Unaligned #13: Computer vision with Plainsight.ai CEO Kit Merker

There we go. It's Computer Vision week here on Unaligned. Who are you?

Hey, I'm Kid Rooker here. I'm the CEO at Plain Sight.

Yeah. And I give it away already. You're doing a lot of computer vision. So give me a like 30,000 foot view of what you're trying to do with your business and then we'll dig into it because there's a lot.

There sure is a lot. Yeah. So Plain Sight, we're a business focused software tool for computer vision. I like to say that we sit between the camera and the spreadsheet. So we're really letting you turn a camera into a sensor. We can extract data. You point the camera at something in your business, maybe cattle or inventory or other produce being created, customers, cars in your drive through whatever it is pointing a camera to something in your business. You want to get data out into a spreadsheet or into an ERP system or BI system which are just complicated spreadsheets. And we create these filters that are powered by Computer Vision. A filter like you would use on social media as the concept, but it's a containerized, easy to deploy application that wraps models. We also have a bunch of tools for model training, but that's really the core idea is that it's cameras and spreadsheets. Plain Slide is the software tool for that.

I'm happy to announce that I am now taking sponsors for Unaligned. I'm proud to have AI Top Tools as our first official sponsor. One of the most important things for me running my podcast is staying on top of the industry, knowing all the players and what they can do. I made a complete list of companies in the AI industry on x.com but no one has as comprehensive a library of tools as AI Top Tools. They break down products by use case like productivity, media, chatbots, customer service and more. A lot of companies are looking for the exact right tools for their needs and there's no better place to find it than AI Top Tools.

Yeah, yeah. Now are you using any of the new multimodal models? They're raining out of the sky it seems. I know of 30 different multimodal models which use cameras. But tell me what you what makes you different than the downloading a model off of hugging face and getting it to work with a camera?

Yeah, so. So first of all, one of the cool things is we do work with a different like a whole variety of model formats and we we deploy on the edge which is usually the, the really key things. Computer Vision's solved problems in Some ways it's all the way to the right of the, the hype cycle, according to Gartner, as opposed to a lot of the stuff like you're talking about is coming out, the generative and multimodal and all these kinds of things. So because we're trying to build models or filters really is what we call them, but for which are apps, we install a model into it that's tuned to a customer's environment. And what we do behind the scenes to create, to create those models is we train on customer data and we make that proprietary. They don't need to have any data scientists to do that, which is very cool. And then we use generative AI from these different providers to create synthetic data. So you might have a, maybe you have a use case where like you have a truck that has tires on it and you want to detect if those tires get popped. This is coming through the production line or whatever. We don't have to go out into the parking lot and actually pop tires to train the model. We use generative AI to create lots of different variations and we can do that for anything from endangered fish that we want to make sure aren't getting caught by the trawlers to broken tiles or other defects on our production lines. We can use that to create a data set that's much richer than you would get from just looking at the real world. And then we take that and feed it into our labeling training process and create a deployable model that's very targeted. We put that into our filters, which then does the data extraction. And so really what that lets you do is now you can iterate very rapidly, get a very fine tuned model for your environment without a lot of effort up front or a lot of data collection. Which is I think a really a novel approach that we're combining both worlds of artificial intelligence.

Yeah. A lot of business people really haven't thought about using cameras with AI to do various things. How do you talk to somebody who's in business, has a business, but hasn't really started thinking about you yet? What do they need to know?

What's interesting about that is a lot of them have assumptions about the complexity and the cost. Those are the two, I'd say biggest barriers to sort of everybody has the idea, wouldn't it be cool if I could stick a camera here and I could take a camera on my food truck and then I could just see all my customers in line and see what people are ordering. At the end of the day I could get my sales report People have that idea, but they think the gap from that to implementation, they might look for a point solution in the market. There isn't a big enough food truck camera market for that to be a real product. No one's going to take that product to market. And there isn't. It's for the enterprise, right? It's for the big businesses that have a large scale operation where it's worth it for them to do it. So really the way that I try to approach this is, is about how do we make it simple and small and easy. And part of that is because instead of us saying oh, we're a computer vision company that focuses on agriculture or manufacturing, we say no, we're focused on a particular point in the software stack, right? And so I kind of use metaphors like becoming the Twilio, right Of that just doing one thing super well. Again camera to spreadsheet, just do that, make it so the software updates are easy. So that's one thing is we really lowered the cost in terms of what does it take to solve a single point problem without having to be a full kind of like venture backed company just trying to solve that one narrow problem. I think the second part in terms of complexity is today I see a lot of tools for AI, data scientists, ML engineers, model creators that are very sophisticated, but that audience is, is limited audience and the Joe business person who's just trying to keep the lights on or whatever and they think, oh, it'd be great to adopt AI. Probably the first thing they think is I need to hire somebody to do this. And because we've made this about the model training process really hides a lot of the complexity. Again partially because we're doing sort of AI 1.0, so to speak, computer vision. And also because we have those generative AI tools behind the scenes, it really reduces that barrier to entry from a data science perspective. And so my goal is that a normal IT person can do it. So what do we do? We use Docker and Kubernetes and we made it so it runs on the edge, it's easy to install and deploy it, industry standard stuff. And so now your IT person becomes your AI, right? Hey, I want to start using AI to track my. I've got this big pile of sand and I want to know if the sand pile is getting bigger or smaller, how am I going to do point a camera at it? My IT team can manage it, we get it trained and it sticks into your ERP system. So now inventory is up to date, suddenly the business is Running better. And I can say, hey guys, we're using AI to, to solve a real honest to goodness problem. And that doesn't, that's not, that's not the. Just, it's not just for the big boys. Right. Like anybody can take that and, and imagine that working their business. Yeah.

If you use Chat GPT's new vision feature, you have to take a picture, it takes a few seconds to give you an answer and then you talk to it. Your systems are aimed at counting a lot of things going by really fast.

Yeah. You saw our.

Tell me a little bit about how your idea or your implementation works differently than a Chat GPT vision model or a Google lens kind of thing.

Yeah. And I think the thing that triggered us getting in touch was the viral sheep counting video that we have of us counting sheep in the field. You're right. We often are trying to get data out of the screen where businesses are not trying to write a haiku about or just describe what's going on in kind of a predictive or creative kind of way. They don't really want hallucination. So if again, thinking about the camera, our metaphor of a camera is a sensor, a filter is saying, what do I pay attention to? And all this data, I don't want to just take the video stream and stick it in my spreadsheet. That doesn't work. I need to filter it down, I need to put it there. So what we care a lot about is accuracy. We care a lot about performance and we care about doing one thing really well. And so in that case, like if we're counting, counting sheep, we need to be able to identify a sheep with very high accuracy. And we may be able to do that for this particular pasture. Right. And fine tune it to these sheep in this pasture. And we gotta worry about variations in weather and other things like that. So we can do generative AI to say, oh, let's do it at sunset, let's do it in with snow on the ground, let's do it in these different variations. The other part about this is really interesting is a lot of times we don't have an Internet connection and we don't want to stream the data, we don't want to stick it in the cloud. And you'd be hard pressed to find the large models that aren't running on extremely expensive Nvidia hardware or in the cloud somewhere. So I think this is one of the other really big advantages is you can take a relatively inexpensive piece of hardware, you can run a high accuracy model at the edge and the total data you're streaming is reduced dramatically. You went from camera, which we connect to over IP to. Now some event data is coming out in JSON blobs. Right. You know, very small, very good compression. It also gives us a point of control for privacy.

500 more widgets went by on the production line or something like that.

Exactly. You're just sending the events. We saw it pass, pass the checks, whatever it is that data streams into downstream BI analytics or data lakes, whatever other tools people want to use.

Yeah, businesses are. There's a spectrum of AI ness in business, right. There's highly AI centric businesses and then there's people who haven't even learned how to spell AI yet. What are you seeing happen in the business world? What are the needs that are going on? And give me a little insight. Insight of how.

Yeah.

How business people are looking at these cameras and looking at this new technology and maybe even starting to think about, oh, I'm going to need an orchestra of AIs to do a variety of tasks. Counting sheep is one of them. But there's other things that AI is going to need. Be needing to do in a business. Right. So tell me where people are.

And I think everybody in 2024 needs an AI story. They either have a story that's a fable or they've got a story that's nonfiction. Okay. And it really doesn't matter whether you're like, I was just giving an example of a food truck that needs an AI story. Right. They click it doesn't really matter the scale or size of business. Everybody, I think in business is asking themselves the question because whether it's writing marketing copy, I. E. Spam, having AI do that for you, or whether it's managing the inventory or looking for, if your employees are wearing their pieces of flair, any of these kinds of things, you're paying a person to pay attention to it and do it. Everybody needs a story for it. So that I think is universally true. As I mentioned before, cost and complexity are a big problem for this mid market. I think there's a very clear mid market emergence now of a recognition because it becomes so much more mainstream. We're productizing it, almost consumerize it. I think of it almost like the prosumer kind of category of product. Right. Which is really drives that, that entry point, which is better for everybody because frankly a lot of the enterprises, I do talk to quite a few of the larger companies that have AI projects. They all have battle scars and PTSDs from consulting projects which were science fair projects that were promised. But if you go through this whole engineering cycle just to prove whether we can do it or not, then you wave a magic wand and you say, do we get the insights we wanted? And then, oh, wait, privacy, security updates, rollout, right. Training, integration, all these other things that didn't do in the science fair project. So I think that a big part of what I'm trying to do with educating people about AI is about accessibility. It's, look, we're making it easy. We're making it so that you can start small. And that doesn't mean start small in a lab. It means start small with a single kind of point. But because it's a part of the IT stack, it's like a well defined software stack infrastructure. Right, the infrastructure for building computer vision as a capability. And now you can imagine that scaling up and down, applying DevOps principles, really good software management principles into that. So we can do security patching and all this other stuff. And so when I look at the businesses, they want the AI story and they've been hurt in the past and I think now there's a new hope. I would say, right, we can now adopt this in a way that's productive. I believe that the companies that are going to survive the AI hype cycle we're going through right now are the ones that solve real problems. Doesn't necessarily have to be a business problem, a person, either product or an urgent public problem or a personal problem. We, we have to solve something people can actually adopt. I see a lot of promises out there for sure, and it's exciting. But the promises only get me so far is I think the feeling on the street.

So I've been blessed to go to a lot of businesses that manufacture a variety of things or make, for instance, pistachio farmer, right, gets loads and loads of new pistachios in every few minutes, has machines to sort those by size and quality. Is that a place where a camera could be useful and can it see that kind of small difference in size between a bigger pistachio and a smaller.

Pistachio kind of thing? Yeah. So I would say a couple things. One is some problems are at a scale that cameras don't make the most sense. Vision is still a tool for the job, and there's some cases where it makes a lot more sense than others. The second thing is for a highly specialized operation. So in that, in a, I don't know, pistachio sorting, I don't know the business, I would assume that there's probably some shakers and some holes that things are falling through, or there's probably some mechanical processes it's going through. It's not literally like a camera and then a hand comes by and pulls out bounce. You might have something like that you've seen where it's like foreign object detection and you have an Airstream or something to remove things. But if you're really doing one thing and you're really able to hyper specialize, right, you build a big piece of equipment, you, you funnel it on there, you run that process and everybody's happy. I think of the computer vision world that we're seeing now is more like 3D printing. It's, it's more kind of batch bespoke. We do this once in a while. It's, it's confusing. We're paying people to stand around and fill out spreadsheets. That's really what I look for is, hey, if you have somebody who's going around counting, I, I'll give you an example. We work with a, a, a company that does cattle. They sell cattle at an auction and they have the ranch hands that are counting the cattle as the auction's happening, auctioneers dropping the gavel and talking real fast and they're counting all the cattle. And then what's interesting is they have to have multiple people doing it because county cattle's tough and you want to check the counts and then the auctioneer every night after dinner watches the video and makes sure the counts are right as well. So they're spending a lot of energy doing what seems like a simple task. But the reality of it is it's not, it's a mind numbingly hard thing to do. And you have to be right because these are expensive commodities. And so that's the kind of thing where I see it applying much more clearly. I will also say though in these production processes that you might go buy a specialized piece of equipment that has computer vision built in. And what we're seeing now is as computer vision is becoming more available, you end up now with these silos, right? Okay, I bought this one solution for security about this other solution for slip and fall. I bought this other solution for customer vip, another one for threat theft detection. And now I've got this, I've got an IT mess is basically the situation you come in. So where we can come into that and say, look, you got one piece of infrastructure, one camera, one set of compute, deploy a Kubernetes cluster that runs all these filters and processes the data and gives that data out to the different teams. So the security guy or gal gets notified if there's a security issue, and the customer care team gets notified of this customer care issue. And now that's a software solution. Right. We're really abstracting away. It's not a smart camera. It's like we have this camera infrastructure, this computer infrastructure, and then software on top of it that's giving us this AI capability to go solve a business problem. And that architecture, I think, really takes what you described as this very specialized process. And now you can start to think of this with a little bit more imagination, which could lead to, hey, maybe there's more opportunity to create differentiated products. Where before we had to have a single production line. Now we could have, oh, there's different flavors or different saltiness levels of the pistachios. Right. And that's where you might have a more more dynamic approach. Because now we have the ability to use software as a tool. It's just so much easier to update than these physical processes.

Yeah. If you go to Louisville Bat company, Right, they make a lot of bats and they burn logos into their bats. Computer could your style, computer vision look at a bat and say, oh, that was made properly.

Yeah, that do some quality control. Fairly trivial quality control issue. We can do with computer vision. A great example of something where especially if perhaps you had in the past you had artisan people who really took pride in their work and looked at it, and you could really trust them because you're doing small batches now, you've got a more complex workforce that maybe doesn't have the. The same kind of artisan attention to detail.

Right.

And so now you want to create consistency across different manufacturing locations. What's interesting on this topic, related but different, is labeling of food. Labeling of food is a very important issue. And it. It has to do with regulations that are changing constantly. And you think about international packaged food companies. And I heard a story, I don't know, I won't say what large company it was, but they accidentally mixed up the printing between Lebanon and Israel and shipped food with the wrong labels to the wrong countries because the person who was doing the labeling couldn't read the languages. They had to recall the whole thing. That's the story I heard. I don't know. It's interesting. And so you think about that. So your point about Louisville Slugger. Okay, well, maybe they want to do the same kind of thing, make a bat in another language. Right. And now you're equality, control across different languages. I can't even read it. I'm producing it in a country made with lower costs. This global supply chain issue, now all of a sudden our software can differentiate between these different, these different products and now we can get quality control, consistency and we can report it back to home based. I think this is really the important part is the person who cares about that sitting in an erp, she's looking at her ERP system and trusting the data that's flowing through all these different games of telephone. What if you just stick a camera right at the source? Right. That's really the key.

If I was working for a Hertz rental car, I'd probably not like to see lines of people waiting to talk to an agent. So could we build a camera system that looks at like the lobby people would be standing in line and figure out, okay, how long did that customer number one stand in line to get to the counter and report that and give everybody a look at. Oh, at one in the morning we had a lot of people standing in line because we didn't have enough staff on hand or whatever. So let's fix that problem.

Great use case.

Is that the kind of thing that you guys can do?

Well, yeah. And you think about that scenario. So waiting in line and being able to do like your workforce planning or even being able to, hey, send an alert. I heard a similar one for hey, tell the manager to go to start making coffee or go notify people, go call people in if there's a huge heard. But if you think about the car return scenario, identifying the cars by their bumper stickers or excuse me, their might be bumper stickers or license plates. Scratch detection. Right. So you can think about that. You look at the tires, there's a lot of work that goes into that. Walk around on the cars and deciding on that. There's a lot of different use cases there that are interesting. And again for us, we think of those as filters. And it's a simple way to think about the problem, which is I build a filter that's really looking for one thing and paying attention to it and then rendering that data or outputting that data. So, and then I can string them together. So I might have filters for counting people like you said, and then there might be some, some overnight processing we do later with the cloud where we can now go and do some, some. Okay, are we seeing any patterns from that data in terms of the same employee that's upsetting all the customers and we have the conversation with them or we're doing facial detection to get sentiment analysis. Right. Are they, are These people happy? Are they not happy? What are we doing? Trying to dig in a bit deeper and understand it and build some prediction around it. We don't do that part. Our goal is to give them to give you really solid data so that you can do that analysis downstream. And I think this is where starting to use different AI tools for the different parts of the job. Right. The generative stuff for us, great input for training, training it at the edge or running at the edge for inference about what's going on and paying attention to what's happening in the world. Sticking that into a data lake, then using data analytics, AI tools, predictive analytics and other kinds of analysis and categorization. And now that gives you a much richer view versus what I think a lot of people's conception is. The sort of layperson conception of AI is. It's, it's jack GPT and there's just so many different variations to what, what can be used again in concert to give people a technology solution. Picking the right tool for the job.

Yeah, if I owned a grocery store, I might want to be using computer vision on dwell times. How long are people hanging out in the cookie aisle and how can we increase that or increase sales?

People who are standing or see an empty shelf and generate a pick list for the inventory team to come and refill. Like there's there when you start peeling back. And this is what's so fun about this. You're, you should be on our, our idea generation team. Right. It's like we, we're always thinking about. So we have a channel on our Slack called Filter Ideas. And so anytime anybody has a filter, a lot of times for me it's two in the morning, I come up with an idea, we post it there and we, we're always talking about different ways to build filters for customers. And we get, we hear a lot from them. They have ideas that they have as well. But yeah, the opportunities are, are limitless. I think the, the biggest thing I found is that when you go into a business, you have a lot of ideas about where we can apply AI. Right. And customers have lots of ideas. The, the difference between one that's really good and one that's less good is if they're already doing something, they already have a place to put the data. They're already trying to solve this problem today and they're doing it manually. It's usually a good indication that it's a priority to do it and it's not going to be something where the technology is going to the Tail that wags the dog, right? It's going to be like solving a real problem. If you come into an idea and say, oh, hey, wouldn't it be great if you could tell this thing and then you could start sending coupons from the blah, blah, blah. You start coming up this complicated story about how the business like theoretically could be run. It doesn't really work. And, and I've seen too many times these types of solutions where the impact is not what we expect, frankly. Checkout. Dragging my grocery self checkout to me is the bane of my own personal existence. My wife insists on using self checkout. I think it's like a marriage bonding ritual for us to move the stuff. I get so frustrated with these machines because the latency, the accuracy. Obviously I'm over 21 when I'm buying my bottle of wine. I got to check my ID now. It's just, it's frustrating. And I think that a lot of what two things, one is I think ahead of the technology curve, right? A lot of this got implemented. I've heard a lot of them are getting torn out now, which is funny. The second thing though is I think they built, my friend built them for.

Walmart and they decided not to go that way because customer one, they were seeing increase in theft. Theft two, customers didn't like it. Right. They wanted to talk to a human being. And that might be the one time where they get a little luxury in their life. They have somebody else take care of their checking.

It's a great job. It's a great, definitely a great job for a lot of people. I have actually two siblings that are both have worked in the grocery business before and started as checkers and did inventory, stuff like that. So I appreciate it as I believe people should have jobs. And this is where I think the difficulty is when technology and a great idea about what technology can do collides with the real world and society and what can be done. One, one thing I wanted to share with you on that topic is wildfires. I don't know if this is an issue you care about, but one I care about a lot living in the Pacific Northwest is we have some really terrible wildfires here and to the point where actually there was a local winery that won't release their 2021 vintage because the smoke was so bad in this area. And we just recently launched a new website called wildfirewatch.org and the idea was pretty straightforward. We took all the public cameras in Washington state. We have 1547 cameras that are run by the Washington State Department of Transportation. They're all publicly available on the Internet. So what do we do? We built a wildfire detection filter and we roll it out so that every five minutes we are scanning it. Now we're still early days. We have a lot of false alarms. We have issues with headlights, we have issues with solar or lens flares. We have issues with sunsets. We have smoke that is setting it off sometimes. What we're doing is we're working with the community, the fire community, Department of Natural Resources and others to get better data accuracy. And so we're doing the same exact techniques we're using for generative AI to generate these images, do the tagging improvement. But right now, you know, you can go on there and see like a map of where we think there might be fire that's worth checking out. So you can filter down your view and then you can see the live feeds overlaid with the fire detection, which is pretty cool. And so we're looking for people who want to contribute, additional cameras or any researchers that want to get involved in this. We don't want, I don't want to make money off wildfires, frankly. So I'd like to have a free resource out there for the world. And yeah, it's pretty cool. So there you go.

No, that's cool. How do you charge businesses? If I'm a grocery store and I want to watch my checkout lanes and for dwell times or for line length or something like that, how would I be thinking of budgeting for that?

Yeah, the way that we charge, I think is pretty, I think it's pretty fair. I'm a big believer in being fair. So the first thing is that we sell a product. We don't sell services. We sell a product. We make sure that there's a lot of care that goes into getting it rolled out and installed and deployed and everything else. So we very simple, get a subscription, annual subscription. We charge based on the number of filters. We give you up to 10 camera feeds are included in the filter and then we have a very small add on feed for devices past that. So basically you have the. Each, each camera gets cheaper the more you buy. So it scales very nicely. We include in that, by the way, the software development. So if there's customizations, et cetera, it's included. The training cost is included, the data acquisition and generative AI stuff all included, all the labels included. And the reason is because I think it doesn't make sense to try to nickel and dime customers to do this. We have a continuous improvement process we run. So when you become a customer, we're going to quickly show you that it's possible. It doesn't take very much to show it's possible. If you sign on with us, you're going to get software updates, security patches, all of that on an annual basis. And the other interesting thing that surprises people is that we're not really a SaaS platform. Everybody at enterprise Software is a SaaS platform. We're really a downloadable, installable piece of software. You can run it in cloud, you can run at the edge. It's Docker, it's Kubernetes. Very simple to deploy and run. And so that, that kind of really builds, I think, a very cool business model.

Oh, that's interesting. Could it run on a cheap computer, like a Raspberry PI kind of thing?

That's right.

Or do you need a full Nvidia card around on the computer also?

Great question. So one of, one of the things we're working on right now is model optimization for different hardware specs. And so what we do is we have a hardware guy who basically has a lab where he's got different hardware specifications that we can benchmark it. So what we'll do is depending on the intensity of the compute requirement. So in some case you might have a whole, you know, big box store with lots of cameras, lots of processing. What's cool in our world is you can just stick one big box there and you can run a Kubernetes cluster on it and you can just process all the video you want and you don't have to worry about. In other cases you want to have a smaller device or as small as a chip. And it really has to do with computing complexity, model accuracy, how much data you're processing. But it scales up and down pretty nicely. And depending on the use case, it could be different. Different levels of hardware. We try to make sure that's price. We don't sell the hardware, by the way. We actually do a hardware starter kit we mail out to our customers. Includes a couple cameras, Jetson Orin Nvidia device and some extra, you know, goodies, some tripods and things like that. And what's cool is we send that out to customers when they sign a little getting started agreement. If they convert into a paid customer, they get to keep it, and if not, they just mail it back. So it's a very simple way for us to get past the hardware issue and then we help them scale from there, figure out the right hardware strategy.

Yeah, you, you mentioned inventory a Little bit. If we were doing a inventory for a Walmart, that's a lot of shelves to look at. Right. Do, do you start thinking about putting a camera on each shelf or do you think about moving that around in a, in a robot or a drone or that kind of thing? How do you think about doing inventory even for a small store, like a little liquor store or something like that?

Yeah, yeah. This is one of those sort of, it depends situations. If we can get overhead cameras or existing cameras, we usually start there. We usually start with what cameras do you have? Let's see how good we can get it with what you have. And then if there's some blind spots or we're not getting the accuracy we want, we can make some recommendations around that. I, I, the drone use case is really interesting. More applicable, I would say in orchards or in, in, in groves and those kinds of agricultural environments. We do have some, some drone missions that we support, which is a very cool thing where you can go and fly drones and upload the videos later. There's a variety of different techniques. The surprisingly, a lot of grocery stores already have eye level cameras that you don't even know about that are in the shelving. And so if we can get hooked into their, their sort of camera management system and scale that work. Pretty much everybody's got security cameras and they're underutilized. And so one of the things you can do is a starting point is, hey, let's take a look at what security cameras you have. And then let's like hook into them through RTSP and then let's see if we can start adding some filters to it and get some value out of it. Again, it's really, it's so specific to the space, but if there's a big enough need where they see the business value to take action and to track the data better, and you're paying somebody to do it already, that maybe is inaccurate, expensive, or maybe it goes on vacation. Right? We can replace that. But if you say, oh, I'm going to go spend a hundred thousand dollars to go deploy cameras out to the store and is it really worth it? We will quickly tell you that, hey, don't waste your money. Let's be smart about this and make sure that it's worth it to me, that's a really big part of this. I don't want to sell technology to people who aren't going to see benefit. I will say though, that from a pricing perspective we usually are cheaper. An annual subscription is usually cheaper than a proof of concept with any sort of other company. So our total cost for a year is like about the same price as doing a PSC with a big company which usually pencils out pretty well. Yeah.

Yeah, that's cool. What else do people need to know about this new computer vision world? It's old, but it's getting new. I've seen computer vision for a decade, but now it's getting smarter and smarter and the systems that it's getting integrated into are getting smarter and smarter.

Yeah. So I think what people should know is that if you have an idea for it, we'd be happy to talk about it. And no obligation. I'm always very happy to listen to, to students, to researchers, to public officials, to business executives to it people that are either have an idea but aren't sure where to get started or have a project where they feel like maybe they're either overpaying or it's too complicated, too risky. I'm always happy to chat about it and I try to spend most of my energy like spending with the community and out talking to people so I can learn more and more about what's going on. That'd be the number one thing is don't be afraid to ask. And we would be happy to chat about it. And we also on our website we have a couple resources I think are interesting. One is we actually have a listings for filters. So if you go on plainsight AI you can click Explore filters and you can see some of the filters that are out there and that sometimes is inspiring for people and they say oh this is something I would want to do. Of course it's not exhaustive of everything we could do. So we also have a box there so you have an idea, you can submit an idea and we'll reach out and chat about it. We also launched something called Filterbox which is our dockerized runtime and we made it available as a demo right now. So we have a face blur. We have what we call Seymour Pong where it takes your hands and lets you play Pong on the screen. Seymour is our mascot because we like to say Seymour Business. Our elephant mascot, Seymour. And then there's an object detector. I think it detects 80 common objects. You can hold them up and it'll put boxes and show them so you can show kind of a simple way to get started with with filters yourself and see some pre built models. We're trying to make that more easier to do, more accessible, run more platforms over time. Would love feedback from anybody who tries it out.

Very Cool. Where do we learn more about before we get into that? But give me a go into your business a little bit. How many people work there? How were you funded? Give us some fundamentals behind this.

Oh, yeah, yeah. So it's an interesting. It's an interesting story. I'll give you the short version, but I joined recently and we've actually created a brand new company. In this economy, of course, companies have gone through ups and downs and Plain Sight Technologies is no different. So completely rebooted it recently. In terms of funding, we're privately. Privately funded by an investor in Chicago who's done quite well and is really interested in the space and given us a really strong mandate to build, I think a very unique and scalable business around this concept of taking computer vision to market for businesses. And yeah, roughly 20 employees, mostly engineering, some sales as well and some delivery folks. But it's a pretty lean team, which is. I really enjoy that. I like running small companies where we can wear a lot of hats, get a lot done. I've been very focused on the product and how we take the product to market and then thinking a lot about building something pretty unique from a culture perspective and from execution perspective. So fun. And yeah, hopefully we're going to be. Hopefully we'll be big someday. I don't know, we'll grow up. Yeah.

A lot of computer vision jobs out there to do, so.

That's right.

Yeah. Until they're all done. Where do we learn more about you?

Plainsight AI and the AI is right in Plain Sight. So plai and S I g H T AI is the best place to go and you can see all about us and all of our filters and filter box and how it works and get in touch that way.

And see the sheep counter.

And see the sheep counter right there on the homepage. That's right.

Thank you so much. It was great. This is a new world. It's really interesting to see what businesses are doing with cameras as cameras continue coming down in cost and the capabilities continue going up. So exciting times. Thanks for being on the show.

Thanks so much for having me. Great to meet you and I hope we get to do it again someday. Yeah, definitely.

Keep in touch. And maybe you got some new robots coming with this kind of technology or whatever. Seems like that's the other theme of the day. I just visit a landscaping business that's building robots and they're using computer vision to find the edge of the lawn and stuff like that.

That.

So there's all sorts of things to do with computer vision, which is a lot of fun.

Amazing. Yeah, very cool.