“Today, video AI agents exist that turn your existing cameras into AI teammates across a range of applications. If you can leverage this technology to make your assembly line operate 1% more efficiently, that can result in hundreds of millions of dollars of revenue in a year.”

In this episode of the Manufacturing Matters podcast, Dunchadhn Lyons, Director of Engineering and Allison Lilly, Director Product Marketing at Spot AI joined TECH B2B Marketing’s Winn Hardin and Jimmy Carroll to discuss talk about the concept of video AI agents, what this means and how it works, and the different industries and applications in which the technology can add value. While businesses like car washes and offices can benefit, global manufacturers that are looking to improve safety outcomes on manufacturing floors and warehouses and to improve operational efficiency can as well, according to Spot AI. Additionally, the episode covers real-life examples of productivity gains, reception of the Spot AI product from plant floor workers, the overall topic of AI in the industrial space today, and more.

Winn Hardin: [00:00:06] Hello everybody and welcome to Manufacturing Matters, where we talk about the technology and trends reshaping the global manufacturing industry. My name is Winn Hardin. I’m one of your co-hosts today. And I’m with my good friend Jimmy Carroll.

 

Jimmy Carroll: [00:00:17] Hey everybody.

 

Winn Hardin: [00:00:17] And today we’re lucky enough to be with Allison Lilly, director of product marketing at Spot AI, and Dunchadhn Lyons, director of engineering at Spot AI. I thank you so much for joining us today.

 

Dunchadhn Lyons: [00:00:27] Thank you for having us.

 

Winn Hardin: [00:00:28] Is this your first time at Automate?

 

Allison Lilly: [00:00:29] It is.

 

Winn Hardin: [00:00:30] Awesome. How’s the initial — how’s it been so far?

 

Dunchadhn Lyons: [00:00:35] It’s been incredible. Really vibrant community. So many interesting sessions, talking a lot about automation, physical AI, and robotics. And so we’ve just been learning so much, bouncing ideas off of people. It’s really great collaborative energy.

 

Winn Hardin: [00:00:49] Yeah. I’m sure you enjoyed today’s keynote from the video about physical AI.

 

Allison Lilly: [00:00:52] I tweeted a quote, yeah.

 

Winn Hardin: [00:00:54] Awesome.

 

Allison Lilly: [00:00:56] I mean, they’re saying we’re in the era of ChatGPT for robotics. We believe we’re in the era of ChatGPT for video. You know, it’s just bringing that accessibility to all the latest automation technologies.

 

Jimmy Carroll: [00:01:10] So tell us a little bit about that. For those who don’t know, and tell us about Spot AI, what you do and I guess why you’re here for the first time and why you think it makes sense.

 

Dunchadhn Lyons: [00:01:18] Yeah, absolutely. So at Spot AI, we’re building video AI agents for your dumb IP, your security cameras. Basically, we turn those into AI teammates to support your safety, your operations, perhaps security kind of use cases. And what we found is there is a ton of demand in the manufacturing space. Everyone here, 30,000 people, are looking to make their assembly lines more efficient, their people more efficient and safer. And so we’re here to learn and share what we know and find better ways to build these AI teammates for the manufacturing world.

 

Jimmy Carroll: [00:01:51] It’s funny that we kind of had a serendipitous connection on LinkedIn. And then I’m looking at your company and thinking, this is really interesting. And we reached out and we had this conversation set up. But you know, I was looking through your posts and your LinkedIn page and some of the different things that you’ve been talking about lately. And one of the things was something about telling — what’s the name of the AI agent?

 

Dunchadhn Lyons: [00:02:12] Iris.

 

Jimmy Carroll: [00:02:16] Iris. You asked Iris to tell how many people were wearing sunglasses throughout a week or something like that, right? So if I had one at home, how does it work? Would I be able to have have Iris look at how long my kids are brushing their teeth or how many cookies they’re stealing on the counter? And if so, how do I sign up?

 

Dunchadhn Lyons: [00:02:33] Yeah, absolutely. That’s great. So Iris is our custom video AI agent builder for the video world, for cameras. And like you said, Iris can help you build a video agent that understands and acts on basically anything. So if you had cameras set up in your house and you wanted to know when your kid was skimping on their two-minute toothbrush routine, Iris can help you build an agent to understand and maybe play a message over a speaker saying, “Hey, you didn’t finish brushing your teeth yet.” Or maybe a cookie jar monitor that understands where the cookie jar is on your counter. And if your child is sneaking in late at night to try and steal a cookie out of the cookie jar, send you a text notification or turn on the lights, whatever it may be. We’re not explicitly addressing the consumer market today, but I think those really well embodied the kinds of things that our AI teammates are able to do.

 

Allison Lilly: [00:03:27] What that would look like for you is you could sit at your kitchen table, with five minutes, in your laptop, and you would tell it, “Hey, I want to build this agent. Here’s 10 images of what my kid looks like brushing his teeth. Here’s 10 of when they’re not brushing their teeth.” And it’s trained in minutes. That is building an AI model in minutes. And anyone can do it through just talking to it like we do to ChatGPT, bringing it back. Prior, you needed a machine learning degree. You needed eight weeks. You needed code. You needed training, testing, validation, tons of times. Tons of data to tell it what to look for and when and why. Now it’s five minutes at your kitchen table.

 

Jimmy Carroll: [00:04:09] It might work against me though. I could see my wife being like, “Did you have an extra bourbon? Iris told me that you did.”

 

Allison Lilly: [00:04:17] It’s funny — we test and play with it all the time in the office. We just did the sunglass one. And yesterday I just saw that someone built a “carrying a drink” agent. That’s the latest: grabbing a beer from our fridge in the office. But apply it to anything. You know, we had a client. Someone was stealing a bunch of copper in their outside facilities through bolt cutters. So we built a bolt cutter agent in minutes. There’s some when the conveyor belts are starting to back up, it looks like a very specific thing. One was a bread manufacturer, so bread kept backing up, so we trained it on, “Here’s when there’s too much bread on your conveyor belt.” And you do that in minutes.

 

Winn Hardin: [00:04:56] So what industries are you guys playing in? We’re talking about a lot of fun applications, but where are you finding success so far? How long has Spot AI been around?

 

Dunchadhn Lyons: [00:05:02] Yeah, Spot AI has been around for about five years. And we started essentially as the YouTube for IP cameras, for security cameras. When we entered the space, there were close to 100 million deployed cameras across U.S. businesses, but no good way to even view that footage. People in your organization didn’t have access to it. So we started by building a platform that integrates all these cameras together, gives you a dashboard to be able to just view footage. And we started layering on additional AI technology to power search for basic alerts. And now where we are today is these video AI agents that turn your cameras into AI teammates. And we work across 17 different customer segments, for example, in the car wash space. We’ve built AI teammates for improving operations around kiosks. We’ve all had an experience where we’re sitting in a drive-through for 10 minutes, not understanding why the traffic in front of us isn’t flowing. So our AI agents are watching a live stream. They can understand when there’s somebody at the kiosk but they’re having trouble and there’s no human attendant there. They can send a notification to the manager on the ground immediately to make sure somebody gets out there, helps them, keeps traffic flowing, and keeps business moving.

 

Dunchadhn Lyons: [00:06:11] And one of our car wash customers was able to parlay that into a 54% increase in subscription sales over the course of just two months, simply by ensuring that there was always a good customer service experience. In the retail space, we’ve been working with a lot of retailers for understanding how people flow through their businesses, through their locations, so they can target advertising, understanding volume, throughout the day, of patrons, so that they can plan for staffing accordingly. And then also, similar to the unattended kiosk, understanding when there are people waiting in line to check out but there’s no human attendant. You give your patrons a better customer experience. They come back more, they spend more money. So there’s real ROI. But what we’ve been seeing over the last 12 to 18 months is especially in the manufacturing space, tons and tons of inbound, primarily looking to improve safety outcomes on manufacturing floors and warehouses and to improve operational efficiency. If you can make your assembly line operate 1% more efficient, that can result in hundreds of millions of dollars of revenue in a year. So very interesting stuff.

 

Winn Hardin: [00:07:19] I could see this being your key link in the chain, especially as we move towards autonomous mobile robotics. In terms of measuring, optimizing missions and which warehouse, throughput efficiency gains. And I was really curious, you answered the question, which was what was the key value proposition that you’re bringing here to industry? Safety was top of mind. And then obviously there’s overall efficiencies of operations.

 

Allison Lilly: [00:07:42] It all starts with safety because there’s a lot of common safety challenges in the industrial space. Are people wearing their proper gear? Are people falling? Is there a spill that someone might fall on? Of course forklift behavior. That’s the number one thing. If you search our sales calls: forklifts. The number one thing that comes up is where are the forklifts going? How fast are they going? Are there any dangerous behaviors? So we built a “forklift near miss agent.” It’s our most common agent, just to understand: Are forklifts getting too close to other people, forklifts, racks, walls, etc.? And that’s kind of a low-hanging fruit that people start with. They get a couple of these safety agents build out. Then it’s like, okay, expanding — yes, we’re mitigating risks — but next is how can we bring efficiency to the broader operations and get that bird’s-eye view of the flow and the movement of people and traffic and goods and the order in which things are done and where are the spots of downtime and bottlenecks, loading and unloading sometimes. Sometimes it’s material handling.  Sometimes it’s along the assembly line. Sometimes it’s the packaging piece of things, the distribution piece of things. So then we’ll start tackling the bigger efficiency and also, of course, higher-yield problems ultimately.

 

Winn Hardin: [00:08:55] Yeah. Do you have custom dashboards that help people? Because if you’ve got 16 locations within a warehouse or a production factory measuring all those individuals and interactions, I can see that being quite complex. So how do you handle that sort of simplifying the data to bring intelligence to operation?

 

Dunchadhn Lyons: [00:09:11] Absolutely. So we do have a centralized dashboard for each of our customers. It’s available web and also on mobile. And with these AI agents, not only are they watching and understanding and reasoning about all of your video 24/7, they’re selecting only the portions of the video that are important to share with you, and then they can automatically generate a report for you. So you want to see just the events that occurred in these three specific locations over the last two weeks. Two seconds later, you’ve got the report, you have the relevant clips, and you can make decisions and garner insights from that.

 

Allison Lilly: [00:09:44] And we should tell a story about that — one of our early manufacturing customer’s story. If you think of some of the data around one of these use cases. So one of our manufacturing customers, they have five facilities. They do a bunch of really heavy stone, granite cabinet-making machinery. They hired a person to sit and watch all their video feeds for safety incidents.

 

Dunchadhn Lyons: [00:10:10] Six hundred cameras.

 

Allison Lilly: [00:10:11] Six hundred cameras, which translates to . . .

 

Dunchadhn Lyons: [00:10:13] Fifteen thousand hours of video a day, something like that.

 

Allison Lilly: [00:10:17] Which is obviously impossible for any one person. So this person, this is her whole job. She was catching about two of these near-miss incidents per day. Two, which is a lot. Any one of those could be catastrophic, right? They put AI agents in place, and within minutes, because you could look historically, they’re saying that they were getting more like 10 a day in certain spots of their facility. So a 5x increase in what’s happening. But it’s done in minutes, not hours. And it just takes the responsibility off of her lap. It’s that force multiplier to get her back to higher value: training on safety protocols and what does she really want to be doing and not watching that.

 

Winn Hardin: [00:10:56] One of the most mind-numbing jobs in the market.

 

Dunchadhn Lyons: [00:10:59] I can only imagine, as much footage as I’ve had to watch over the last three years as we’ve been building these AI agents.

 

Winn Hardin: [00:11:05] Especially when you have 600 cameras, it’s rotating through. So, I mean, you’re not even getting 100% coverage at 600. Not to mention you couldn’t divide your attention amongst that.

 

Dunchadhn Lyons: [00:11:11] That’s exactly right. Yeah, of these 100 million cameras that are deployed across U.S. businesses, only about 1% of that video is actually ever watched by a human.

 

Winn Hardin: [00:11:23] And we only know the problem when there’s a catastrophic failure.

 

Allison Lilly: [00:11:26] Exactly.

 

Dunchadhn Lyons: [00:11:27] Exactly, and as you mentioned, in a lot of those cases, you’re flipping from camera to camera. So 95% of even that 1% is missed. 

 

Winn Hardin: [00:11:38] Cool. I love this product. I can’t wait to hear more about it. One last question: Do you see you guys moving into defect detection or a product like standard machine vision? Obviously, you’ve got an AI agent. Is that something that you see in your future roadmap?

 

Dunchadhn Lyons: [00:11:54] Yeah, absolutely. So right now, we think of our video AI agents as kind of junior team members. You can bring them in like you would a new employee. You can train them up, you can coach them. And they’re very good at executing on specific tasks. And if they understand what they’re supposed to do and what outcomes they’re supposed to be providing, what actions they’re supposed to be taking, they’re very good at that. But in the next few years, with the advent of foundational models and generative AI, the reasoning capabilities have gotten a lot stronger, and we’re starting to integrate that. And so solving more complex use cases, like anomaly detection, that’s more open-ended — very soon these junior AI teammates, they’re going to be promoted. They’re going to get feedback. They’re going to get smarter. They’re going to be able to reason better. And rather than being assigned to task with specific actions, you’ll be able to assign them objectives to complete however they see fit. And so when you’re talking about something like anomaly detection along a conveyor belt in your manufacturing line, that’s absolutely something that these agents will be capable of in the very near future.

 

Jimmy Carroll: [00:12:59] So it’s very interesting. This conversation has shifted. And like you guys said, it’s your first Automate. You know, seven or eight years ago, the conversation was all around robots. Are robots taking jobs? And A3 put out this report in conjunction with somebody and saying basically: No, it’s the opposite. It’s creating more jobs. And it seems sort of like people are getting over the hump of robots taking jobs. But now AI comes in, and it’s a very similar conversation. Like everyone says: dull, dirty, and dangerous. Those are the jobs that robots are taking. But AI is doing that now too. And I’m seeing that it’s not a bad thing. Nobody wants to go through all that video — it’s a mindless thing — or just pick oranges and put them in. These technologies are all advancing. Anyway, I’m going too far down this path and waxing poetic AI. We spoke last week on the phone, and something I thought was interesting I wanted to ask on the podcast. You said that companies are finding success that have mandates to use AI. I just thought that was a very interesting concept. First of all, what are some examples of where they’re finding success? And then, what are your thoughts on companies mandating the use of AI? I mean, it’s great for you, but like overall?

 

Dunchadhn Lyons: [00:14:13] Do you wanna take that one? Allison.

 

Allison Lilly: [00:14:15] You can kick it off.

 

Dunchadhn Lyons: [00:14:16] Sure. Yeah. So we are seeing these kinds of AI mandates or AI initiatives across U.S. businesses. And the way I think about these is we had a massive revolution in the ’70s and ’80s with the advent of the personal computer. And they completely changed how business was done. And although people were afraid of computers at first, what they really did was augment and empower people who use them. Then in the 90s into the 2000s, we had another way. It was the internet. It was the same thing. Massive change and revolution in how we do business. And AI is that next wave. It is going to transform every person at work over the next 10 to 20 years. And executives realize this. And although just a few years ago, a lot of this generative AI, LLMs, ChatGPT were still very much experimental. We’re now seeing enterprise-level deployments of this kind of AI technology, solving real problems primarily in the digital space. But what we’re trying to do is pull that into the physical world. So these businesses are trying to get ahead of the curve, retrain their people to leverage this technology, because it can be a massive force multiplier for your people. This technology empowers people, augments them, and allows them to spend their time thinking and strategizing in a much higher level than some of these more mundane tasks that AI can very easily automate for them.

 

Jimmy Carroll: [00:15:39] Interesting. So what are some examples of maybe some quick wins for these companies that have adopted your product and in a way that maybe you were surprised by or they were surprised by?

 

Winn Hardin: [00:15:49] Especially in the industrial environment. I mean, you guys touched on safety, in identifying your business, but we’d love to hear more.

 

Dunchadhn Lyons: [00:15:55] Yeah, absolutely. So another manufacturing customer focusing on safety outcomes used AI agents to understand missing personal protective equipment. Primarily safety vests and gloves in the warehouse space, which are crucial. And over the course of about two months, by the agent being able to identify, alert, and build these reports, they were able to better train their people and reduce those incidents by nearly 40% just in a two-month period. And Allison also mentioned we’re working with another manufacturing customer to understand when they have bottlenecks on their processing lines, and they kind of classify this into green: Everything’s flowing as it should. Yellow: Things are starting to get a bit a little bit bottlenecked, so if we can get a human to intervene at this point, we can save a lot of downtime. Or red: where everything’s completely shut down. And what we’ve been able to do is build agents that understand that yellow zone very, very well. And so again, it’s AI empowering a human. The agent is watching 24/7, understands when we’re bordering on yellow. There could be a potential problem soon. We’re going to contact the operational manager on the ground right now. Get them to intervene. And so there’s no downtime, no lost revenue. So like I said, AI teammates. These really are your teammates. They’re empowering you. They’re making your job easier and allowing you to focus your attention on higher-value things.

 

Winn Hardin: [00:17:18] You know what’s interesting is these returns keep coming in. I mean, you can almost envision an environment where you might get a break on worker’s compensation insurance premiums, because now we’re verifying that people are following the best safety protective measures and saving a whole lot of employees from getting hurt and taking time off. It’s funny how tangential things will drive. You know, if the insurance company feels — which is a massive industry — if they find ways that are going to help save them money and allocation. So my mind is just a little bit blown in terms of all the different directions, and we’ve been living in AI for I don’t even know how many years. We’ve worked with a lot of companies in that space.

 

Allison Lilly: [00:18:01] Now AI has eyes. Think of it like that: your AI has eyes. And if it’s seeing and sensing and understanding your world, really it’s up to you how you make the most of what you need and how it can act there for you.

 

Winn Hardin: [00:18:13] Almost the only limitation is the imagination.

 

Allison Lilly: [00:18:15] Our customers say that to us.

 

Jimmy Carroll: [00:18:18] You’re going to scare some people, right, by saying AI has eyes. So talk to me a little bit about that. Like when you tell people that maybe are outside of the industry that you work at an AI company, I’m certain that some people are like, oh, what does that mean? I was talking about it yesterday on a machine vision panel and Automate Live, and one of the guys was saying, yeah, people are afraid of AI either taking their jobs or enslaving us. So how do you dispel the notion that AI is a bad thing to people who might think that way?

 

Winn Hardin: [00:18:49] I’m sure education is a key component.

 

Allison Lilly: [00:18:50] We see that. Absolutely. Some of our best customers who roll it out, they are so transparent about what the cameras do and what they don’t. And in most states, facial rec is not even there at all. Most organizations don’t use it to target a specific person or face or to discipline a singular person. It’s about the trends and  using the trends to figure out your next steps accordingly and as efficiently as possible. Sometimes that might be a new training program. Sometimes it’s a good behavior catch. So we have “good catch” agents basically, where they want to know when people are wearing their PPE, and they can congratulate that shift or reward that shift or that pocket or that zone where people are doing a really good job. The best is, like you said, transparency: This is what they do. This is what they don’t. Here’s how it’s ultimately going to help you. It’s going to be a training and a coaching practice as well. And also, with our UI, there’s some quotes in the industry that some of the best AI has the best UI, meaning the way you interact and experience the AI is really in your hands. We make sure all the toggles of what people are detecting and not detecting are completely in our champions’ hands, and it’s really up to them to customize what their org needs based on just what they need on the floor but also their culture and philosophy around bringing AI technology to their factory floors.

 

Dunchadhn Lyons: [00:20:21] And you mentioned the notion of education. And we very much see ourselves as educators in this space. We do have plenty of customers come to us who are very technologically savvy and they understand AI, but we have an equivalent number who come to us who understand the technology very little. And so it’s our job to walk them through it, make sure they understand exactly how it works and, again, how it can empower their people, how it can be a true teammate. And you ask about: Will robots or will AI enslave us? The way I like to talk about that, particularly in philosophical questions with friends, is: As mature and powerful as ChatGPT and similar foundational models are today, at the end of the day, they’re just a massive probabilistic equation that’s been trained on tons and tons of human data, but there is no sentience. And at this point in time, we are very far from anything that looks like human sentience.

 

Winn Hardin: [00:21:20] Yeah. It’s a general AI conversation a lot with folks outside of the industry. I’m gonna follow up on the other question about acceptance because when machine vision, robotics, in the ’70s, and when machine vision became more prominent in the ’80s and ’90s, there was pushback from the guys working on and the ladies working on the floor. They were afraid, again, that it would take their job. And then ultimately, they came to realize that no, actually, this makes my job easier, more palatable.  It’s helping me actually. And then this makes my company more profitable, which then allows them to increase capacity, which gives me more employment, you know, insurance: I’m not going to lose my job actually. It’s actually working for me to make sure I keep my job. In your early deployments, especially in industry, what’s the reception you guys are getting from the plant floor workers? Any kind of anecdotal information about how that’s going? I think you would expect some speed bumps up front, I mean with every major change, and this is certainly a major change.

 

Allison Lilly: [00:22:19] Huge changes. You know, one of our customers, because it’s detecting what they were already trying to detect with their eyes and sometimes clipboards or sometimes watching the camera, one of our customers said to us, “Wow, it’s what I’m supposed to do, but it’s served on a dinner plate to me.” You know, it cuts through the noise and it watches what you need to watch for you. So I think many are seeing it as an augmentation, but it’s not. You have to see and experience AI. It’s kind of like the ChatGPT moment. And so we’re all about making sure when we’re rolling out an org, we have everyone that’s going to be involved in it on the call, getting trained and communicated with and at the same time getting to ask those questions, so they understand what it is and what it isn’t, and what will exactly be the tangible physical experience of interacting with these agents for your job?

 

Winn Hardin: [00:23:12] I could see the safety officers in the management level just loving this 100%, giving them the tools they need to actually make their company more productive and keep their workers safer. But I wonder have you had any kind of interactions with labor unions? I’m just curious. I know we’re kind of getting off target a little bit, but back in the day I believe they pushed back on where they could put cameras within the plant. You know, whether it would be in the break room or it would be other places, so they can have some privacy.

 

Allison Lilly: [00:23:39] Well, one very recent example, huge manufacturer. They wanted to test one very specific thing and then collaborate on that very specific thing with their union before going to the bigger, broader, many-agent use cases. Very specific agent, bring it with them, bring the union folks along, and then potentially roll it out and show the benefits to workers.

 

Winn Hardin: [00:24:04] That’s a brilliant approach, you know, because you need to do collaboration early on, and that allows the education component to come into play and then takes down the fear level.

 

Allison Lilly: [00:24:12] Exactly. And see and touch and feel. AI is just such a nebulous term. It means something different to everyone. Some people, they don’t even use ChatGTP. Some people have automated travel booking with it. And so until you see and touch and feel it, there may be a small specific win or low-hanging fruit. You may not know, and it may remain in that box. We hear a lot about the black box aspect of AI. And you know, that’s why we do focus so much on design and bringing that ChatGPT-like experience to our technology, because we want people to feel that it’s accessible and within their reach.

 

Winn Hardin: [00:24:51] Yeah, yeah.

 

Jimmy Carroll: [00:24:53] Beyond the obvious of networking and potentially meeting new customers, how do you see yourself fitting in here? Like are you looking for potential partnerships with camera companies? Can it run on any camera? Does it require additional processing? So would you need to partner with GPU companies or industrial PC companies? What are your primary goals when you come to Automate?

 

Dunchadhn Lyons: [00:25:19] Absolutely. Well, like we said, this is our first time in Automate. And it’s mostly for us to learn and to meet people as you said. You asked if our system can run on any camera. Any camera that you can pull a stream from, yes, it can operate on. So while primarily today we work on top of IP cameras or security cameras, what I’ve realized from talking to camera makers here at the show is there’s potential for us in the future to be able to integrate with cameras that are attached directly to machinery or built into machinery. And if we can combine them with other cameras, we can provide much better context to the automation, more of a bird’s-eye view, understanding of all the components of your warehouse. So I think that’s a very interesting potential there. So certainly we’re open to partnership, but really we’re trying to learn as much as we can and meet as many people as we can.

 

Winn Hardin: [00:26:14] Talking with people, digital twinning is becoming a much bigger deal. And whole plant and/or production line simulation. You’re seeing a lot more third-party entrants come in there to help people visualize before they actually go in to build. Is there a kind of a synergy between those service providers and yourself?

 

Dunchadhn Lyons: [00:26:30] I think there could be in the future. You know, we’re mostly on the perception side, understanding what’s happening in real time and then responding to it, whereas digital world twins are all about simulation, so that you can optimize your automation in the simulation of the digital world before you bring it to to the physical world. And I think there’s definitely opportunity for partnership there in the future. If we are already integrated with all of your cameras, we understand the layout of your manufacturing space, we can identify all of the different objects and machines that could then be pulled into a simulation to make it that much better and representative of the physical world. So I think there’s certainly opportunities for partnership there. That’s not something we’ve explored today.

 

Winn Hardin: [00:27:16] Theoretically, it might give you the ability to bring in more and more data from actual production without having to do all the expensive recabling and IoT. I mean, maybe one camera could do the work of 20 individual sensors, point sensors, or something.

 

Dunchadhn Lyons: [00:27:30] Yeah. And that’s the incredible thing, right? We’re mostly leveraging cameras that are already there. We send you a box of hardware, PC size. You plug it into power, and your local network, and any cameras that are connected to that local network you can pull in and turn into AI agents and AI teammates just that quickly. A lot of our customers aren’t actually putting in any new infrastructure in order to support our system.

 

Winn Hardin: [00:27:52] That is very important. Okay, there’s the soundbite for this episode by the way. Massive new feature and functionality without giant CapEx expenditure. And do you have the ability to do cloud, on premise, hybrid? You give everyone the option in terms of where their data is going to be stored?

 

Dunchadhn Lyons: [00:28:07] Absolutely. Yeah, so we primarily operate on the edge. Most of our AI processing is happening on the edge. And we actually do partner with Nvidia for GPU acceleration of our AI models. And for security and privacy reasons, most of our customers choose to have their video stored on prem as well

 

Winn Hardin: [00:28:23] Especially manufacturing.

 

Dunchadhn Lyons: [00:28:25] Exactly. That’s right. But we do have the ability for cloud storage. And with some of our more advanced capabilities that we’re working on now, we do leverage cloud computing as well to pull in things like foundational models. So we do have a very flexible hybrid architecture between the edge and the cloud, depending on the use cases that our customers are trying to solve.

 

Jimmy Carroll: [00:28:47] What else haven’t we talked about? I’ve got a lot of questions, but it would be selfish and kind of silly to ask all of them.

 

Allison Lilly: [00:28:53] Still thinking about your kids not brushing their teeth?

 

Winn Hardin: [00:28:57] Well, that’s where we started, so this is a great place to close the loop.

 

Jimmy Carroll: [00:29:01] One of the things I’m thinking of, and you were talking about it Winn, is I could see this technology being really useful in in-vehicle monitoring for drivers. Like for the insurance company, for example. I was just thinking about stuff like that or truck drivers or Uber drivers.

 

Winn Hardin: [00:29:16] It’d make me feel better about my 16-year-old daughter who just got a minivan. I’m telling you. I’m trying to talk to my wife: I’m not sure a 16-year-old needs a car that she can fit 10 friends in.

 

Allison Lilly: [00:29:28] I know exactly why she did that.

 

Winn Hardin: [00:29:32] Of course it’s got a big steel shell. Before we move on, I do want to point out one other thing you said earlier, Dunchadhn, which is that the advice you just gave to the audience is similar to what we see across many AI startups. Find a small project that’s actually going to yield some good returns. It’s going to fix and solve something that’s really important. Deploy it. Be collaborative with all elements of the enterprise. And then, once everyone’s got that education and that comfort zone, scale. So I just think that’s something that we need to say a lot. And that’ll make the mystery of AI a little bit lighter, a little bit easier to see through. Better transparency issues.

 

Allison Lilly: [00:30:08] Absolutely. Find one predefined, really important use case. Figure out what you’re going to measure. Go roll that out with one team. It might be your operations person. Then expand in operations. Then bring in your safety director. What do they need to solve right now? What’s their hottest spot? Is it forklift driving or is it training around PPE? What is it? Then go solve that thing. Measurable, the whole way through.

 

Winn Hardin: [00:30:29] One hundred percent, without major CapEx expenditures. Thank you so much Dunchadhn and Allison for joining us today on Manufacturing Matters. We hope you enjoy the rest of this week here at Automate. And I’m sure we’ll see you more and more at other shows. If anybody out there has any questions for Spot AI, for Allison or for Dunchadhn, please just put them below in the comments. We’ll make sure they get to them. If you want to check out any past episodes, you can go to manufacturing-matters.com and see all past episodes or look on your favorite podcast platform because we are everywhere. So until next episode, Jimmy, it’s been a pleasure.

 

Jimmy Carroll: [00:31:10] Thanks everybody. Thanks for joining us. Thank you guys.

 

Winn Hardin: [00:31:12] We’ll see you soon.

 

Dunchadhn Lyons: [00:31:13] Thanks a lot.

 

Allison Lilly: [00:31:13] Thank you.