Skip to content

Your Machines Are Talking. Pittsburgh-Based Ramp AI Helps You Listen.

Interview by Jonathan Kersting

AMP'd

What if your machines could explain why they stopped working—and help your team find the problem faster?

On this episode of TechVibe's AMP’d Pittsburgh, host Jonathan Kersting of the Pittsburgh Technology Council sits down with Joel Reed, CEO of Ramp AI, to explore how purpose-built artificial intelligence is helping manufacturers understand their equipment, troubleshoot faults and preserve hard-earned expertise.

Born out of Premier Labs and its relationship with Premier Automation, Ramp AI combines machine manuals, maintenance records, technician knowledge and programmable logic controller (PLC) code to help operators and engineers make sense of complex industrial systems. Reed shares findings from an initial pilot with Pittsburgh-area train-coupler manufacturer M&T, including reported fault-resolution times approximately 6.5 times faster.

In this conversation:

  • How industrial AI helps teams move from confusing symptoms toward root causes.
  • Why capturing veteran technicians’ knowledge matters for the next generation.
  • How both junior technicians and experienced engineers use diagnostic assistance.
  • What separates purpose-built manufacturing AI from general-purpose chatbots.
  • Where machine intelligence could go next, from understanding legacy equipment to supporting new installations.

It’s a practical look at AI built to assist people—not replace their judgment—and why Pittsburgh’s combination of manufacturing experience, automation expertise and AI talent makes it a compelling place to build it.

Listen for a shop-floor perspective on smarter troubleshooting and the future of advanced manufacturing.

The Pittsburgh Technology Council produces TechVibe to explore Pittsburgh's technology and manufacturing industries.

Transcript:

Introduction: Ramp AI and Joel Reed
---

[00:00:00] 

Jonathan: But really, what we can claim is that we're seeing about six and a half time faster time to resolution on faults. This is Jonathan Kersting right here with AMP'd Pittsburgh. We're talking about advanced manufacturing Pittsburgh. This is one of my favorite series to do, 'cause my heart lies so much in Pittsburgh's industrial roots and the technology we've been developing here for the past 50 years. And today, we have a company that I think is living exactly in the middle of this awesome Venn diagram where technology overlaps industry, and I am talking about a company that is called Ramp AI that came out of PremierLabs.io.

This is a multi-level story. Yeah, that's right. And I have Joel Reed here, who is the CEO of Ramp AI. And Joel's got quite the roots in Pittsburgh's technology and robotics ecosystems. It's pretty amazing stuff, and I'm... I think you're living your best life right now, Joel, doing what you're doing, leading up Ramp AI, like all these different things you've done over the years.

Everything I did- Yeah. ... up until now, yeah, [00:01:00] put me in this position. That's exactly right. It's so much fun to see it, and when I got this news release about Ramp AI being deployed in a local manufacturer that's been around for about 150 years, I was like, "Oh, we gotta get Joel to stop by," 'cause this is exactly the story I wanna tell on AMP'd.

And if you're talking about the story, I'm talking about the idea of advanced technology, AI, helping old-school manufacturers Make better stuff. Simple as that. At the end of the day, it's like your machines are talking to you, but you can't hear them all the time. This technology allows you to hear what the machines are doing so you can keep on top of it all.

I know I've way oversimplified it, but I just- You teed up our tagline. I love it. Thank you very much. I did my research. I did my research 'cause like I said, I was really nerding out just looking about what you guys are up to. And this all kinda starts with Premier Labs, which which spun this company out- Yeah, that's right

which is cool. So tell us about Premier Labs and what that's about, and your role with it, but then we're gonna jump into how Rampant AI is just really, for lack of a better term, ramping things up- Yeah ... for some local manufacturers. Yeah, ramp, ramping up, the name is purposeful. Yeah. First of all I have to thank you, [00:02:00] Jonathan- Yeah

for exploring this area and starting to put a greater light and more focus on, and bringing your energy to it as well- thank you ... 'cause- I'm trying, man ... 'cause it's really important. As Southwestern Pennsylvania has all of these strengths and- It do ... all of these capabilities, and we're doing things across the world that most of us don't know about in manufacturing and industrial places, so- Exactly

and we have to talk about it, right? We do- and yeah ... because no one else does- That's right ... because here's the problem is you guys are just so busy kicking butt that there's no time to do the storytelling. So that's why- A lot of great stories ... but- So m- for on behalf of the entire community, thanks for telling these stories.


Premier Labs and Premier Automation
---

Jonathan: Even better yet. Thank you, Joe. Yeah. I appreciate that. So Premier Labs. Yeah, tell us about Premier Labs, yeah. It's a venture studio which is a relatively, still relatively new what's called an asset class in investment- ... and commercial creation but that is really earlier than an incubator and an accelerator.

So it typically is happening within an industry, and novel concept of talking to customers- Yeah ... embedded with customers to really understand what their most pressing problems are, and then to start to innovate and ideate around that and come up with [00:03:00] proof of concepts. And then what a venture studio will do is that they have all the structure and resources necessary to be able to help build those companies, grow 'em, and commercialize 'em.

Now, we're a corporate venture studio- ... to add another layer. Yes. And we're affiliated with a really great company in Southwestern Pennsylvania- 100% ... that's got over 20 years of experience and a lot of their leaders, you know, 30-plus l- years of experience globally in industrial settings called Premier Automation.

Exactly. And that really is our- are real competitive advantage, right? Because it put, the things that early stage companies lack, they have technology, they have intellectual property, they have vision. But many times what they have to work to develop over time is an understanding of the market they're going to.

Making sure that their product actually fits what the customer- Yes. ... is looking for. Exactly. I know. And then even when you get beyond all of that is having the relationships with the people that you need to sell to, and then lastly, having the confidence of that market that you're coming into.

And so all of that is baked into this thing called, Premier Labs, which is a corporate [00:04:00] venture studio. ~Yeah. I remember when you guys were starting to cook this thing up a few years ago. I heard about it. We ran into each other at CES, I think, of all places. Yeah. I remember you had just moved over to there, and I was thinking, "This is really exciting," because first off, I don't know how many corporate venture studios are out there that are around the industrial and automation space.~

~And much less in Pennsylvania. I've I hear the term, venture studio a lot out in California and places like that. Yeah. And that, that's cool. But to have this in Pittsburgh, I think is very unique. It's really exploded in the last, three to five years. When we started and we started looking out at the landscape, and it's not a, it's not an an immediately new concept.~

~It's been around for about 10 years, but there were less than 1,000 of these entities- I believe it ... five years ago worldwide. Yeah, exactly. And now it's just exploded, right? Now it's how people do... 'cause they see how this works so well, 'cause it can be so specific- And I think we knew- ... and very targeted.~

~Yeah ... we knew about it in the past as corporate innovation, right? Okay. Or just, building new concepts in, within companies and then trying to spin that out. Exactly. And it's, again it's it's a, a new approach on some older, ideas, and a lot of them that you'll see, and the more popular ones, are ones that are developing software solutions.~

~Exactly. So that- And especially with a lot of vibe coding, capabilities now, and that's all really good. That's, exactly. There's a lot of companies in six months that are, you know- For sure ... achieving incredible valuations. Exactly. And so- But they're not in automation and in manufacturing.~

~In manufacturing, which is a harder, by definition- Damn straight ... market to go after, so- 'Cause Pittsburgh makes stuff, dude. Simple as that, right? Yeah. And that's what makes this so exciting to me because I, when I look at the the challenges that confront our manufacturers- And what they have to do to get product out the door to keep it cost competitive, to keep it efficient, and the crunch on supply chains, everything going on.~

~And when you can bring in advanced technology to make them competitive- Yeah ... come on, man. That's what it's all about to me. I just get excited about that. You've seen the opening promos for, Monday Night Football where they go into our steel mills locally. Yes. I'm going into those environments and- Exactly, yeah.~

~and I'm standing on those platforms seeing the huge cauldrons, pouring, the castings and, building their products and that has been happening all along, even though while we've been developing a lot of this digital and information technology- Yeah ... we still need to make things.~

~We still gotta build the stuff. And I'll tell you, and maybe this is a thread that we can pull on in this conversation- Yeah, it's good ... is that, I would say, y- you know my history of going into warehouse robotics. And I would say that I've had a, a experience working in conventional industries where, adoption of technology 20 years ago was an uphill, battle.~

~Oh, yeah. But that has changed across all of the global industries, and it's true in manufacturing industrial settings as well. I'm glad to hear that. Okay, so you're actually seeing there's more of an open mind to saying- we don't have to educate, right? That's amazing. See, that- Yeah~

~I didn't realize that. I, 'cause I remember back- on some level we do. Yeah, of course. Absolutely. But even AI and especially due to the nature of AI, it's being experimented with and tested every day by engineers on the front lines. Gotcha. And so when you come in and say that we've got a, an AI solution built to solve some of your more pressing problems, we don't have to we don't have to educate on what AI means.~

~It, it's immediately known. We don't have to deal with, built-in bias and objections- Exactly ... because, and we could talk about this as well, there are really compelling, pressing reasons why this technology is needed at this point in time. ~I think we can do all that by talking about how Ramp AI works, 'cause I think this is where I think our viewers and our listeners here are gonna be like, "Wait a second.

This is amazing technology that Ramp AI i- is offering." So give me the general pitch for what Ramp AI does. Yeah. And if we can kinda nerd out on that case study that you guys sent me- Yeah, absolutely ... with M&T would be so much fun- Yeah ... because first off, I'm also a railroad nerd, too. And I was like, wait a second.

There's only four companies in the world that make train couplers? You... Yes. Yeah, and they're one of the market leaders in it, so- Exactly ... if not the market leader in it. Yeah, right here in Pittsburgh for 150 years. 


Ramp AI’s Engineer-Led Origins
---

Jonathan: Before I talk about- Yeah ... ramp- Okay ... just to tie off the connection with Premier Labs- Yes

which is the venture studio when we started, we went into Premier Automation, our corporate partner. And, had built this process for discovering ideas going through different steps of validation, and we immediately went to their field ed engineers and said- "what problems are you seeing that are shared across, different customers? How are they looking at [00:05:00] technology? What are some things that you've tried to do in the past, but didn't work out for one reason or another?" And it's really difficult when you start in a service world to try to build a product company.

So Premier Labs is really set up to be that bridge of innovation to try to create scalable solutions. So what one of the engineers, one of the better engineers within Premier Automation who's been there for a material time and works with global companies and gets called in the middle of the night and flies in, the next day to solve, extremely challenging and important problems- was already using AI. His name is Brent Breigar. Okay. He's, A leading automation engineer and advanced engineer, with Premier Automation. And again, he's the guy who is on the phone in the middle of the night troubleshooting when when a a steel mill furnace is down, right?

Exactly. We need it back up. Or a stamping machine is down. Exactly. And he brought it into our process. And, what we did at that point is we stopped and we're like... and it's every person who has an idea from, those that are, spinning out of universities to people like [00:06:00] Brent who are embedded in an industry and have all of these really great ideas for their solutions they have this vision.

~But we follow a process to be able to test that core idea, and then to mold it in a way to where we're talking about product, the, the classical product market fit from day one, right? And it goes through these several gates, and it won't pass through a gate if we don't have the validation and the market evidence- I love hearing that~

~that we're looking for, right? I mean- Now, it takes longer. There's a trade-off. Yeah. But it's so focused. And when you get through those gates, you know you got something in your hands where it's got a way better chance of catching, right? Nine out of 10, nine, 10 out of startups fail. Of those that fail, about 60, 70% of them are due to lack of product market fit.~

~It's a classical problem. Oh. You've had these conversations- We've seen it a million times over the years ... over and over again. I know. And it's tough, and I've been in these startups as well where you go in with a set of assumptions, but you won't find out until year two some critical things, and you're already a million and a half into building your technology platform.~

Exactly. So I wanted to tell that story, right? No, I'm glad you did. Because it came directly from- I'm glad you did ... the industry. So that's the origin. Yeah. And then we went and talked to customers and started going through some iterative cycles. And then we were very fortunate enough to to be matched up with an AI developer who had significant machine learning experience- and artificial development experience. Okay. And he was actually working on a similar solution, in a, in another use case, but it had a lot of the same architectural components. Gotcha. Yeah, absolutely. And so our team started forming. I love this. You're kinda taking these pieces and you're putting them toge- you're manufacturing them.

Yeah. To build this thing that can now create this, this technology that's become Ramp AI. 


The Cost of Downtime and Lost Know-How
---

Jonathan: So what Ramp AI became was, is it's an agentic layer that works together with plant site collection tools to become a machine interpreter. We're addressing, like a lot of companies, the problem of downtime.

It's about a $1.5 trillion [00:07:00] problem that is characterized within the top 500 companies globally. Okay. Wait, back up. $1.5 trillion. The problem represents- Wow ... about 11% of the- Yeah ... top 500 global companies' revenue. Interesting. Okay? And and there's validated reports out there that, that show that on average the cost of downtime's about $125,000 per hour.

Oh my God. Now, it varies across different industries and different processes. Of course. Yeah, but when you average it out, I mean- It can be as high as $2.3 million- Yeah ... in automotive. Yeah, I can only imagine. When you think about how tight supply chains are, so this... Something goes down, a stamping machine goes down, you can't get those parts out, it just has that ripple effect, and it just can impact what's happening further down the line, and things don't happen.

That just becomes absolutely crazy. So top line, bottom line, companies are losing money every time that, every hour that machine's not running. Machine goes down. And this happens in, in, in almost every, industrial setting where you've got complex manufacturing machines.

Exactly. And so what happens is- Yeah ... is that you have a maintenance [00:08:00] team. And, they're a mix of mechanical and electrical engineers that are doing swapping of parts. Chasing a lot of problems with electrical systems, so on and so forth. But the really, the person that has the understanding of how the machine works, which we, we now commonly refer to as tribal knowledge, is typically a guy, one person within...

If they're lucky, right? A, a lot of, Exactly ... in, in a lot of the discrete manufacturing segments that we're talking to, they don't even have, a sufficient maintenance team and/or that type of automation engineer. That's why they call somebody like Brent in and they- Exactly.

work together with Premier Automation. But there's typically a guy, and it is typically, historically been a guy. But this individual has that instinctive knowledge of what's happened before. They- It's funny you say that. Yeah I've literally seen that in play when I've taken some tours of some manufacturers where...

I was at PEAD Manufacturing like a year or so ago, and to see as they're melting metal into castings and stuff, it's like this person knows exactly what's going on with every furnace. We've been on [00:09:00] tours- And you're just like, "Oh, my goodness." We've been on tours with these individuals, and they can walk by a machine- Yeah

and hear a pitch- And know it's off ... and know something's not right or something's going- Or, or- ... not going to be right coming up. Exactly. Exactly. And that is this knowledge that's not written down, it's not organized, it's not structured, and it's very difficult to access by younger engineers- Yeah

Less senior engineers who are... and- or operators that are on the front line. And one, one of our interviews, someone talked about something, this was real interesting to me 'cause I could relate to it on a personal level. He said, "When something breaks, the operator's gonna sit there and scr- you know, scratch their head, and they're gonna do what he called a rain dance."

It's kinda these- ... let's try all the things- Exactly. ... to get this working again. And then what happens is they'll call in the technician, and it could be a junior technician that maybe with two years, you know, of experience, and that individual's gonna... They... W- what instinctively happens is the person starts all over again.

And 'cause they have their own- Yeah ... set of steps- Yeah, ways to go through it ... to be able to solve the problem. Did they ever give it the old Fonzie bump? [00:10:00] Hey, like this, and it- Not on these machines. Okay. Damn. Yeah. Yeah, I know. I mean- It's a shame ... we might do that to our lawnmower, but not to these complex, casting, machines that, that make the couplers.

So any event, this goes all the way through an escalation pattern. And- Yeah ... and the problems that we found in talking to the the market is that documentation is outdated the day the machine is set up. All right, so you have the manuals and operating procedures, so on and so forth. There's a limited amount of alarms, so sets of alerts that try to tell people what's going on that come from the manufacturers.

They could be misleading, right? That you're looking at symptoms and not causes. Okay. And then there's this tribal knowledge problem. Yeah. So- ... 


How Ramp AI Reads Machines and PLC Code
---

Jonathan: we built a, an agentic layer that reads a lot of static information like manuals, like maintenance history logs. We also- Okay ... build up that tribal knowledge through interviews with people that have these things- Really?

So you're actually interviewing some of the folks to, to- Yeah ... to train the AI on what the people know. Yeah, that's right. Wow. That's right. That's so cool. And that's something we're gonna continue to- Yeah ... to race out and continue to develop. And then the most important distinctive [00:11:00] feature is that we are reading and building an AI level of understanding of the code.

So it's a- Okay ... it's called a PLC, which I'm sure you've heard. Yes. It's- Programmed logic- Yeah, logic controller ... controller, man. Which is the digital bridge between this mechanical world and- Yeah ... this digital world that we live in. It's like the heart of it all in so many ways.

It's the brain. The brain, should I say. Yeah. Yeah. It really is the brain. And And the thing about that code is that, again, it's ch- typically just one person or a limited number of people that, A really understand it in principle. They understand that set of code. Because as we know with code, you can go in there and look at lines of code, and there's comments, there's explanations, there's tags, and that was all done by individuals that could've been 5, 10, 15 years ago.

So it's really understanding, what was the intent of that line of code, to be able... We just know the machine runs, exactly. If you have someone who comes in for the first time, you know the machine runs- Yeah ... and you kinda know how it all works together. And we hope it keeps on running.

And we're talking about a high level of complexity. Oh. Where you've got [00:12:00] sensor sensor platforms, you've got automated settings. You've got a lot of complicated conditions about what's the humidity in the plant that day. It gets that granular. Do, do- Yeah. Okay ... yep. Do we have the right raw materials that are going in, and have that been set up the right way by the operator?

So it's really a, this huge... So what we do is to be able to talk to the machine as you said in the opening- is we we read and ingest all of that, and then we build a baseline of understanding on our AI lay- layer. And so this now becomes what we're all commonly experienced with now working with ChatGPT and Claude.

Yeah. We can actually prompt- ... and ask questions- I love it ... and explore. Yeah. But then one step further, we've built deterministic algorithms, which again, is our secret sauce, of how and where to look when a problem arises. So when we have to, it's called the diagnostic pathway, right?

And again, that comes from- Yeah ... tribal knowledge and experience. Exactly, yeah. And so we read it, we reason and then we have advanced features that [00:13:00] understand your level of experience. Okay, so it'll actually know based upon what your level that it- It talks to you at your level ... at your level.

Oh, that's so amazing. And that, that creates a very unique feature, which is we are valuable to someone who's only got a year of experience, but then we provide a heightened level of value to someone who has 15- Okay ... plus years of experience. And that was one of the things that came out of our experience at Montgomery and Torley- Yeah

is both groups found tremendous value in it. That is amazing. Yeah. See as you describe this in more detail now oh my goodness. I think so too, but I love your enthus- No ... I love your enthusiasm. I just, I don't know, I just see the detail and that, that's going into this and how it can be so helpful.


M&T Pilot: Faster Fault Diagnosis
---

Jonathan: So as you've deployed this at M&T, this is a company that makes train couplers. Yes, that's right. Okay. And so you work with them on installing this on a line. Yeah. What are the results that they've been able to see so far as they're making this thing? Yeah, it, it was really just an initial pilot.

And then we're working to expand the use of it- Yeah ... currently with them. Great. But really, what we can claim is that we're seeing about six and a half time faster time [00:14:00] to resolution on faults. Okay. And again- And yeah ... we think about 70 to 80% of the time of let's say a standard hour of downtime is really just sitting around doing the rain dance and thinking- Yeah

about what's going on. Thinking about which headdress you're gonna put on for the rain dance, right? Yeah. Yeah, exactly. Sometimes we say it's not the wrench time, which is, ordering the part- ... replacing the pa- parts, so on and so forth. And of course, there are delays to, in terms of supply chain availability.

But it ultimately comes down to really understanding it. We we had some sit- situations that we put into the use case so that two separate faults, right? One, one was a valve that was malfunctioning, but they couldn't pinpoint that as the actual cause. They just knew what the symptoms were, that the machine had stopped working.

And then another one was an automated sensor that's supposed to kick on at a certain point of the process and it wasn't kicking on. Wasn't kicking on. Okay. That was in the code. So it wasn't like that there was something broken on the machine It was a code thing. There w- Yeah ... it was originating in the code.

In both of those cases that they had happened before, so we had some history on them. And they would take about six to seven hours to diagnose, [00:15:00] right? Wow. And to get to that problem. What is that- And then we were able- I'm curious- ... to get it in under an hour. In fact- Wow ... in both cases it was 30 minutes each time to understanding.

Okay. Back up real fast. This is so interesting to me. So I'm thinking if they're down for six hours and you have a batch of metal that's ready to be poured to make a coupler and- Yeah ... all of a sudden you can't pour that metal, whatever that metal may be- Now, I don't know specifically- Yeah ... what their outage costs are, right?

Yeah. But if you take the industry average, $125,000- Exactly ... times six hours- Right. Exactly ... you're talking about some significant savings. . Yeah, so that's where I'm just like, "Oh, my..." It's like literally I feel like you're pulling people's bacon out of the fire- just think about-

on certain areas ... how we're using these tools ourselves in different ways- ... right? It it increases our level of quality it helps us do more, and it helps us arrive at certain answers faster. Exactly. Yeah. So it's about efficiency and productivity. And we're really, we are leveraging the general LLM models, for that as the backend engine.

Yeah. But the real value is in this structured AI layer and how we talk to the machines, because these are restrictive environments. I was gonna say, yeah. Exact- [00:16:00] exactly. But also, so I'm curious As, M&T is using this system, as you guys are, were piloting this it's actually learning now from when something went down, so that becomes part of its knowledge.

So if that happened again, the machine your technology would probably know it was fixed by doing this last time. There's actually a step- Yeah ... in the chat session- ... when an operator or a technician says, "Okay, w- that was the problem. We fixed it," we have a mode that says, "Okay, let's go ahead and log that fix."

Exactly. And and it's a lot of technology. We can become a de facto historian- ... layer. Sure. That's the kind of the, the solution that a lot of people do offer on the marketplace. Or we can integrate with a data historian platform. Okay. Gotcha. A lot of smaller SMEs, small and medium sized enterprises don't have an extensive IT, network or solution.

And so really you can use the the, the in-solution aspect of that to be able to to get, d- like you said, not only make the, the product smarter, but then you can go and ask questions about things that had happened in the past- Exactly ... and get to that data. Okay. [00:17:00] A lot of this data today, and it's not gonna be a surprise- Yeah

is being is either m- the majority of it is like an Excel or, digital, Word documents. Yeah, exactly. Exactly. But s- we certainly have seen cases where it's on both pen and paper- Yeah ... and a clipboard. Then you're like scanning that in. Or not, or it's sitting in a binder.

No. Then you need to so you can- Sitting in a binder ... train the machine- Yeah ... on that. 


Helping Veteran and Junior Technicians
---

Jonathan: So the other thing I'm wondering about is, as you were going through this pilot, you mentioned earlier that you had people that were new and older on this. I think someone had 17 years experience running a machine, someone who was maybe there for just a few months, and what I love about you're talking about are the people- Yeah

here, right? 'Cause I feel like this is helping the human. So our AI- Yeah ... AI solution is an assistant. And it is for those people to be able to use. Exactly. In fact, we're not making recommendations about how to fix or exactly to fix 'cause we're not gonna tell somebody to open up a panel- Right.

Yeah ... and touch an electrical lead. Exactly. Because there's significant human safety and health, Of course ... implications there. So we really are really trying to help them understand the problem [00:18:00] and then make some recommendations about what to do. But at their particular level, they need to rely on their training to be able to make those decisions.

The biggest thing that came out of it I'm happy- Okay ... that you... that caught your attention as well. The 17-year veteran- more often than not is being asked to do strategic things for their business. But they can spend 60 to 80% of their time troubleshooting, fighting fires, right?

Makes sense. Yeah. And the f- the comment that we got out of the individual was is that I, I, "I don't have a lot of time to waste sitting around guessing." Phew. And so there's this, level of impatience of "Ugh, this problem again, but it's slightly different this time, and now I'm gonna have to go ahead and spend four hours or five hours really sh- and but yet I have that upgrade that I'm working on."

And so the comment there was is that this 100% takes away the guesswork. It gives them more detail for the information. Now, if you think about the person who has only been there a year and in a lot of the customers that we've talked to, it could take up to two years to bring someone up to a proficiency level.

To get them to that l- level to where they even understand everything that's kinda going on. Yeah, that's right. And just think about any one [00:19:00] of us. Even if we have functional expertise, you go into a new organization, it takes you some time to really understand how things work internally.

It took me five years to figure out where the bathrooms are here at the Tech Council. You know what I'm saying? So yeah. It's been almost 30 years. It's been almost 30 years, yes. Have you got- do you have it down yet? I- it's all down, yes. So two years to bring a junior technician up to a sufficient level of proficiency.

And so the comment that we ... And we interviewed 17 of the users through the pilot, and this one the, the junior technician said, when something s- when it happens on my watch and that senior person is not there, I don't know where to start." Yeah. Where do I start?

How many times are they calling that senior person? Almost all the time. Exactly. Or you're on the third shift, and you have a different, different set of skillsets on the third shift. And being able to try to troubleshoot and kinda figure stuff.

Even if you're trying to set it up to be fixed on the f- the first shift the next morning, right? And so the comment there was is like, "This gives me a place to start. This gives me a level of understanding." The system [00:20:00] was built for downtime events and for diagnostic assistance, but we had this ask and explore functionality.

And the younger technicians just really adopted that immediately and we see them using it because, they're asking questions of, "Why does it do this? How do I- ... do that?" And even some of them took the... And while we're not positioned as a training solution- ... they built their own like we do with general LLM tools, they built their own training program- I was gonna say, absolutely

for themselves. Yeah, exactly. That's what it's all about. So they took the initiative to do that- ... and that was a really great- I was gonna... Was that kind of a- ... experience ... surprise for you guys to see that? Like- Yeah. Yeah. I think- Yeah, I would say it was ... 'cause it, once again, it goes to show you that we've got these great machines, for lack of a better term, these, these tools to help us run these machines, and the software to help us run these machines, but still, the human's gonna figure out another way to use that because- Yeah.

Yeah. And as you know with startups, you, it's you gotta be focused, right? So we are coming to the market focused with a troubleshooting app, but we realize our value goes beyond just waiting for something to fail. Exactly. [00:21:00] And- ... this was just one of these natural ways- ... where that team Was getting some-

immediate value out- Absolutely. 


Beyond Troubleshooting: New Use Cases
---

Jonathan: We're now talking about additional use cases. Tell me about that, because I feel like this is what a cool launching point for this. And I'm thinking "Man," "the future's wide open for you guys." Yeah. Like how... Where do you go next? How do you keep building?

What do you learn? So- Where are you going? ... at M&T, I'll talk about at high level- Sure ... 'cause we're still in the process- Of course ... of, doing discovery. Yeah. But even in the early days, we knew with the technology that had just showed up in our lives, right? It seemed like- Yeah ... with these chat these chat tools.

Engineers are taking pictures of panels, they're taking pictures of parts, and they're throwing it in the LLM models and saying, it might be a legacy system that's, 30 years old, and they're saying- Who made this part? Exactly. Where can I find the manual of it? Or I've got a problem and I need to f- I need to replace a subcomponent and there's no known supplier for it.

Or a beehive of of of wires and alerts and everything else, and literally taking pictures and then saying to the [00:22:00] system, "Okay, this is the operation I was doing, and now here's the- here's how the panel looks. Can we try to reverse engineer them?" Yeah. Reverse engineer this.

Yeah. Yeah. So it's things like that are starting to emerge. What we also found in talking to other customers, it's not just about fixing things that are broken. It is about the reverse engineering. There's a lot of legacy equipment- ... especially in North American manufacturing markets. And so really trying to figure out how it works what it does.

Code might have been updated 10, 15 years ago- ... by an engineer that's no longer there. Like 15-year-old code. It's oh my God. No notes, no commenting. No. A, a lot of systems are from European OE- OEMs, and so could literally be written in Italian- Yeah. ... or French. Oh my goodness. This is common.

I know, but it's so crazy- German ... to think about that ... with Siemens being one of the leading- Yes ... pLC providers you might have it written in another language. And of course, there's translators that are out there, but just really, again, trying to come to an understanding of- Right

of what's happening. And then finally, even new equipment installation again, talking to that machine while you're doing onboarding and [00:23:00] provisioning, right? As you're starting to think about your recipes and your standard procedures, what if I do this? What can I expect? What if I s- what if I take this one sensor setting and put it up to the highest level?

What is that going to mean to- ... another subcomponent? And these weren't things that we went to market purposely thinking about, but we knew- But that's what's coming up as you, as you- We know ... install it. Yeah. Yeah, we know. Yeah, there's things going on there. So we actually, and I don't know if you picked up on it, but, we believe we're defining a new segment in the marketplace, which is called machine intelligence.

Exactly. Yeah, that makes- the ability to really have those machines- Okay ... talk to you and tell you- How- ... what's happening and to be able to work together with them. How does that coincide with... I had this really interesting interview with the Arm Institute around physical AI- ... and how we're seeing so many more sensors now on the manufacturing line, and they're being, powered by other companies like Powercast and other members of the tech council, and they're able to power these things remotely.

But it seems like the more sensors that are out there, the more data that you could be grabbing and pulling into your system. Is that something [00:24:00] I'm thinking about in the right way or...? It's come up in conversations- Okay ... with customers and investors- Yeah ... recently. Right now, when I tell you that we read and build a baseline, we're really centered around that brain, that PLC.

That's, you're still on the PLC. Gotcha. But you have sub-components like industrial - robots- ... drive systems, electrical drive systems- ... sensor, architecture. Yeah. The next step for us is to really understand and bring those into- So cool ... the platform.

So that's why- Yeah ... that's why I get so excited- Yeah ... because when I look at how the technology keeps evolving and how you, it just, you can keep adding to what you're doing. Yeah. And I- It's like phew, man ... and I wanna, and it you made me think of something, and I do wanna make sure I say this, is that- Yeah


Purpose-Built AI vs. General LLMs
---

Jonathan: There's been a lot of discussion in the popular press that the LLMs are just gonna eat up everything. And they're gonna, they're gonna become our financial systems and our healthcare systems. And, and we personally don't Buy into, that future Yeah, I don't either.

Yeah and I think very quickly- Yeah ... and this might have been a, a somewhat of a novel thing to say, one or two years ago, but today it's now c- more commonly understood. [00:25:00] We have a purpose-built AI solution, right? And this is how we think a lot of solutions are going to evolve, right?

Because- Okay. Yeah ... you st- like context in an AI setting, context matters, right? That, that's everything. Yeah And a lot of what we do is not just ingesting this huge amount of data- ... and then spitting out an answer. We're narrowing down to where the answer actually lies much faster, right?

We have run benchmark bench-top tests against some of the general LLMs, and they do a pretty good job. You can go and ask about a particular OEM's machine and you can get a lot of useful information and troubleshoot it. But we capture that tribal knowledge because every plant, every facility- It's different

every customer is different. Yes. Yes I would think almost every shift could be different to a certain degree. Every run. Because every run could be different. Whenever you have a production run. Businesses are constantly changing. Yeah. And so when you talk about making manufacturing, better, faster, smarter, what we have seen over the last 10 to 20 years is the ability for them to do personalization in the marketplace.

So you now have a, a wider range of products that [00:26:00] go out. Yeah. So you're not just selling one widget. You're, you might be selling 100 of them. Exactly, yeah. Or you might be doing special orders, for your customers as, as their needs change. And so they have already been in the mindset of building flexible industrial processes to be able to do that.

We only s- play into that, more. Yeah. Which, because now you can think about, okay, if that, if I'm gonna go ahead and, make a design change to a coupler, because of a new application either in an existing market or a new market that we wanna, that M&T wants to pursue you could start running through some of those scenarios, using it.

So now we start moving into becoming an analytical, platform as well. Exactly. So the key part, the first part is understanding, and that's really what we're focused on right now. And that's the journey now. Yeah. Yeah. That makes total sense. 


Why Build Industrial AI in Pittsburgh?
---

Jonathan: So I feel like Pittsburgh is the perfect place to be building this.

I feel I don't know, like- This wouldn't work in California, man. I just don't think it would there's not a big base of opera- manufacturing in California Exactly right and they're really focused on doing... they're changing the world and driving and affecting society and, In their ways

in software- Absolutely ... and other technologies. Yeah And this and actually, and a lot of what you're doing takes me [00:27:00] back to when I interviewed Andrew Moore from Lovelace this past summer, and he gave me this- By the way, I'm really glad that you think of Andrew Moore when we're having the conversation.

Yeah. No I think that... Yeah. Sorry, just wanna... and you'll see, and you'll be double stoked when I drop this bomb on you here, Joel. So when I was interviewing him, he's talking about, like, why he's building Lovelace here in Pittsburgh, and he's "You can build wonderful things on the on the West Coast.

Cannot, will not disparage anything about the West Coast," he says, "But when I wanna build something that works-" "... I build it in Pittsburgh." That's what you guys are doing. This is building stuff that works. This gets back to Premier Automation. Yeah. And again society continues to evolve, and in my mind, always get better with these new technology.

It... Net, net, right? Exactly. Of course there's a lot of problems. Always, yeah. That's the way it's gonna be ... but he's absolutely right and it's a big part of the value of the Premier Labs and our relationship with Premier Automation. The systems that they work on, and, I can't name customers of theirs, but, like- Pretty important companies out there- I mean-

are relying on Premier ... they're steel mills. They're- Exactly ... they're manufacturing companies. Yeah. They're waste water [00:28:00] plants. They are the ride that you're riding at the amusement park that it- ... it, the, the tower goes up- Yeah. ... and it gets stuck, and you c- that's all driven by control systems, electrical control systems. Yep. They're also working in with some of the leading companies that are taking us to space f- w- that are working on space exploration. Exactly. And so a lot of those important subsystems need to work, and they need to be, at a, a level of, high reliability.

And I can tell you, and again, I've had many great experiences as w- you alluded to of companies that are still out there changing their market spaces. But, just the first couple meetings with the engineers, with Premier Automation, sometimes they would say something in the first ideation meeting or proof of concept meeting that it would've taken me two years in startups that I've been part of- No way.

Okay ... to gain a realiz- yeah ... to at least be aware of. And by the way they're, like, practical. Why are we architecting, different system, different project, but why are we doing it with one PLC? Why don't we just throw another PLC in there because it's gonna go ahead and control this part of it and [00:29:00] isolate it?

Yeah. I have enough examples, and, and- That's what I'm talking about. That's bigger ... that's in other companies- Yeah ... where you find out two years later that your assumption was flawed and it's hard to go back and fix it. And so it's been a real joy working, with these engineers.

I'm not an engineer so it's just... It's sometimes I just sit and observe and really marvel at it. Got it. Yeah, it's... 'Cause like I said, when Andrew said, he says there's an engineering culture in Pittsburgh that he says is just so unique. Yeah. And that's why he wants to be here. Our- And-

our greatest strengths are our manufacturing base- Yep ... and understanding the history, of it. And that's still here. We're not the largest like we used to be, at one point of history, but we still are doing s- We're still making stuff. We're still making a lot of the really important stuff, and you're- Exactly

telling these stories. Yeah. But then, as you, you mentioned, the Venn diagram, it overlaps with this new industry of first robotics- ... and now being recognized as one of the birthplaces of AI, and now this emergence of this, new segment called physical AI. Yeah. When I [00:30:00] was with the PRN, you know how I like to make bold statements.

And I believe in that, but it's okay. We're one of the top three places in the world- 100% ... for doing that. Absolutely. And there are other great regions that are doing amazing things. Of course. And we really should be networked in, to them. Absolutely. But, you know- We're- ... even as this grows, top 10, top five globally, it's happening right here


Closing Thoughts
---

Jonathan: We're holding our weight. And I like the way you brought back my circular thing with our Venn diagram here. A great way to wrap this up, 'cause I like it when those two circles overlap and there's something really kick butt in the middle, and that's the type of work that, Ramp AI is doing.

Yeah. And Joel, what... How much fun talking to you today about this. No, it's always... you- yeah ... hey, I'm jazzed, but when I talk to you, I get even more jazzed. That's what I want- ... because that's what happens. Dare I say I get amped? Dare you say you get amped on what we're talking about today-

because I like to turn up the volume on all things manufacturing. There, I one-upped you on my puns, which is a great way to end this podcast, Joel. Always a pleasure having you in the Huntington Bank Studio so we can have these conversations. You want more content like this, if you wanna nerd out on all the great [00:31:00] manufacturing and technology happening here in Pittsburgh, I'm your dude, simple as that.

It's been Jonathan Kersting with the Pittsburgh Technology Council and TechVibe bringing you amped right here.