Future-Proofing Enterprises: AI Readiness and GPU Innovations

Unlock the future of enterprise AI with our guest from Plainsight Technologies, who brings their expertise from industry giants like Microsoft and Google. Discover how Plainsight is transforming businesses by turning everyday cameras into powerful data-generating tools, creating digital twins of the physical world. This episode promises insights into the groundbreaking concept of "digital world building," which is revolutionizing sectors like agriculture and manufacturing with AI-driven visual data.

Explore the intricate challenges and innovative solutions in deploying machine learning models for computer vision. Learn about Filterbox, Plainsight's Dockerized framework that simplifies AI adoption for enterprises, even those without specialized expertise. Vision data filters, customizable computer vision apps, are highlighted as key tools that integrate seamlessly into existing IT infrastructures, empowering businesses to focus on leveraging data for insights without the burden of maintaining complex AI models.

Examine the crucial role of technology in quality control and compliance within the manufacturing sector. With early defect detection and consistent quality as focal points, we delve into software-defined compliance and the resurgence of American manufacturing. Our conversation extends to the AI readiness program aimed at making AI more accessible and impactful. Finally, we touch upon future innovations in GPU efficiency for edge computing and the collaborative efforts needed to drive scalable, cost-effective AI solutions. Tune in for a comprehensive understanding of how AI is set to revolutionize business processes and operational excellence across industries.

More at https://linktr.ee/EvanKirstel

Transcript

Speaker 1: 0:03

Hey everybody, fascinating discussion today on practical applications and use cases of AI around computer vision technology, a company that Plainsight is focused on, kit. How are you, hey, evan? Good to see you.

Speaker 2: 0:36

Good to see you Really excited for this chat. You're doing some amazing work and really want to dive in. Before that, maybe introduce yourself a little bit about Pl in the software business for a little too long 20 plus years, you know started out at Microsoft and then worked at Google as a product manager for Kubernetes, which is an open source project you may have heard of. We just celebrated our 10 year anniversary of Kubernetes, which was very exciting to be in Mountain View with the rest of the team. And, yeah, I've been building, you know, software, open source communities, developer tools, cloud infrastructure for many years. This is my first foray into artificial intelligence, which, of course, is so hot right now, and I've joined Planeside Technologies six months ago. It's actually a brand new company. We just started and we were able to acquire some yeah, some intellectual property and some logos and patents and things like that from some other companies, which was great.

Speaker 2: 1:31

And we kind of formulated this new thing and we're focused on computer vision, but I think with a pretty unique spin, which is that we want to generate enterprise data from cameras. We turn cameras into spreadsheets is kind of the cheeky way of putting it, but the idea is that businesses today are in the process of what I call digital world building. If you kind of think about, yeah, there you go, make your cameras count, that's us you think about building the digital world you know you've heard of digital twins and you think about how companies in the future are going to be all optimized and making decisions and building scenarios and everything from the digital world. They need to have that representation of the physical world. I think one of the key ways it's going to happen is people are going to use cameras and images and video data to generate that map, that digital map of their enterprise. And so that's what PlainSight does.

Speaker 2: 2:30

We help them build AI filters, we call them, which are basically apps that wrap AI models and let you generate really interesting enterprise data for everything from inventory to quality control, yield estimation, defect analysis. All of this is done using our filters, which are Dockerized apps. They deploy in Kubernetes. You run them at the edge, run them in the cloud and you can process loads of data and generate useful data that you flow into your existing workflows and data pipelines. So think about how you might have inventory data in SAP. Wouldn't it be great if, instead of that being entered by hand, you had a camera that just updated it automatically and kept your you know, pallets or gravel pile or closets full of materials up to date at your fingertips, so that the people who are kind of in the data side of the world could have a better grasp on what's happening in the physical side of the world. That's really the idea of what we're building at Plantset Technologies.

Speaker 1: 3:31

Fantastic and I love how practical this approach is. You're dealing with real world challenges and so diverse the number of sectors and industries you're in I saw agriculture, manufacturing, as just a few. Tell us about that approach and the kind of problems you're encountering these days.

Speaker 2: 3:51

Yeah, it's interesting because a lot of the strategic feedback and advice that I get from people, especially when I was first taking on this new challenge, it was like, hey, go pick a vertical and go win that vertical, and I think that that seems like a very wise direction because you want to really go deep and build expertise. But I kind of look at it a slightly different way, coming from an infrastructure background. I think about this computer vision capability as an infrastructure capability, and most applications and enterprise workloads don't need that much computer vision. They need a little bit at the kind of at the what I call like the front end of their data pipeline. Right, if you think about, like the camera as the input, a very powerful input sensor that then can flow into data pipelines, then go into data lakes and other things, and you kind of put on this infrastructure hat which requires a certain level of abstraction, right, you have to have a little bit abstract thought here and realize that you know a cow wandering in the prairie and a car and a drive-through, you know those are actually kind of similar things as far as the computer is concerned. And by giving developers and architects these apps right, these filters, which are really enterprise data apps in order to take camera data and generate tabular data. It becomes almost like a new component, right? A building block, or a Lego brick, if you will, for building these enterprise architectures that need some computer vision data. But it is not the entire app, right? They have many other parts to how they're collecting data, through sensors or through manual entry or through, you know, internet traffic, et cetera. So how do we combine that visual data into the system? So I really like to think about it, as our focus is on a particular point in the software stack, which is really this, you know, using AI and computer vision techniques which, by the way, it's not just about machine learning for object detection, it's also calculating the physics right, how do we convert a two dimensional image into geospatial data, right, and physics data? How do we understand the size and shape and position of different objects within the physical world? You know that capability as a, as a computer vision app, as a filter app, but then also is outputting data that can flow very easily into, you know, the Databricks or BigQuery or whatever downstream system. So the focus becomes on this software, you know, point in the stack, and that, to me, is really the way to think about it. So that applies, then, to many different industries Now, where we have chosen to kind of go to market and where we've looked for customers, identified use cases, et cetera.

Speaker 2: 6:26

Where we've seen traction, we'll put it right Is it a lot of sort of livestock and agriculture. We see a lot of traction there. Manufacturing, particularly construction materials, roofing tiles and and solar panels, things like things of that nature. We've seen traction in a variety of food production scenarios.

Speaker 2: 6:51

Medical supplies Interestingly, medical supplies are an interesting one in the US in particular, because your storage facilities are not actually your facilities, right, you have medical supplies in hospitals. So how do you keep track of all the medical supplies that have gone out to third party storage? So there's a variety of these different kind of workflows that today are done manually and tomorrow will be human plus machine augmented. And the advantage of having the machine learning and this kind of data layer to it is, you know, you can make it more accurate, you can do it with it's reviewable, you can quickly integrate that data across a variety of sites, right, you gain some consistency. It doesn't necessarily remove the human element, because there's other things that you know, obviously people have to do, but it does become almost like a, you know, collaborative I hate to say co-pilot, right, but it's. It's a sort of you know, almost cyborg collaboration where you're able to use these tools for for recognizing images, you know, and it can be from a variety of different image sources too. Fixed camera is definitely a big one, but we also see applications with mobile, with drones and all of that kind of combined video and image data into this larger data set that then goes through a data-to-model lifecycle where we're creating proprietary models with customer data.

Speaker 2: 8:14

Obviously, segmenting the data for privacy reasons it's a big issue. I hear about a lot is enterprises that I talk to want to protect their data and gain a strategic advantage from AI. They don't want to give their data to, you know, some you know third party who might use it to resell to their competitors. You know like, you know jokingly like not my customers, but like, jokingly like if Pizza Hut doesn't want their data going to Domino's, you know it's like that's kind of the idea of of the data privacy privacy I think we see in AI today. So those kinds of concerns, I think, become a big part of the practical nature of making this work. But there's so many potential applications for computer vision.

Speaker 2: 9:11

I guess the other key thing to realize is the limitations. We're not in a position where, for example, you know, detecting criminal behavior or criminal intent is not something that you know we can claim or anyone can claim to really do, despite what you might hear. There's limitations when it comes to the size of objects and the speed of objects and things like that. So we have we spent a lot of time educating customers on you know, kind of the readiness and what's what. And then the other, so that's one angle is like well, what's feasible, right, and like our view is like if the data is not in the frame, we can't really express it to you as data. So that's an important part of the story.

Speaker 2: 9:34

The other side of it is what's the ROI? And you know, again, being a practical, you know this is what you were talking about real world practical applications of AI. You know we have a kind of a little rubric which is like if anybody starts with that, wouldn't it be cool if that's usually a signal that they haven't done the ROI calculation on the AI. And so you know I take customers through this journey of kind of talking them through it like okay, so are you going to stop doing the manual approach? You know? Are you using this data today? How does this data helping you? How are we improving the data accuracy? And the significant investment of GPUs which is an important part of the story cameras, et cetera, lighting all these issues that make up a good computer vision solution? Is this going to pencil out for you?

Speaker 2: 10:18

We really spend a lot of time educating and helping build the business case for specific things. And then one of the nice things about our business model with this sort of app-centric approach where we're licensing apps, we're not trying to sell computer vision to people who don't have a computer vision problem. You know it's called me old fashioned, but I think the customers should feel like they're getting a huge benefit from adding computer vision capability to a very specific place and not feel like, oh, the sales guy is trying to get me to buy computer vision everywhere because they're in the computer vision business. I want them to say, look, no, I get this one filter. It solves this one problem, gives me a new piece of data I didn't have before, and then, all of a sudden, my business is more efficient. My customers are happier. That's kind of the idea of what we're trying to do strategically.

Speaker 1: 11:03

Wonderful insights and examples there and, just looking at your website, you talk about something called computer vision filters, and I assume this isn't like a Snapchat or Instagram filter. What is a computer vision filter and how does it work across all these different use cases, from preventing wildfires to the counting of sheep, as you just illustrated?

Speaker 2: 11:27

Yeah, yeah. So we were inspired. I gotta say we were inspired by Instagram and Snapchat filters, because it's actually an easy way to understand as an app. Right, you think about it as an app. I think it's really important.

Speaker 2: 11:39

When we talk about machine learning, people like to talk about models, right? They'll say, oh, we created a machine learning model. A model sort of just sits there, right, you spend a bunch of time training it and it can do cool stuff. But a model on its own. How do I host it? How do I deploy it? How do I version it? How do I add business logic to it? Right? That sort of left as an exercise to the reader, so to speak. Right.

Speaker 2: 12:04

And what I think is for people to adopt new technologies, they need a frame of reference. They almost need a black box. Frankly, I mean, If you want to adopt computer vision, you don't want to become a computer vision expert. You don't want to become a machine learning expert. By the way, most enterprises don't have access to all of the people doing machine learning. Those people are gobbled up into some of the big businesses that are really leading this AI revolution. You don't have access to those people.

Speaker 2: 12:32

So what do you do, and what we try to come up with is a simple way to express an application framework for building computer vision apps and the filter naming. Kind of formally we call them vision data filters, sometimes vision intelligence filters. The idea is that it's an app and it can include a variety of different models. It can also include business logic about the physical world or the objects you care about for your business. It can be customized, and so we built this kind of runtime called Filterbox, and Filterbox is a Dockerized framework that gives you all the basic services you need to create a computer vision app. So think about, like how do I get access to the video feed? How do I get access to the GPU? How do I output data through an MQTT endpoint? It gives you all the basic services that you would need in order to build a generic computer vision app and then allows you to write that app and to insert a model that you've trained, and the model can go through its own lifecycle, so you can kind of think about an application lifecycle and a model lifecycle. The marriage of those becomes this sort of application version that can then be flipped. It's very DevOps we jokingly call it filter ops, by the way sometimes, but the sort of DevOps mindset that really you want to keep software up to date, continuous improvement of the software and, in this world that we're in now, with AI managing those updates.

Speaker 2: 13:57

Not only is it about versioning like, oh, I have more data, so therefore I can have a better model, I have security patches, I have software updates, but there's actually an even bigger change which is I can't predict and I don't think you or any of our listeners here can predict what AI technology will emerge in the next 12 or 18 months and how will we consume the next transformer that changes the entire way we do CNNs and then transformers. Now we want to do something else. Or there's a new model on hugging face, or there's a new model created by an open source or academia. How do we know? We don't know. We have no idea. We don't know what's coming, okay, so the thing we do know is that we're gonna have to change and we're gonna have to keep up to date. We know that and the best model that I've seen for software update infrastructure as containers, right Artifact management, continuous delivery, continuous deployment, having a defined process for updating software, and so the theory we're going on is okay. Ai is going to change.

Speaker 2: 14:58

If you're an enterprise, you're stuck between a rock and a hard place. You know you have to adopt AI, but if you adopt AI, you might adopt the wrong thing and something is going to come out that's going to outload what you decided. You're in a really tough position. So we're offering a way to get started using filters as this application framework, and our promise is that when new AI technology comes out, we will build application frameworks around it so you can continue to just consume Dockerized apps. Right, and you already know how to deploy Dockerized apps. Your IT team already knows how to do this. You're already deploying them all over the place, day in, day out. So let us help you adopt AI of today and of tomorrow using this Dockerized framework and deploy them, and we'll keep that up to date. So, again, your model, your application logic and your security patches can all flow to your environment at the edge or in the cloud over time, and let you really focus on getting those apps deployed, getting the value out of that enterprise data, and not on monkeying with.

Speaker 2: 15:53

You know the various incompatible versions of PyTorch. Let's get you working on. You know the things that matter to your business and you know that also, for us, provides another advantage, which is we can sell to the business, right, and their IT team already knows how to deploy it. And that is a huge advantage as well, because today, if you go in and you're saying, hey, you guys should do computer vision, choose a computer vision platform, you know, please buy ours. You know it's a little bit of a difficult conversation because, like, who's going to use this thing? Now you're going to go sell them an army of professional services, people, and that that's you know. You're in a different business at that point.

Speaker 2: 16:26

So I think, uh, for us, that is really what the filter is all about. It's a new abstraction. It's building on the easy to understand Snapchat, instagram kind of way of thinking about a filter as a computer vision app. And they're customized and you pick the right tool for the job. So the one used for counting sheep is different than the one used to assess solar panel defects. Right, those are two different filters. They're both built on the same application framework. They have the same sort of primitives, the same application lifecycle, but they're apps that are in an app store repository that you can use the right tool for the job.

Speaker 1: 17:03

A wonderful approach. Let's talk about one area that you mentioned that's undergoing a lot of change, a lot of disruption, new thinking, and that's manufacturing across the board. But machine vision and automation and AI robotics has been around in that area since the 80s, as you know, but what isn't being done or what needs to be done better, what's new on the horizon for manufacturing?

Speaker 2: 17:31

Yeah, yeah, yeah. So well, first of all, manufacturing is many, many, many, many different industries unto itself. We like to paint it with a broad brush, but manufacturing cars versus manufacturing, you know, on Etsy, two very different things. You know. I was just in Frankfurt visiting a ceramic tile plant and looking at computer vision solutions that they have in place there and they've been building over many years, and looking at how machine learning can enhance what we're doing. And I think the quality control aspect of it, I think, is one that very clearly comes as an idea people want to do, they want to figure out how they can add. Quality control comes as an idea people want to do.

Speaker 2: 18:12

They want to figure out how they can add quality control. Obviously in manufacturing, the earlier in the process you can identify defects, particularly material defects, before joining two or more objects or layers right. So if you can find there's a defect in one half of the thing before you combine them, it'd be better to get rid of the thing, and those kinds of process improvements can be done there. I think the broad question I think is around democratization of the technology. I also think that in distributed supply chains I like to think of this as software-defined compliance or software-defined quality. So if you imagine a broadly distributed supply chain where you have many pieces coming together across corporate boundaries, across supplier boundaries but also geographic boundaries, and let's say you want to ensure some quality control upstream and also downstream in that process, how do you communicate quality expectations of the different manufacturing processes? And I think this is where the software-defined machine learning-defined rules and you kind of think of it as like hey, I can build a filter that I can ship to my supplier, let them run that to get the exact definition of quality that I would expect and have that not be a verbal or even contractual arrangement but really literally be a software arrangement right, we have the ability to give the quality control tools directly to our suppliers I think can be a fundamental shift in how we manage quality across supply chains. I think it's still early and, you know, I think right now manufacturers are trying to figure out. Kind of the first step I mean the sort of broader changes happening in manufacturing, I think is we're seeing a lot of it come back to. You know, I'm an American. I like to see manufacturing come back to America and I expect that if we do see an expansion of domestic manufacturing in states that you know there's going to be a strong technology push to make that happen, I think that's going to be a reality of labor costs and things like that. So again, I don't think we can eliminate the labor associated with these different processes, but the combination of the two will be really powerful. The problem is that if you go and try to adopt technology straight up and you don't balance it from an ROI perspective and process perspective, it can end up being too expensive and cost prohibitive and you try to sort of over-engineer a manufacturing process, it can be, can be something. So I think you can't really just say, okay, we're going to add AI everywhere in the process. There's a lot of process design that has to go in place. Some of it's physical, some of it has to do with the human processes and then robotics and things like that.

Speaker 2: 20:50

So, for you know, and then just like to bring it home, for us, like we have a number of manufacturing customers, primarily we're working in quality control. That's our primary use case for for machine learning, machine learning and computer vision. We also do inventory. I mean inventory is a big part of manufacturing as well, right, making sure you've got the right supplies and parts in place in time for manufacturing. So those are probably the two key areas that we play. I would like to see us expand more to human process manufacturing.

Speaker 2: 21:20

Quality control. It's a difficult problem, right? So you like, you know hand stitching the baseball, you know looking at the behaviors of the people in the process. I think could be a really great growth area for us, but I think technologically very difficult and also it's hard because of the camera placement and also the number of GPUs required in order to process it at the edge. So those are kind of some of the challenges with the human manufacturing, people manufacturing I don't know what you call that, but like people putting stuff together, quality control. You kind of have to do that a little differently, yeah.

Speaker 1: 21:57

Well, interesting, interesting topic. So much to dive into there. But one of the other things you mentioned on your website is compliance tracking, which I thought was interesting. So many industries are heavily regulated, and for good reasons. I was just thinking the other day I got a couple of prescriptions from my local provider, from the pharmacy, and there were pills missing. I actually took the opportunity to count the number of pills. They provided 30-day prescription and there were like a dozen pills missing and I thought, wow, that's strange. I called a three-letter prescription provider and said you know, there are pills missing. So they researched it and they got the video of the actual pharmacist counting the pills and they're like you're right, there were seven, eight pills missing. And I thought, wow, at first I was impressed that they could actually get the video and they recorded that, but then I thought, well, this might be relevant for our conversation. Why isn't there machine counting of those pills and how many?

Speaker 1: 23:06

other people are missing pills, who would never even think about counting pills. I guess the point is how can technology simplify this process? Is that a use case that you foresee, for example?

Speaker 2: 23:17

I want you to share the details with me offline so I can contact them. Listen, here's the thing when the customer already has the video and this is where you just made the technological leap right we already knew the process was happening, we already were able to audit it by hand, and so now the question is can we make it cost effective for them to add in an AI-driven computer vision check that would you know, certify and verify. You know what they're putting into the bottle and it's great to be able to go back and audit right. That's a you know, if you think about kind of the maturity right, the maturity levels, being able to like have the video and audit it on demand, that's actually a pretty strong capability and that's actually almost a prerequisite right.

Speaker 2: 24:00

Many companies I talked to want to jump to the. You know, wouldn't it be cool if we just had cameras for all the pharmacists? But they haven't done the intermediate step of saying, well, if a customer questioned the count, can we go back and check the tape? Can I instant replay? And actually somebody told me that that what plain sight technologies does is sort of like instant replay for business. I thought it was a pretty good, a pretty good analogy. For what?

Speaker 2: 24:21

we're trying to do. Yeah, so, but in that particular scenario, just imagine there was a filter that was like the pill counter filter, that was running and can catch it in real time before the before the pills went out the door. You wouldn't necessarily replace the pharmacist, right? You want the human making the decision, you want the human doing the task, right. Very, very professional jobs, you want that. But wouldn't it be a great add on tool?

Speaker 2: 24:46

And I think the reason why people wouldn't do it is because they would imagine it to be very cost prohibitive or they would imagine they would need to go hire a team and all this kind of stuff, and so this is what we're trying to do is to lower the bar for just adding one more step of computer vision to these existing processes, where they already have the infrastructure in terms of cameras, they already have the ability to do the recall. Now the question is how do we make it more proactive, and that becomes a perfectly ripe customer opportunity for us. This is what we're looking for with all of our very careful marketing that we're doing to reach out to people. We're looking for exactly those situations where we can just take one step further and increase the maturity dramatically.

Speaker 1: 25:31

Fantastic. Well, lots of problems to solve like that. And talk about the other challenge here, that traditionally tools like yours were in the hands of data scientists and researchers and software engineers, not everyday developers or IT business folks. Talk about the user experience and what has to happen or what you're doing with your, your tools yeah, yeah.

Speaker 2: 25:58

So so the the first thing, which is a little hard for people to wrap their head around, is that you kind of have to imagine the filters as part of an app store and and this is the best metaphor I've come up with for people understand if you think about a repository, you know, if you're familiar with docker hub um, where there's, you know, docker hub, you can go and download all kinds of open source and free apps, but also proprietary apps or internally developed apps. You can have your own docker hub. I used to work at jfrog. We had a product called artifactory. It's very popular for artifact management and uh and in fact we are, you know, we use jfrog, uh, artifactory at, uh, plain sight technologies.

Speaker 2: 26:33

So the idea of artifact management is the first thing, right. And so when you're buying filters from us, what you're really buying is a subscription to get updates to those filters that you can then go and deploy in your environment, and what comes with that is data acquisition, labeling, training, software updates, patches, all of that, including the business logic development. And we do have trusted third parties that can build their own filters on our filter box framework. But for the most part, we're taking pretty, you know, as an early startup, we're taking big responsibility for our customers and we're developing, you know, some configuration and customization for customers. The idea is that it's all reusable and accrues to product value, where we'll have more and more of these filters. So the concept is that because we're creating apps that are reusable and the customers can plug in their own models, train on their own data, we have an incredibly high amount of reusability. We can provide white glove service to customers, really give them the full meal deal, but we limit the scope and we're not doing custom development, integration et cetera.

Speaker 2: 27:42

We have partners that are helpful with that, but what we're really doing is we're shipping you a container that has this app. You have the model that we've helped you source the data and train and you're now able to deploy that out to your edge or cloud environment and the data flows into your data systems and you can manage it from there. And this setup works out pretty well for enterprises because they have the data science or at least the business intelligence we'll say right, those are kind of similar concepts to handle the data. The part they don't have the capability around is the computer vision, and so we wrap that computer vision up as a black box essentially, and ship them updates and they can deploy it and install it and then it just starts flowing data into their systems.

Speaker 2: 28:25

That's kind of like the arrangement, and over time I'm hoping that we will have more and more third-party apps. I think there's a big opportunity for us to create third-party apps and frameworks that other businesses can use and sell and share and really build a profound, you know, kind of capability, especially when they've put in the energy to collect and train models that are, you know, are perhaps specialized or proprietary. I think that's a great community. So we want to get there eventually. In the very short term we're working directly with customers and we're expanding the kinds of use cases and the kind of computer vision we can do and also the data integrations where we can have data flow downstream once the data is processed with the filter.

Speaker 1: 29:19

Fantastic. And how do you suggest businesses, customers, potential customers, get started? Is there a certain playbook or step-by-step approach that you recommend, because this is a new venture for many industries outside of big tech or you know traditionally, you know tech businesses.

Speaker 2: 29:40

Well, I'm glad, yeah, so we're. We're actually in the midst of launching and we're just doing this with our first couple couple companies cohort what we're calling an AI readiness program, and it's not even on the website yet. It's not even on the website yet. But the AI readiness is really about how can a company prepare themselves for AI that provides a good ROI and meets them where they are but also prepares them for the future, and so we've started to develop this. As I was kind of mentioning this maturity model around digital world building, how do we get to this place where our company has a digital map of its operations, facilities, customers and business objects? How do we get the right GPUs, the right hardware infrastructure, the right cameras? How do we start to build up the expertise and infrastructure capabilities? And we're offering this to customers as a way of preparing themselves, because so often we have kind of like your pill example, we have people come to us who don't have the necessary, let's say, prerequisites to get started down the path of computer vision, machine learning. We want to help them. I don't want to sell them something that they can't be successful with. So, if they're ready and they can start to engage with us on a commercial level, great, we will help them get going. More often, though, frankly, people are confused. They know they need an ai strategy. They're under a lot of pressure to get an ai strategy together. They want to prove that they can. They can do it. They want to start small, and the challenge I see is often what happens when people are in that situation is they're not preparing themselves for production, they're doing a proof of concept. They're in a science fair project. That's a dead end, and a lot of money is being made and has been made on science fair projects and proof of concepts that will never see the light of production, and that's actually, for them, is actually a great thing. Everybody's very happy with that outcome. Hey, we built this demo. Let's pray to God they don't ask us to take it to production, because that's going to be a disaster, because it only worked in that context.

Speaker 2: 31:33

I want to help customers get ready. I know it's maybe I don't know crazy, but like I don't want people to adopt AI for no reason, I don't want them to, I want them to do it because it's solving a real business problem. I see AI as a tool, just like any other software technology. It's a tool to solve a problem and we have to understand what the problem is. We have to understand the business case, we have to understand how they will adopt it and how we'll operationalize it. And if they're not ready to do that, then let's put them on a different path. And this is, you know, I guess, what I've always found as a way to help customers be successful, whether it was an open source or cloud adoption or anything else software reliability.

Speaker 2: 32:09

I spend a bunch of time in software reliability. Meeting customers where they are often involves a community. It's education, it's sharing, it's learning, it's trying stuff out, but when it's not, it's selling people stuff they don't need. So that's where we're pushing people now. And if anybody out there wants to learn more about kind of the readiness, it's a no sales pitch, no framework, something we're doing collaboratively. It's still not ready, so we're kind of in the .1 version of it. We're trying to make it better. And one of the cool things is we're collecting a lot of materials from around the market and kind of putting it into a little bit of a library. So we're looking at how we can crowdsource that and build something up over time. So I'm really excited about what we're doing there. I'm probably sharing a little early, cause we're still kind of building it, but it's

Speaker 2: 33:00

a. It's kind of a cool thing, I think, to be able to talk to a customer or a prospect and tell them you know, you know what you're not ready to buy, but let us help you get ready to make a good decision. I will also mention we have a computer vision buyer's guide that we put out, as well as a pdf couple page pdf and it has, you know, really just a list of criteria to think about. You know, it gives you the ability to kind of like stack rank and prioritize different things that you should care about, gives you a framework for thinking about roi. You know, free, freely available on our website. Uh, I can find the link. I think it was supposed to be more prominent than it is, but we have this buyer's's guide, so if you're interested, you can find that. But yeah, we're, you know, fill out a form on our website. We'll help you go through the readiness program. We're very low pressure from a sales perspective, so you know.

Speaker 1: 33:47

Well, great opportunity for someone listening or watching. Thanks for that. And I guess, as a startup, I'm sure you're focused on the here and now and the next months and a year, but what long-term innovations do you see with your futurist hat on? What are you most excited about in the next couple of years?

Speaker 2: 34:06

Well, yeah, I think for me, I am expecting that we're going to see more efficiency of GPUs, or I'm hoping. I think this is one of the biggest bottlenecks in the industry. I think we're almost in a weird way. We've got this legacy of the NVIDIA chip as a graphics card that kind of just became the de facto, and they've done amazing work to build the platform for it. But I think the challenge I see is really about the edge and for the theory that we're going to have a lot of edge running computer vision and other machine learning workloads, it's pretty cost prohibitive the way that it works today, and I think what we need to have is, from an inference not to get in the weeds on this a little bit but, like you know, from an inference, the edge perspective, we need an architecture that will work well. With that. I would like to see something that's really optimized for containers, because I think containers are the bee's knees and we've kind of made a bet on containers and so being able to you know, time slicing is maybe one way of thinking about this, but really it's about load multiple models into memory for the GPUs and get good inference and be able to share across different application workloads and really high utilization of GPUs. This to me, I think, is one of the big areas that needs innovation. It's not innovation that we're necessarily going to develop ourselves, but we're collaborating with our partners at the different chip makers and different operating system companies to try to understand the best path forward, make sure we have a cost effective and scalable approach. When I have a customer who says, hey, I have 500 stores, I want to install 12 cameras at each and run filters on them, and I go, okay, well, let's see how many GPUs is that exactly? And it starts to become problematic, and so we're having to drive that efficiency at the software level. I think it'd be great to have it at the full ecosystem. So that's one area I know. Maybe, again, I'm a very practical guy, so it's like I went to a practical matter.

Speaker 2: 36:09

I think that there's a lot of emphasis right now on centralized, you know, consumer grade artificial intelligence and a lot of futurism around how we're going to change the way we work and the way we live and the way we do everything in our lives, which, you know, there's probably some negative truth to it.

Speaker 2: 36:31

But I'm more interested in highly specialized, highly targeted models that solve very specific problems for businesses and you know you look at the state of technology adoption in enterprises, writ large and, like you know, we're a long way away from this, and what I've seen come out in AI has been mostly consumer grade type stuff, whether it's productivity tools or text generation.

Speaker 2: 36:59

It's a lot of consumer grade stuff and I think the challenge is really in the enterprise grade stuff, like where we're going to bring that to bear and how we're going to sit on top of the large kind of business operating systems like the SAPs and NetSuites of the world. And how are we going to bring something far more sophisticated than a chatbot, something that really drives inventory yield, operational excellence, quality control, into these systems that basically hold the world's supply chain in memory? How are we going to use AI to make that a better way of running the physical world for so much of materials that affect us on a daily basis, whether it's food or pills or or construction materials or or other goods? This is what I'm excited about, and so I think that's where we're going to put our energy and you know, yeah.

Speaker 1: 37:52

Well, so exciting, so many opportunities. Congratulations on the vision and the mission. I can't wait to see your story unfold onwards and upwards. Thanks, kit. Thanks for joining and sharing with the viewers and listeners here.

Speaker 2: 38:05

Thanks, evan, anytime Happy to do it again. Fantastic, have a great four Thanks everyone have a great week.