Beyond Defect Detection: How Overview AI Is Transforming Machine Vision with Russell Nibbelink

Why This AI Vision Stands Out

SPEAKER_04 0:00

Welcome to Automation Ladies, the only podcast we know of where girls talk about industrial automation. This episode is brought to you by Inductive Automation. It's just Nicky again. I am actually recording today with Russell from Overview AI, a company that I met and and I met Russell briefly at Automate this year. As uh a thank you to, I will say to Jessica G at NVIDIA for introducing us. I got to go um and have do an interview and learn about uh a little bit about their product. And I was uh surpr I can't I shouldn't say this, but I was surprised that I was impressed because I've been um, you know, we've all been hearing about AI vision systems for a really long time now. Um, and they all seem really promising. Um, and I'm sure they all have their their great features and things, problems that they can solve that weren't solvable before. But it's just one of those things like I feel like there were over the last few years a dime a dozen of them. And I wasn't really seeing anything that that was particularly like jumping out at me as impressive. And I was I was really happy to see um overview and what they were doing. Um, and the focus of that particular interview was me looking at kind of their new uh generative AI feature of generating defects for training, which is powered by NVIDIA. So they were trying to kind of get, you know, everybody in their ecosystem to get a little bit more uh, I guess, people knowing about what they do and how it's powered by NVIDIA and stuff like that. And uh, as you may know, if you've been following me for a while, I've been kind of paying attention to the NVIDIA ecosystem for a couple years now. Um, they've taken a bigger role at Automate, which is kind of cool. Instead of just, I think last year they didn't have a booth at all. They just had a bunch of partners. This year they had their own booth and then also like a partner directory um and a tour of a bunch of their different partners. So, anyway, with that long introduction, I wanted to get Russell on the show and talk about overview and what he does. And so this is a typical episode we're gonna, you know, kind of hear about his journey and everything. But as an entrepreneur, we're probably gonna focus a little bit more on his company and what he does and why he's solving the problem that he's solving. Um, and then I'm um I just made this uh commitment um as we were preparing for the episode, but I genuinely would love to see a live demo um of the system. And I think that you guys would too, as the audience. So stay tuned. We're gonna start our live demos on LinkedIn again probably next year. Um, to be honest, we're already like in November. It's crazy how fast 2026 has gone by. So, with that, Russell, thank you so much for coming on the show. It's great to have you with us. How are you?

SPEAKER_00 2:40

I'm good. Thanks for having me, Nikki. I'm happy to be here, happy to chat. Um, and it was great to meet you at Automate and uh appreciate the kind words about some of the stuff that we're working on and and how we're able to um try to differentiate from, as you said, uh a lot of AI vision systems the last few years have been dime a dozen. But how how do you build a lasting business and you know, while maintaining, you know, a customer uh focus so that it's it's actually useful and you can get into production, get value out of it.

Why Vision Success Depends On Install

SPEAKER_04 3:10

Yeah, yeah. It's kind of, I feel like so. I actually um I may have told you this already, but I started out as a machine vision sales engineer for KeInts in um the beginning of my career, uh sometime between 15 and 20 years ago now. And it was new at the time. Like the technology was relatively new and it was all rules-based, of course. But and and I don't know that this has really changed that much. Of course, we've gotten like very much more mature about it since then. But vision is one of those things where like the the application, the success of the application always depends on the installation. Like there are so many factors that go into it. It's not a product that you can just sell and it's, you know, like a vacuum, it vacuumes dirt. Like no matter what home you're in, it's gonna vacuum the dirt. Um, vision systems are not the same. They're they're looking at like it's always kind of a customized problem set. And even back then, there were, you know, we would have a catalog that showed all these different applications that were easy to solve. Like, oh, check if there's a cap on the bottle or not. And at the time that I was selling vision systems, those were super easy to do. Like it was rare that we found a cap that wasn't detectable by a vision system. But I was in this in in the Bay Area, and so I was working with a lot of kind of cutting-edge technology companies that were not buying a system to see if a cap was on or not. They were trying to solve a problem that was very difficult to solve with vision or any other technology for that matter. And vision systems at the time were so freaking expensive that like you really wouldn't want to use one unless you had a problem that was otherwise difficult to solve. And I have a feeling that we're still in this phase where there are the like the frontier applications now are the ones that have been difficult to solve so far with existing vision technology. Um so, anyway, with all that being said, uh I yeah, let's just start with our first question because otherwise I'm gonna get into like all kinds of crazy stuff right off the bat without even without the audience even knowing who you are. So if you could start us off with giving an introduction of yourself and like how did you get into what you're doing now? Um and in particular, like for us and our audience, we like to know how the peck people got into industrial automation as a field, because most people don't even know about it. Um so I'll I'll let you take the floor, Russell.

Russell’s Path From Tesla To Factories

SPEAKER_00 5:26

Thank you. Yeah, I'm I'm Russell. I'm I'm one of the co-founders um of Overview AI. Uh we've been around for almost eight years now, and really with the with the focus to um make uh machine vision uh as easy as possible for the end user to to ultimately get uh those cameras into production to reduce scrap, ultimately improve yield, and and produce better quality in the factories. And uh that that's really been the mission um from the early days. And the reason why we started it was kind of a very clear idea that AI was advancing, especially in 2018-2019. Everyone in Silicon Valley and in San Francisco, where where we're based, thought that self-driving cars were right around the corner. Two of my co-founders were at Tesla. And at Tesla, they were combining on one hand, ramping up manufacturing um extremely quickly as they were you know growing like crazy. And on the other hand, they had a big self-driving car team trying to put self-driving into the Teslas back then. Um, self-driving cars took a lot, were a lot harder and took a lot longer than we all thought. However, there was a lot of optimism back then, and so this jump from my co-founders at Tesla of like, can we just apply this self-driving car technology onto the factory floor uh to create value was was very um was not a very large leap to take when you were in the right place at the right time, especially around this era in Tesla. And so then they pulled me in, and I had been at a self-driving car research center at UC Berkeley, where I went to school, um, and it was immediately obvious that this was transferable and in fact uh significantly easier to do um than putting a soft driving car on the road would be to put AI into a camera that can do uh valuable things in a factory. And so we started with defect detection. So AI-based defect detection uh in the industrial automation space. So that's how we got into it. And so we we all all three of us, and then we met up with the fourth guy who was a camera hardware expert, and we we started over for you.

SPEAKER_04 7:52

And your background is in the software side of things as well, right? Or can you tell us like what kind of engineering um did you get into and what did you think you were gonna do in your career before you ended up here?

SPEAKER_00 8:04

Yeah, so I studied electrical engineering and computer science. Um, and I thought I was gonna be a software engineer or maybe even an AI researcher.

SPEAKER_01 8:13

Okay.

SPEAKER_00 8:13

And I ended up um after um leaving, after graduating and leaving the lab, we um I ended up working at Salesforce on AI essentially infrastructure for uh data center optimization, which is slightly tangential, but I think there was at the time in the research, there was a little bit of a disillusion in how long it would take to get these AI models um to work for self-driving cars because there was just you had to get 99.99999%, otherwise people die. And that challenge um wasn't as interesting as getting that, you know, these core kind of discoveries done, and now you're solving a tail problem. And um and then when I was um connected with Chris and Austin through Tesla, and they pitched me on this idea that you could, you know, apply this in a much more controlled environment, I was immediately not only it made sense, but I felt like I we had the technical ability to to build this company.

SPEAKER_04 9:18

Had you ever been inside of a factory before?

SPEAKER_00 9:20

Um, I had been do done like a jelly belly uh tour. Okay. So I'd seen, and that's a quite um automated facility in um in the Bay Area.

unknown 9:29

Yeah.

SPEAKER_00 9:29

Um so I had seen some, but it was not um, I would not by any means say I had a lot of manufacturing experience.

SPEAKER_04 9:37

Yeah, it's definitely like a hidden world, I think, to most people, unless people have seen like how it's made, or if they've been lucky to be on a cool factory tour like that. I went on a factory tour of a of a coors brewery in Mexico when I was in college. Um, and so that was like my one and only glimpse into any kind of automated line um manufacturing. It completely caught my interest. When you went to Jelly Belly, did you ever think like was that before you joined um Overview or was that after?

SPEAKER_00 10:05

That was before. I actually went again after and I had a very different I had a very different feeling.

SPEAKER_04 10:10

I know, right? You started looking in all kinds of places after that, especially once you've done vision. Um, like I it's been I don't I'm I don't I'm not as obsessed about it now, but like I couldn't go to the grocery store without pointing out like inconsistent labels or like barcodes that weren't looking right or like laser marked things that were I was like who who sold this thing? Um I think the last thing I did, I I posted like sometime last year, maybe I bought uh a case of Stella and like a bunch of the labels were totally misaligned. And I was just like, how? I think it's 2026. Like, how have we how have they not fixed this by now? Early vision can fix that. Like rules-based vision can fix that. That's super easy.

SPEAKER_00 10:53

Um I mean, I I couldn't agree more. When I, for example, um recently I had a a wearable electronics that um where my battery died uh out of nowhere. And I immediately called the quality team and said, I can solve your problem. And now they're they're a major customer of ours. So so I kind of not only do I look and it annoys me, like labels, whatever. You go to a Home Depot, you go to a Costco. But if it's something I see and I use and I'm like, oh, that's a problem, then I'm immediately looking at that as a sales opportunity for for overview. And uh, you know, this sort of hit close to home with the battery dying. But then turns out, you know, if I if I'm experiencing that as a consumer, there's a good chance they're aware of it and trying to solve it. And we happen to have a solution today that's uh very mature and the right and really the right solution for a lot of um a lot of these sections, whether it's electronics or cosmetic or for consumer.

unknown 11:49

Yeah.

SPEAKER_04 11:50

And one of the things that I so back when I used to do this, it was like big, you'd have to put together obviously you still have to do this, like lenses and lights, and and it's probably a little bit less now with the AI because you can compensate for for different types of images um and and probably more variability than the rules-based stuff in the past. But you'd have to really architect the whole system and then the the the price of those things was insane. Like if you needed a color camera and you needed like five megapixels or something, you'd you'd be looking at like tens of thousands of dollars, multiple tens of thousands of dollars for a system, um, not including the installation and and everything

Transparent Pricing And One-Time Purchase

SPEAKER_04 12:27

like that. And so sometimes it would be like, oh, I can fix this problem. And they'd be like, well, it's not, it's not a big enough problem for us to spend all this money on. And one of the things that did impress me in particular about aside from your technology, was your business model. Tell me a little bit about like why are you guys not doing a SaaS subscription and making it as difficult as possible for people to know what the price of the solution is gonna be up front? Like that. I feel like that's kind of the history of vision. It was like, oh, what's the one of my life models that I've literally applied to almost everything since then is the answer to everything is it depends. Because you could just never upfront tell anybody what a vision solution was gonna look like or how much it was gonna cost until you really like put piece together all those variables and then probably also tested it in the lab. Um and so like you couldn't even ballpark somebody really. But you guys have like a very transparent pricing structure and it's like a one-time thing. So I bet that call is pretty easy when you say, Hey, I saw this problem, I can fix it. And if they ask, like, it's like you actually know roughly what it's gonna cost or you know how it's gonna be implemented, and people only have to pay for it once they don't have to keep paying for it for the rest of their life, which I feel like I just always associated with like, hey, if you're gonna put AI into this thing now, now you're probably gonna charge for software. I mean, you come from the Salesforce world, right? So SaaS is everything where it was for a long time. What made you take a contrarian approach to that from a go-to-market perspective?

SPEAKER_00 13:56

So it's a great question. We we initially started off charging uh our recurring revenue model, like a lot of people do.

SPEAKER_01 14:04

Yeah.

SPEAKER_00 14:05

A few years into it, we realized one that creates friction. Two in a factory, generally the budgets are designed to take uh CapEx expenditures and turn it into OpEx savings. So if you're doing anything that's outside of that model, you're gonna have resistance. However, our cameras are very, very, very, very good at turning CapEx into OpEx savings, reducing scrap, increasing yield, automation efficiency. And so if you instead of trying to fight for recurrent revenue, we really um dove into two things. One, making our products excellent so that we could win a land and expand motion. We we believe in our products so much that if we can sell you one, we think that there's enough problems in a factory that will continuously occur that you will can you will keep buying from us and we will make we will create long-term customers who don't just buy one product and we can get re-occurring revenue from them by solving more problems.

SPEAKER_01 15:11

Yeah.

SPEAKER_00 15:12

And that's been one of the best decisions we've ever made because it's allowed us to scale a lot faster with a lot less friction, and it forced us to make our products so good that people would keep buying them. And then we believed our products were so good, and we the value proposition was so clear that we felt comfortable to have our pricing be transparent. You can go onto our website and we have a shop and you can buy it with your credit card with a clear price tag. And so, but to be able to feel comfortable doing that, we needed the products to actually work. And then the second thing that happened was when we went into this land and expand motion, our business became predicated on how quickly you can get that first camera into production by proving value and working so that they then feel comfortable to buy more systems. And so it really made us focus on uh um uh speed of install and uh speed of deployment. Uh and our systems ultimately are self-service, so it means we try to make the quality of life of the users as easy as possible. And you made a good point of like there's a lot of th factors that go into a vision system. I I like to think of a vision system as a product. People try to sell it like a product, but it's really disguised because it's a project, right? So you're you're doing an engineering project and you're just disguised as a product. And we kind of understand that, but want that project to be as simple and quick in each step of it as possible, so that we can then, you know, get them into production happy and getting a return on investment as quick as

Greenfield Focus Over Retrofitting Old Systems

SPEAKER_00 17:00

possible.

SPEAKER_04 17:00

Uh so part of what I thought too when I saw this was like, okay, it's been a long time since Vision started getting installed and stuff. So I would imagine there's a good amount of opportunity to take your camera and pop it into an existing installation of an older vision system that may not be performing up to par, or because it's you know, older technology. Um and then you don't really have to solve for the entire project all at once, because hopefully the you know existing system may have some of the things right, like the lighting, the position of the camera, the presentation of the parts, things like that. Um and that's I I guess kind of different from solving an entirely new application that has never had a vision system doing that. Um do you see how much do you see uh uh kind of like replacing older technology versus completely green-field applications in in your business?

SPEAKER_00 17:51

Yeah. So there's there's a lot of value in upgrading existing systems because they're already installed. We we made a conscientious effort to not focus on that, even though it would seem very intuitive too. And I want to there's a there's a few reasons why, but I think that something that stuck with me a few years ago was I was interviewing someone from KeIns actually in in Mexico. And I don't, I can't verify this stat, but they said something like 97% of the sales volume from KeAnts in Mexico was coming from greenfield applications. And only 3% was from retrofits. Yeah. And so you know, it made us think long and hard. It's like if we they're very different products to retrofit a camera is a very different product than to have your own set of products, which we do have. And so it's like what were we, it's hard to do two things very well when you're a startup and you're growing. And so we made the decision to focus on an ecosystem of our products so we could control the hardware and then highly optimize the AI to not only be the most accurate in its category, but also have a lot of tools to make it very easy to use because we we we control the whole life cycle of it. But when you're retrofitting a camera, you don't know in advance what that camera's gonna be, the resolution, the connection. And so there's a lot more customization. And with more customization comes less uh friendliness because we we can't be as opinionated and therefore uh make it as easy to use because we just don't have that information in advance.

SPEAKER_04 19:29

Interesting. Yeah, no, that's a good point. That's kind of the opposite of what I was thinking. Um so when do you guys work directly with the customer to do the installation when it's a greenfield project to like select the product and set it up and test it? Or do they is it a completely self-serve, like they do that themselves? How does that work?

SPEAKER_00 19:47

So customer success is is important in any business, right? You need to make sure and customer enablement, you need to make sure that they're happy. And especially vision is still has characteristics of being a project. And and we fundamentally are and our applications engineers and vision engineers are experts in that. And so we build self-service products, right? So everything can be done 100% by the customer. However, for a new customer, we're often gonna hold their hand through it, we're gonna help them with the lensing, the uh, the lighting, um, the setup, just to make it so that they have um the highest chance to be successful quickly. What tends to happen is that for two to three installs, they don't even want our help anymore because they know how to do it and the self-surpose aspect kicks in. But we prefer not to take it to chance that they could mess up the project. So we're there to advise, to consult, and to help from everything from the setup to the outputs of the camera because we have a very flexible um output so we can support all the PLCs, all the MES systems, custom dashboards, et cetera. And so we want to make sure that they they know the optionality they have, so it's as smooth as possible. Um, however, we don't do automation um because that uh is not you know our forte. Uh but we we do want them to be very as successful. So we're gonna, you know, especially with these green fields, we're gonna make sure that it's um has the highest chance of success so that and it's the fastest and optimize optimal way to do it so we can continue doing business with them.

Customer Success Without Doing Automation

SPEAKER_04 21:16

Yeah. If you had um, so you're eight years into this. Uh what do you feel like is kind of the next untapped, or not untapped, but like if you were to look at your plan to kind of grow as a company? Um, and I feel like a lot of success in the marketplace also kind of has to do with having an ecosystem, not gonna do the automation. What are you guys looking for in terms of if somebody's listening to this episode and they're in the industrial automation industry and they're not an end user customer, like they don't work in a factory. Um, are you looking for any kind of like partnerships or or starting to work with like OEM? Play in the sandbox with everybody else that's already in the industry.

SPEAKER_00 21:55

Yeah, it's a great question. Something I spend a lot of time thinking about. The the way that I recently see the world is that you have these tools like ChatGPT and Claude from Anthropic that are are making incredibly difficult and specialized work accessible to someone with no experience. I've never studied finance, I can make a financial document. I've never studied law, I can draft, you know, a legal agreement, I can um do CAD design now with the latest Chetch T model. Um and so it's really democratizing some of these engineering um tasks. And how does this translate to our industry? And I see it as a non-technical person can set up a machine vision system end-to-end if you use an AI agent, if you identify the core steps to being successful in the same way that ChatGPT does with a hardware in the loop system, you can democratize it to a very large distributor uh and value-added reseller and integrator network that already exists today, who a lot of them are very wary to touch machine vision because one, they get bogged down in it. And two, they don't have the expertise to qualify it, to say yes or no, because oftentimes saying no is very important, or you deliver something that doesn't actually work because it's too complicated, or there's some little thing there that they didn't account for, or there's scope creep, and you just you do have an unsuccessful project, and then you're wary for the next project. But we believe that uh, you know, and the first step towards that is our OV Spark, which is the first camera that has an onboard agent, will allow us to blow open the doors to get distributors and um integrators very successful at uh using our products, even if they're non-technical. And then as part of that, we have a companion product called the OV feasibility studio, which is gonna help non-technical people describe the application. And the job of that feasibility studio is to tell them when to say no. And if they say yes, it's it's a full simulated environment of this is the light you need, this is the lens you need, this is the camera model you need, and it creates a full project plan with the quote that they can send to give the customer confidence that the distributor knows what they're doing, and that's all automatic. And so, with those sort of principles, I think that there's gonna be a new paradigm for machine vision um with through these channels.

SPEAKER_04 24:47

That

AI Agents For Integrators And Distributors

SPEAKER_04 24:48

is really interesting. I love that. Um I've actually been in sort of the AI space also for like you you were mentioning back in like 2018 in the Bay Area. Um, that is I was working for an AI startup at the time that was doing um AI and supply chain, so for predicting demand for unreleased products, uh using big data and like Twitter sentiment and um weather data and all kinds of things like that to try to predict like how many people are gonna go to Target to buy this new movie when it gets released, when it's there's no historical data for how it sells. Um and we were doing a lot of uh data analytics, and at the time you had to be a data scientist or at least a sort of business um, you know, technical business user to be able to write SQL queries and things like that to do that type of work. Um so it was very siloed to, you know, there was usually a big bottleneck in these companies for who could get a dashboard made, for instance, right? I I need this dashboard to look at this data to be able to solve this problem. And I am now gonna go to my, you know, data person or my SQL person or whatever and see when they could get me this dashboard. And now, of course, it's very like self-serve. You can ask, you know, pretty much any application nowadays for uh well, I mean, we're looking at this in HMIs as well. Like we just did a demo at PacExpo for WinTech where you could just ask the HMI for a visualization of any given data for like let you know, let me see the OE for this machine for the last month, or you know, different things like that. So you don't have to have these hard-coded, I can only see the dashboards that somebody came up with when we installed the machine. Um so that's kind of been in the in the works for a long time in different areas. I've always been kind of skeptical of the you don't need to be an engineer to do engineering because now there's these tools that tell you what to do. Because if you don't know enough engineering to know if the tools are right or wrong, like, you know, kind of like not the same as with self-driving cars, but in the manufacturing industry, there can be some catastrophic results if you do some bad engineering on the on the plant floor. So I'm gonna give a little shout out um to our friends at Vincian here. I happen to be drinking uh tea from one of their wonderful cups that they gave me that says artificial intelligence, real manufacturing work. Um, and I'm a really big fan of a lot of what they do, including their there uh also have like an online store where you can put together a system and you can price it and you can buy it online, um, which price transparency and sort of the ability to buy things in our industry um has been a frustrating challenge uh for me. I actually worked at an early stage startup also last few years that was trying to solve one of those problems in the distribution space. Um, because it traditionally has been very difficult to buy things in our industry. Like you got to call a distributor and get somebody out to give you a quote, and you know, you got to wait for the PDF and you got to set up an account. And you typically have no idea what things are gonna cost until you talk to somebody. Um anyway, but I one of the first articles I read from Venction was like, you don't need to be an engineer to build machines because our website like walks you through it. And I I don't like big claims like big sweeping claims like that, like or things like, oh, uh we're in the post-plc era now because you know AI-enabled computers can now do they can like pan plan your emotion paths and logic and stuff like that. And I think there's nuance to these things. Yes, there's a lot of leaps that you can make with that, but is the PLC gonna go away? No, I I don't really think so, right? But we want to say these marketing terms. But I think one of the coolest things about AI right now in our industry is like at least lowering that barrier. Like you don't need to have as many years of accumulated experience and expertise to be able to tackle a lot of these things now with the help of AI like guiding you. Um, and that is actually one of the things that I that I initially was pretty impressed with um, because I was shown uh the the sort of process of doing the generated feature or defect detection so that you can train your vision system without having a bunch of examples of the bad parts, which was a big part of what my job was qualifying these applications back in the day, was I needed to get examples of almost all the defects. And to make sure that, you know, we test them in the lab and make sure that we could actually catch all the defects. Um, and then we had to obviously manually train the model to catch all these defects.

Sponsor Break Inductive Automation

SPEAKER_03 29:21

Here's a quick message from our sponsor, Inductive Automation.

SPEAKER_02 29:24

Every path, every project, every goal, every solution begins with an it. What is it? It is your breakthrough moment. It is when you build a foundation for innovation. It is your best idea that will solve unique challenges and revolutionize a process. It is when you integrate new technologies that will transform entire industries. If you have a vision for where it should end, then we know where it should begin. It starts with ignition.

SPEAKER_03 30:04

Thank you to Inductive Automation for sponsoring automation, ladies. Let's go back to the episode.

Generative Defects To Validate Faster

SPEAKER_04 30:11

Can you, I guess, walk us through all the different ways that you guys use uh generative AI or AI? Um, not in the in the actual inspection part itself, but you've mentioned like several use cases now, just in the rest of the business, setting it up and things like that. I don't know if that's the right question or or how you want to like talk about that, but yeah, I want to talk about that. How are we lowering the barrier and all this like time to market and time to value and all that kind of stuff, like using AI, um, aside from what you just mentioned, which is of course, now you don't necessarily have to have a vision specialist on staff or bring in a systems integrator every time you need to set up a new camera.

SPEAKER_00 30:56

Yeah, I think time to value is a great word for uh why generating defects is useful. We we've really crossed this barrier in the last year or so where you can generate such hyper realistic defects because you have these world foundation models that understand the material, the texture, how the physics interacts with it and the light, that you can create a dent in metal or a bubble in glass or laminate that is so realistic that not only can you then send that to your various inspection tools, you know, your AI inspection tools or even your classical tools, uh, and that's big because you can get that data and train your model so much faster. But what's oftentimes the longest period of lag to get to that proof of value is you need to validate the system. You need to make sure that the system's working before you turn on the automation. And if you can if you can generate a bunch of scenarios of things that can go wrong and then back test those against your AI model to prove that it's catching something, uh you can feel not only you can have a data-driven approach to getting validated and qualified faster. So, for example, let's take an example like a scratch. If I can put a scratch on my part and then make the scratch small and big and move it around to every single spot on the part, and then run those 200 images through your inspection as a back test or as you know, a simulated environment, and it catches all of those, uh, you can then have the confidence that when you turn it on, it'll replicate that in the field. And so it really becomes a game changer in in you know, reducing the time to value on both the front end side of uh the AI, which is collecting the data, because you can generate it, or if you have rare defect modes, you can take one of them and expand upon that by generating similar things, smaller, bigger, uh, and then, or just you know, create scenarios that you haven't even seen before that you expect to be able to come because that person has that tribal knowledge from a previous generation of products that this could occur. Um and then of course, yeah, we talked about the validation part. So that's on the generative AI of the the image generation, right? The we call it synthetic data generation on the image side. For us, we call it our defect creator studio. However, there's the back half where you can generate uh using these like AI agents uh um useful uh visualizations, the business intelligence, the manufacturing intelligence, the HMIs, all for the operator, the quality manager, the plant manager to use that data as effectively as possible to not only reduce scrap, but ultimately to drive up yield. Uh for example, if I say I want a dashboard that is giving me a Pareto chart that's then split shift by shift to understand what's going wrong, where to focus my continuous improvement efforts, and then you find the number one failure mode and you want to know did I even fix it, you can use that camera to see over time if you fixed it. And if you have, great, you go to the number two issue, the number three issue. And that is a much faster way and a more data-driven way to uh uh drive higher yield, which is good for uh not only for the margins, but also for the environment to waste less scrap is is is hugely important. Um and you it's very hard to improve what you can't measure, and this is just a way for you to measure now your improvements.

Dashboards That Drive Yield And Less Scrap

SPEAKER_04 34:44

Yeah. I remember I um I had to do a training on a vision system once, and I guess they had such a tight budget that they didn't even install a screen to see what was going on with the system. Like it was just it all it required was a monitor, like a basic computer monitor. Um and they had bought an old vision system from Keynes, so not the latest and greatest with a with a lot of the new features. They Keynes' like model, at least back then, was kind of like if you want you want a cheaper product, you can buy an older product. They don't really make like cheaper models, right? It's like we're gonna keep the existing generation around and that's gonna be the cheaper one. We may lower the price of it when we come out with a better product. But uh yeah, it was it was an injection molding facility, and they had A bought the wrong camera because it was just like they their budget was so tight, and I guess there was miscommunication in it, it was sold to a systems integrator by somebody like out of the territory, and they bought totally the wrong thing. But I just remember thinking, like, how why would you invest in this and you don't even want to see what is going on? Like, you're gonna not spend a couple hundred bucks or whatever it may cost to get a monitor. Um you just what want the product kicked out when it's bad? But I I was like, I was just kind of blown away that like not only did they not even want to see, they didn't care to see what was going on, they were definitely not collecting any data, doing any trending. Like it was just is this a good part or is this a bad part? Um and they ended up with a system that didn't let any defects through, but it definitely scrapped way too many good parts. Um, but again, they didn't really have any good way to use the information that they were getting from the vision system to improve anything, other than just not sending a bad part through. Um so it was really, I think the the thinking back then was can I replace a person that would otherwise have to look at this for quality inspection and just say, is this gonna pass or not? Like can we send this on to our customer? Um, so I'm hoping that more facilities out there are really thinking about it in the in that sense of looking at what data can we gather from this to improve our operations in the long run and the quality of this product and the yield and all of that, and not just, hey, let's make sure that we don't send a bad product out the door. Um Do you find that that that HMI I know that's something that you guys just recently uh improved or released, right? Is some additional um ability to look at the information that you need on the plant floor um from that perspective. Can you tell us a little bit more about that um in terms of the customer demand that drove that or any trends that you're seeing um aside from what you just mentioned?

SPEAKER_00 37:25

Yeah, I think that there's I think I mean I want to live in a world where we're using less energy to produce things, we're less wasteful, and we make better things. And so, you know, making better things, a vision system helps make better things because you you throw stuff away when it's bad. But it's much more exciting if you can actually make the same things for less energy by having a higher yield and you know, reducing scrap, because that's like a real, there's a real benefit to that for the planet and also for the factories on from a you know a return on investment perspective. But to access that, the only way is to use the data that you have to drive better factory processes, continuous improvement, and just make you know, to improve your automation or your processes. And so uh to me, it's always it's been like if you can make that data more accessible, more customizable to their scenario, and then more in their face, they're gonna use it and drive better decisions. And so as you you mentioned, right, with these tools, you with uh these agentic sort of dashboarding tools and HMIs, it's become so much easier that it's it's such it's a kind of a small leap to to push that front and center of the customer so they get the full value of you know this intelligence that's sitting on our factory floor. And you know, we we actually just display a web page with that through our camera, which is on the local web server. You can display it on an HMI, for example, that WineTech sells, or you can display it um on a laptop, or you can be in the office and go to the IP address of the camera, wherever, you know, if they're on the plant network to see it there. And you can have different types of dashboards. But ultimately, I want to drive towards that future of higher yield and less scrap so you can make things more efficiently.

SPEAKER_04 39:12

Yeah, that's one of the things that we also kind of, you know, from an A, and we're considered kind of an old school HMI manufacturer because we've been making those screens, the touch screens, the traditional HMIs that just sit on the machine themselves for the operation of the machine, but then also, you know, trending and data and all that kind of stuff. Um but we have the same kind of philosophies. Like there, there are different users of the data that need it in different ways in different places. So you may have the screen for the operator, you may have a screen for maintenance person, um, and then you have an entirely different set of data that's being shown to you know the plant manager or uh somewhere else. And you need to have some flexibility in where you can see that and what you can do, and ideally not be nickel and dimed for every instance andor every screen that you need to either make or display or whatever. Um, so I love that flexibility because it is really important. If you're gonna really get value out of that data, it's not a it's not a one size fits all model as to what data is useful where. It's there's a number of different types of applications where that data should be served and available and useful.

Safety-Critical Wins And Favorite Use Cases

SPEAKER_04 40:18

What would you say is, can you tell us like what are the coolest applications that you've seen or that you guys have been able to solve that you think would be fun for our audience to hear about, if any, that you're allowed to talk about. We know most things are under NDA.

SPEAKER_00 40:31

Yes, as yeah, factories are always tight-lipped about you know what they're doing. And it's always hard to do co-marketing because no factory wants to admit that they have quality issues, you know. Right.

SPEAKER_04 40:40

So it's always like, oh, you want to co work it?

SPEAKER_00 40:44

Yeah, we solved they don't want to admit that they had a problem in the first place. That's always a uh a funny um, you know, co-marketing challenge, I'll say. Um, in terms of applications, um, you know, I think I I'll speak less broadly, but you know, we we have a customer that was doing some parachute stuff. So it's always you know exciting when you know that your camera is making sure that that parachute is built correctly, because if that parachute is not built correctly and that ripcord is not, you know, made exactly to spec, you know, that that's some serious consequences. So so that's not you know, that's like a you know, a small, it was a small project for us, but it's always reminding me that you know this is this this this matters, right? Yeah, or for example, we do um inspections on the brake assemblies of cars on the welding. So if there's a weld on a brake assembly of a car, you know, you really want to make sure that that brake is not for the car, it's not gonna break on you. Um you you need to stop the car. Um or you know, so that's like the safety critical stuff always makes me proud because it's such it has just such a tangible impact on that. I mean, there's always some fun stuff when you're inspecting um, you know, things that I like, you know, things that I buy. So, you know, without getting into the things that I buy, and I'm like, oh, like I know that I'm buying this, I would buy it anyways, and now I know it's built slightly better uh because of overview is always um exciting, whether it's you know consumer electronics, e-readers, uh wearables, um, that kind of thing is like very, you know, it touches me personally. So that that I think I I enjoy that a lot.

SPEAKER_04 42:22

I um I wonder now if what what my NDA extends to, and it's been it's been some several years. But I remember I once bought a piece of technology that was very new and I wouldn't have otherwise probably bought it. Um because I like to wait a few generations, like when something new and hype comes out. Uh but I I just had to buy it because I was like, well, my product helped design it. And I'm like, I walk into this meeting with this, with these engineering group like all the time, and they're my I feel like they're my friends now. So I'm I've been like I I just wanted to go like wear the thing. Um, because you have this kind of personal connection and you kind of know, like, yeah, these people built something really, really super cool. Uh so that's cool. Have you ever been skydiving?

SPEAKER_00 43:02

I have been once, yeah, on a tandem jump.

SPEAKER_04 43:05

Would you go again?

SPEAKER_00 43:06

Oh, yeah. Now that I know some, at least maybe maybe that's if you're using the parachute brand that you guys work with. Exactly.

SPEAKER_04 43:16

Um, my CTO at one of my old startups, he was a big parachute, uh, big um skydiving guy. And so I always wonder if there's like um the connection between people that are crazy enough to to build startups, um, if they need those constant like ups and downs and adrenaline rushes and your existential fear that you may your life may be over soon or your your business may be. Um, but you guys are doing, I I from what I've heard, and I I hope this isn't any kind of non public information, but um you guys are a profitable startup, which is a an oxymoron in some some ways um from the world that I come from, like VC funded startups usually takes a while to get to that point. Um I to me that was also like another big sign of just being impressed, um, that in The world that you guys operate in, um, in the space that you are being at that point at this point, like eight years in, um, without just having to rely on a bunch of strategic investors that are just willing to keep pumping money because they they really want to see the technology succeed. I feel like there's a lot of that in our space. Um they're not expecting profits for a very, very long time to come. And but they're, you know, they're kind of betting on these really big things. And that's kind of how I see humanoids. It's like they'll eventually be ubiquitous because people just will not stop putting money in them. They may or may not be the right solution for lots of things, but if you have enough people willing to lose enough money over a long enough period of time, it'll eventually happen. Um, but lots of other things are more dependent on you know investor sentiment and shifts in markets, and it's not this like big moonshot thing that everybody wants to pour their money into. Um, and industrial software and hardware tends to be one of those less sexy things.

SPEAKER_00 44:56

Um except humanoids now.

SPEAKER_04 44:58

So well, yeah, that's what I'm saying. I don't know that I would call them that exactly. Uh they're a little bit more flashy than than your industrial tech. Um how long did it take you guys to get profitable, if that's something that you can share, roughly.

Pandemic Pivot And Becoming Profitable

SPEAKER_00 45:12

So, I mean, uh the history of our company is interesting because we really did a big pivot during the pandemic, right? In the pandemic in the US, a lot of factories shut their doors. Uh, no one was investing in in manufacturing. They were, everyone's hunkering down. A lot of people couldn't come, the labor, you know, people couldn't come into the factory to work because of quarantines. And so it was sort of a time for us to do some rich um in um sort of introspection. And I I briefly touched on how we went from a um a recurring revenue business to to do a one-time fee. And that really happened during um uh the pandemic when we had this opportunity where we were able to reinvent ourselves um in a non-disruptive way because we had, you know, there was a shutdown. And so since then, you know, we've just we didn't set out to become profitable, but we wanted to find product market fit and build products that were genuinely useful, good value, and solve problems for customers that could also scale. And I think we did a pretty good job of that. And as we s invested in go to market and sales and marketing, we got more and more customers, we made them happy, they told their friends, they bought more. You know, at some point, relatively quickly, two and a half years after that, um, you know, we ended up just looking at the balance sheet one day and saying, wow, we're profitable. We should start hiring more. And so we were always hiring, but now we're not like, okay, well, we're we should, you know, the venture-backed company is like, if you're profitable, go hire more. So they're like, okay, let's turn on the jets, and we started hiring more. However, the growth sort of continued to grow even faster than that. And so we are really hiring aggressively, and we're trying to recruit um the best people we can around the world to help continue to grow overview. Um, but our growth is just staying ahead of that. And so um, you know, we've been able to remain profitable for the last few years. So it's been it's been a very exciting position, and we are trying to keep growing and keep releasing products that continue to make it easier for customers, more accurate, solve more problems um in this space.

SPEAKER_04 47:23

Yeah, and that is, I think, a real testament to the fact that you built a quality product that is easy to use because your own users will also be your best marketers, right? Because engineers do talk to each other. Um and there's plenty of there is plenty of stuff in our industry that people use they don't love. They use because they have to, they make it work, but they don't go evangelizing about it to their friends, right? And so I think it's a rare treat um when engineers get a chance to use something that really either works as well as advertised or actually exceeds their expectations, right? And it becomes easier to use over time, um, and not just because they went to like an expensive training and finally got the hang of something that has a large learning curve or whatever. Um and and that I think is a is an underlooked at, I think, aspect of go-to-market in our industry because we're so used to like this whole engineering, like you have to kind of go in and like prove the value. Um and you have to, there is there needs to be an expert that isn't the user. Um and so yeah, to me, I I was just like really happy to see that because there's not a lot of companies that take the uh what like you said, um, land and expand, you know, type of uh model with that in terms of making it really easy to get in the door um and just get started and and make that starting experience really good for everybody. Um that is one of the things that really drew me to the company. I can't I worked at an enterprise AI company back in 2018, and so it was not accessible at all to, you know, small companies or uh you couldn't set it up on your own. It was a whole, you know, we'd have to go in and do a multi-month engagement. But that's also the maturity of the AI tools and stuff at the time. It was just the only way to do it, and the only companies that could afford it were ones that were willing to pay a very high recurring cost to that. Um, because there was also a big cost. It was not a you know, set it and forget it on our side. There was a big like SLA delivery model with it, and it models have to continually be retrained, and there's a bunch of oversight and all kinds of stuff like that. Um and I love, and that's part of why we do the show, is like I want when there's a something really cool, like I want everybody to know about it. Um and I remember too when I was working for that company, and the, you know, there was kind of like the beginning of like, hey, there are AI models now that can solve a lot of problems. And I would meet with people that would come ask me if we could do this or do that with AI. And I was like, well, that's not our niche. So, like you know, said it's you have to be careful not to stretch yourself too thin and really focus on specific applications. So I think also at the time, too many people thought AI was something magical that could do many different things. And it's I mean, the the foundation models today are kind of magical and they can do a lot of different things. But at least back then it was like everything has to be very specifically trained for a very specific application. Um and so I'd have to meet I'd have to meet with people and they'd be like, Can can your AI do this? I'm like, no. I mean, it could if we put enough effort and and money into it, but it but it doesn't. Um and I remember thinking, like, oh, there should be like a marketplace for AI models where people could discover what AI is out there that's capable of doing different things, like an angel list, but like an AI list. Um because it just seems such an inefficient market strategy to me to be like, hey, I have this crazy cool thing that does this. Um, but only people that will like let me knock on their door and take my meeting are gonna know about it. And you know, everybody else will have no clue. Um but I think now, like obviously, with the the internet was around at that time, but there's just I think a little bit more like parity in in finding information, especially too with the with the AI, if people are searching, you know, using Chat GPT or whatever. Um you guys focus on that? I'm curious. This is a whole nother conversation, but the marketing side of things, like if you ask Chat GPT, hey, what what kind of vision system do I need? Do does it recommend overview AI? And do you guys put a lot of thought into that?

SPEAKER_00 51:33

Yeah, we definitely put thought into that. Um it's it's newer. So we think that we have an outsize advantage in getting recommended because we're all the whole world is starting at the same place. And so for us, it's like um it's definitely something that we see we see that we can put an outsized effort to get outsized results into it. So for sure, uh a big focus for us.

SPEAKER_04 51:56

Cool. Well, I think we're coming up on on time here.

Events Hiring And How To Follow

SPEAKER_04 51:59

Is there anything that we haven't talked about that you want to tell my audience about or make sure that we know about overview? Before ahead of us doing an actual demo where people can see, it's super frustrating to talk about like dashboards and not be able to see them. To me, anyway. I'm a visual learner in a lot of ways.

SPEAKER_00 52:15

Yeah, no, I think that this is a great conversation. Um, yeah, we we uh it was a wide ranging and we we covered some good topics.

SPEAKER_04 52:22

I'm all over the place and I like to go on lots of tangents and I like to ask a lot of compound questions that don't always make sense. So I appreciate your uh your willingness to stick with me here.

SPEAKER_00 52:31

Thank you for leading leading the conversation to interesting places.

SPEAKER_04 52:34

Yeah, no, no worries. Um so yeah, my last question is typically like tell our audience if they what like who would you like to reach out to overview and and how? And then if there's anything cool from like if people just they're kind of you know tangentially interested in following you guys, where should they do that? And is there anything cool coming up that we should be aware of um in terms of events or releases or other other things?

SPEAKER_00 53:01

Yeah, I I would say that if people want to follow us, we're we're on LinkedIn at Overview AI. You can sign up for our newsletter on our website. We're also gonna be at Electronica um in Munich on November 10th to the 13th, where we're debuting uh a new unsupervised uh anomaly detection model that's incredibly easy to set up for PCBA. Um so for you know any sort of electronic server, we can look for uh defects, um, missing components with only five good examples and uh and a highly trained model that can uh detect very small things uh in in a very cheap um and fast to set up way within our ecosystem. So we're gonna be debuting that at Electronica Munich and um yeah on November 10th. So we're excited about that.

SPEAKER_04 53:46

Very cool. Um, what kind of roles are you hiring from? And are any of them, if they're like more so related to the industrial automation side than maybe like AI product development on your side? But tell us um if you're pulling anybody from our kind of side of the world, uh, what types of job opportunities do you have open that people should be looking out for?

SPEAKER_00 54:05

Yeah, absolutely. We're looking for automation engineers, we're looking for controls engineers, um, manufacturing um engineering people who want to be more customer-facing. So we're always looking for, you know, we call them forward-deployed engineers, we're looking for solutions engineers, and we usually pull from, you know, uh X usually extroverted manufacturing engineers or that. We're looking for vision engineers, um, marketing folks, uh, technical marketing folks who have manufacturing backgrounds um as well. We're also looking for pure sales folks, um, engineers on the software side, the hardware side, the embedded side. Um, and that applies not just to the US, but Mexico, Canada, uh, China, Taiwan, Japan, Korea. So you name it. We're we're hiring.

SPEAKER_04 54:46

Oh, wow. Okay. So I'm assuming that's overview.ai slash careers or something like that.

SPEAKER_00 54:51

Slash jobs, yeah.

SPEAKER_04 54:52

Slash jobs. Okay, cool.

SPEAKER_00 54:53

And um, yeah, we we pay above market rate and we try to uh give everyone um equity in the company so they can uh be a part of the growth um that we you know we all build it together.

SPEAKER_04 55:03

Yeah, and that is a really great opportunity right now um in terms of especially with your positioning, I think in the market, your uh profitability and stuff like that. Um I I've definitely worked for some equity that didn't turn into anything. So I know that's you know, some people have that startup bug. They want to try things that are a little bit riskier for a potentially outsized reward um or just for fixing a freaking problem that that's been bugging them for a really long time. Um but uh I, you know, I don't give any investment advice, but I would say it'd probably be a pretty good bet that you may get something from uh Overview AI equity if you decide to join this team, from what I can tell. And that's just my unsolicited opinion.

SPEAKER_00 55:44

That's the hope and the plan. So we'll see, we'll see.

SPEAKER_04 55:46

Yeah. Absolutely. So oh uh, I guess that's another question. Um do you have uh well, maybe, maybe not for this conversation, but I'll be like, what's your exit plan?

SPEAKER_00 55:57

Are you trying to IPO or um Yeah, I mean, right now we're we're we're growing so fast, we want to continue the growth. We want to release more and more products, um, not just um in the 2D vision, but eventually 3D vision. And then eventually um, how can you put that into different form factors? So mobile, uh, whether that combines with you know robots or drones, um, that kind of thing, you know, there's just so much. If you can see the world, understand the world, and then you know extract um value, there the roadmap is extremely exciting, especially with the advances in physical AI and these world foundation models. Um you can apply that to some incredible things in factories, and we have that bug to turn those uh incredible things and make it accessible, easy to use, and very deployable and scalable. So that's what we want to continue to do. Um, and we'll see where where that takes us.

SPEAKER_04 56:48

Very cool, very exciting. Thank you so much, Russell. I appreciate it. It was a great conversation. I'm very interested to see what you guys come out with next. Um, I'm not gonna be at Electronica, but I feel like that is an application I remember from back in the day um being very expensive to do due to the resolution and like the all the different ways that you have to present a board. I mean, the training is just ridiculous, like that the amount of time. I'm sure it's gotten better since I was doing it, but uh that is probably a very high value use case. I can see why you're excited about showing that off. So, with that, thank you so much. Thank you guys for listening. Reach out to me if you want an introduction to anybody at Overview, um, if the other bajillion ways to get a hold of them don't work for you, or if you want to just you know have a little chat about why I'm excited about this stuff and and these guys and their company. Um, and then Russell will have a uh guest profile like all of our guests on automationladies.io where there will be some links as well to their company and other things. And uh if you're interested in this, definitely keep an eye out for us doing a live demo on LinkedIn sometime in the near future. Thank you so much. Bye.

SPEAKER_00 57:54

Thank you, Nikki.

SPEAKER_04 57:56

Thank you for listening to AutomationLadies. If you like our content and you want to stay in touch, please connect with us on LinkedIn, follow the show page, subscribe to our YouTube channel, and you can send us a message or a copy on our website, automationladies.io. We look forward to getting to know you.