there will be some of that. But we, you know, I like to think of it as we're providing technology so that the people can do their jobs better and the companies will be more efficient, and therefore we can produce product at a lower cost with fewer defects, with less downtime, so the companies can be more, you know, we can bring more manufacturing back to the United States. It may require less manual work, but we're able to do it better.

Roopinder 42:38

Have you seen that that cartoon where somebody, a salesman, let's say, a case, a salesman is presenting the wheel to other people, and the people are pulling, let's say, a raft along the ground, dragging a raft along the ground because they're too busy to think about a new technology like the wheel. So is that an issue now? Because although as good as this sounds, you go into a potential customer and they're just too busy to look at a solution. To they're you know, too busy getting keeping their factories humming, you know, dragging their factories along as they were.

Closing Thoughts And How To Reach Us

John Harrington 43:14

You know, I think every person in a factory, every person in a company has what's really important to them right now. And that's what's top of mind. And if you can find the people whose top of mind issue is how do I get, how do we become more efficient, how do we leverage information to improve our quality, to reduce our costs, to improve our uptime, then those are the people that are thinking about solutions like what we provide. Um, those are the people who are trying to put together platforms so that they can leverage data to drive their business. Um, there are other people who may be thinking about how do I make sure that my that my workforce. Shows up tomorrow. And in those cases, they're not in the right headspace to be dealing with us. So, you know, every every company has its challenges, and uh every person within the company has the challenges that they're trying to solve. And the the key is really to find the person that's trying to solve the challenge that we're we're providing. And um HighByte provides a data platform, a data ops, industrial data ops platform that makes leveraging data so much more effective in manufacturing companies. The challenge is finding those people.

Roopinder 44:32

Sounds really good. I wish you continued success. Thanks a lot for being on the show and explaining things on a very basic level that even I can understand. It's been great having you on the show.

John Harrington 44:43

Thank you for the opportunity to uh speak with you. It's nice to meet you.

Roopinder 44:48

Thank you for listening to FoDES. the Future of Design and Engineering Software Show, brought to you by ENGTechnica. I hope you have learned of a new application or technology that will help you with your job. If you have an application you think would be of interest to other engineers, please let me know by emailing me at roopinder at enchtechnica.com or message me on LinkedIn.