THE HARDTECH PODCAST
ClearObject - Four Tenths of a Second to Save a Life
Guests: Derek Bleyle, VP of Product, ClearObject
Tim Hagan, Principal Data Scientist, ClearObject
Hosts: Grant Chapman, CEO, Glassboard
Elijah Clarke, Hardware Engineer, Glassboard (guest host)
================================================================================
NOTE: This is a machine-generated transcript, lightly cleaned for proper nouns
(ClearObject, Glassboard, Wi-Fi HaLow, guest names). Speaker attribution was not
reliable in the source file, so lines are timestamped but not labeled by speaker.
Verify any quote against the audio before publishing it.
================================================================================
[00:00] Welcome to the podcast.
[00:08] Everyone, welcome back to the Hardtech Podcast. I have not the usual suspect with me. Oh, Aisha, thanks for joining. Hey, how's it going? So we are here with some fun friends from ClearObject.
[00:18] Derek and Tim and our team have been working together on some really neat tech. Part of the reason we had them on today is the journey that they've been on from a software centric product and going through the crawl, walk, run of hardware. I think it'll be really good for our listeners to hear. This is truly how you get an idea out, get it tested, and then start getting user feedback before you hit volume production.
[00:39] So that's the story that I think school for the listeners, but I think for everyone that is building a hardware product today, it almost always has a software feature or focus in it. And you guys can finally tell that narrative because we haven't had that many software folks on. So thanks for joining us.
[00:53] Thank you for having us.
[00:54] Thanks. Thank you. So Derek, I'm going to pick you first. All right. Derek Bleyle, senior VP of product and operations have ClearObject. So got the responsibility of the engineering and delivery across the organization. Condolences. Yes.
[01:12] Excited to talk about you know, one of our you know, I'd say probably the project that we're all the most proud of at the moment today here. So looking forward to the conversation. Awesome. Tim. Yeah. So Tim Hagan, principal data scientist at ClearObject. Yeah, that I kind of split that in two parts principal and data scientists.
[01:31] So, principal, I kind of run a lot of our slide opportunities and a lot of our vision stuff. And then data science responsible for data science team.
[01:40] Yeah. Like Derek said, really excited to talk about the product we've been building alongside you guys. Awesome. And then you're you've been a guest host before. But what is what is your role been playing with these guys on this cool product.
[01:50] Yeah. So Elijah Clarke I'm a hardware engineer here at Glassboard, and I was brought onto the project with them as a hardware designer. So designed all the circuit boards in the product and then also just worked on the larger side of the project, doing a lot of experimenting, playing around with different technologies and helping them to achieve the goal here.
[02:09] So let's start with ClearObject. What is ClearObject? You guys are in software, which is really broad word, you know, what do you guys do as a as a core business. How does this fit into that. The product that we're going to talking about.
[02:22] Yeah. So I've been with ClearObject eight years. So I'll talk a little bit about the journey we've been on, because it's changed a lot in those eight years about what we have done.
[02:32] So we started as an IoT technology company, meaning, you know, eight years ago, everybody was wanting to say, I do have hardware. Now I've built it. How do I connect it? How do I make use of the data coming off of that software? So started with a lot of connecting to those products. We weren't we weren't developing them.
[02:52] Right? We were more of a we were helping them with their products, build the digital solution to that in the software side of that solution, from that, you know, we returned as an organization, continued to focus a lot on not just what sensors, but how could probably the most powerful sensor out there as a camera, right. So it's multimodal, what you can extract from that.
[03:17] So we really started to lean into the space around vision AI. So extracting information from cameras, making real time intelligence out of that as a connected product, turning that into whether that's insights for manufacturing or food and beverage retail, all of the above. So that's really been our focus over the last I'd say 4 or 5 years. And this broad project that we're going to talk about today started about a year and a half ago, specifically looking at how can we use vision, a technology, to solve a problem specific to it had a lot of similarities of what manufacturing has.
[03:58] So manufacturing you need real time intelligence, meaning, you know, I don't have any latency talking, you know, 30 frames per second type of inferencing speed. All because I need to be able to understand what's happening, make a decision, infer, communicate that
[04:17] to something to take action in real time. Also, you know, if the
[04:23] network goes, there's no Wi-Fi connection.
[04:27] Manufacturing facilities still need to manage. So we'll talk a lot about the the work zone solution here today. But I think it ultimately combined a lot of our specialties, from connected devices to software to how we use the connect, the data coming off of connected devices to
[04:43] provide intelligence. Yeah. Well, I think it's always fascinating to see the evolution of a tech like dispersion in the ecosystem.
[04:51] Right. When we all got started in connectivity in the 2010s. Right. Like the 2010s, we all thought it was going to be industry 4.0, because B2B is usually how most technologies break out and find the first fit. And we thought industry and manufacturing and connectivity and it's big business was going to take off. And we were all wrong because everyone got a smart door lock and a smart thermostat and smart toaster first.
[05:13] And then now we're starting to see industry pick it up and medicine is now going into it. But the data that is coming out of each one of these sensors keeps getting more and more rich is kind of like the narrative. I keep seeing the industry. It's a very simple thermostat or on off switch. Is my door locked or not locked?
[05:30] And all these very binary or individual number metrics that I'm sure you guys were building on of iOS apps and dashboards and databases to collect graph over time, and alert when number went above or below a certain number. And over time these started to get more and more integrated and connected with multiple sensors connecting to make if then statements right or automations happen.
[05:51] And then in the industrial automation world, speed started to matter. And I think this is where you guys got some of your really cool first applications from hearing about your history is you started to be able to take really fast action on a fast moving assembly line that isn't a mechanical operator anymore, right? A lot of times there's mechanical light beam, brake sensors and things like that.
[06:09] That would trigger a PLC, but you guys are adding way more intelligence for some of your clients behind the scenes, stealing what you took from IoT and putting it into a fast paced factor that has to run locally, not a large distributed network.
[06:21] That's exactly it. And I use the example of, you know, this is the simplest example of what is the type of intelligence today.
[06:29] You might have somebody on the line projecting, rejecting a component because they see a defect. Well, how can I augment that? I can automatically do that. But when I do that automatically with an AI system, I can know not only there was a defect where the defect, where was it, where it was the timeliness of that. And so I can track that information and start to do real time process control, plus tie that back to variables upstream to say, okay, what are the actual variables impacting that
[06:56] that are causing and find the root cause, be able to correct those.
[07:00] And if you put it into a big enough database with disparate enough data, you can do the really neat thing that we're all learning how to do, which is find the needle in the haystack. Hey, we see that the upper right hand corner is get bent in this part when it's really hot outside. And we haven't serviced this one, you know, in a while on one of the presses.
[07:19] And you can put these disparate pieces of data together if it all is in one place. And with modern compute and inference and AI and other algorithms, we're starting to find needles and haystack in this industrial in medicine and things like that. But you guys have taken large subset of data to a whole new level with the product we're talking about.
[07:36] So we moved on to vision of what you guys are building in this. And can you guys tell us what the product is so I can sink my fangs in and talk about the neat tech you guys
[07:44] build? Yeah, yeah. So I'll talk about it. We call it Clear Work Zone. But ultimately from the simplest perspective, it is built for field crews that are out working day to day doing operations along the highway.
[07:59] So we're all familiar. We're all driving every day. We see the cones on the side of the road. You know, you might, may or may not pay attention and see that crew there. Tim and I can both tell you from being out in the field, how scary that really is to be, alongside traffic flowing that supposedly 45 miles an hour, right?
[08:18] We all know that's not the case. And so we were approached by a client and what they asked was, hey, we've seen what you guys could do with vision and other use cases. They they were forward thinking New York State Thruway Authority. They've been great partners, really, really leaders in this space to say we want to do something different to protect our workers.
[08:42] So we did a design thinking workshop with them. You go back to the thermostat, funny enough, right? We always started those workshops with tell me a new way to design a thermostat. Then we'd rephrase the question of tell me a better way to control the temperature of the room, right? So taking that back, we said, hey, what is the what is the best way that we can protect a work zone for those workers?
[09:05] And when we did that, we looked at other, other, other people who have tried this in the past. What were the limitations of what they've done? And really the feedback that we heard from workers is, I want something that is simple to set up. I don't want to spend any more time in the works on than I need to, because I got to set up the safety barrier.
[09:21] It has to be local. We work in areas where there's no cell connectivity, there's nothing else. Right. Which we'll come back to that when we talk about the wearable technology and why we selected those pieces and working with you all.
[09:36] Excuse me and be reliable. So ultimately, what we developed is a vision solution that gets put on work trucks and monitors those zones. So we could define protective zones to think of an imaginary boundary that's protecting that zone. And anytime a vehicle intrudes in that area, we in real time send notifications by our wearable, which is where we worked with you all to help develop that capability, where the workers received then a multimodal type of notification.
[10:07] So it's got haptic, visual, inaudible, knowing that they might be performing many different jobs. And we need something that no matter what they are doing, there's a form factor that can try to get their attention, given them every second and or what we say every millisecond
[10:22] of time that they can to get to a safe area. Yeah. Well, I think it's it's so neat what you guys have been able to do with the UX of the vision system.
[10:30] So for everyone listening, I, we all have bloodstream knowing exactly how this thing works. So when they say on the back of a truck, you guys have your camera system that is facing backwards, looking into traffic that's coming into the work zone. And you can use on a like a tablet, see that screen. And the operator can just draw.
[10:45] These lanes are open. These lanes are my, you know, orange caution area. And these lanes are completely closed alert if someone's going, you know, fast in the orange lanes, alert of someone's even getting close to coming into the out of bounds lanes. And that is the vision system that we've been talking about. How cool it is, how fast you guys can process that data.
[11:04] Because every millisecond here matters. You guys then send that down to your computer that does that locally. That alerts out you have some alerting on the truck. I think this was like the very first version of what you guys built is it wasn't actually on the user yet. You had an alarm system that could set off an alarm.
[11:19] We had danger. Will Robinson. Something bad is happening. And you realize that sometimes these workers are jack hammering into the concrete. Sometimes they've got their on a piece of machinery or they're too far away from the work truck. These work zones
[11:32] aren't tiny. How long is your lungs work? So we walked to the end of one once and made it 1300 feet.
[11:41] Couldn't see the trucks anymore. It was close, you know, the two miles back from where we started. So two miles back to, you know, another quarter of a mile
[11:50] forward. So yeah, you're talking two and a half mile closures that could be even larger than that. Right. Which means sound is out. Right. Just an alarm isn't going to get everyone that you're trying to capture here or cover up.
[12:00] And and you had to go reach them where they're at. And this is part of the design thinking how do we alert the user. Not only how do we detect, how do we even reach them in all the weird ways that they're working? Because your your use case user is so different than everyone else in everyday life, right?
[12:16] It's hot outside, it's bright, it's loud. It's all these things that aren't normal for even an industrial zone sometimes. So you guys have to tackle that. And that was a huge amount of fun. So how did you guys get started in the journey? Right. So we're design thinking. We're fever dreaming. Really cool ways to make people safe. We detect fast cars or cars going the wrong way.
[12:36] I mean let's start there. What are all the things you guys tend to detect? Like what are what are the modalities you guys can pick up on on call it drivers being bad behavior.
[12:45] Really we're just tracking the cars. So we run a couple models as the system running and we're looking for cars, people, cones, lanes, all that good type of stuff.
[12:56] And frame by frame. We're just monitoring the behavior of those vehicles with pedestrians and whatnot. So really the action that we care to find is that a vehicle starts outside of the zone and enters the zone, but particularly at speed, and then we send an alert for that. The specifics around that is important because there's plenty of trucks already in the zone.
[13:18] Right? There's the foreman's jeep that's already parked on site. There's all the work trucks and all that type of stuff. So, the system, while simple and intuitive for people to use and understand, does need to be somewhat complex to filter out day to day things of lots of cars will be in this. We need to not alert when those things are happening, but do alert when
[13:39] you know the bad thing does happen.
[13:40] And false positives are tough, right? Because this kind of worker, if they're experiencing a false positive once a day, they're going to get through this thing overnight. Yeah.
[13:47] That was that was the biggest feedback as well during the during that workshop. Right. They said, hey, the systems who tried to pass we got alert fatigue on day one.
[13:57] Yeah. More notifications than they were ever going to respond to and say we would never get our job done if we had to take action every time we got one of these alerts. So Tim alluded to it. We don't try to predict that someone's going to enter a zone, and that's on purpose. If you look at this data, we looked at it right.
[14:14] The fraction of second that you might see a car going in a straight line first suddenly veer into into that work zone. Even if you could predict there's maybe a fraction of a second that you can see that difference in trying to tell if they're going to impede in the zone for they're going to correct. So the cost of that many false positives versus saying, hey, when they
[14:36] impact the zone, that's where we want to send.
[14:38] And relative to the human reaction time that's going to be, you know, your computer is gonna be able to react way faster and it's not actually going to make a real difference
[14:44] for the people. Yes, exactly. So in that that is one of the benefits of a vision system like this as well. Every time one of those alerts goes off, we have the ability to capture that information and record it and show the user.
[14:57] So if they're like, hey, why did that one go off? Well, here, here's the F2 50 that just came and knocked down two barrels
[15:02] before it corrected itself. Or watch this guy's mirror get folded in as he tapped the barrel. Right. Yeah, exactly.
[15:09] Exactly. So you did ask, how did we get started with this? So we worked very closely with Google, has a rapid innovation team that's focused specifically in the sled area, trying to say, hey, where are the problems that are industry wide, where we could take some of our best partners?
[15:26] Clear object being one of Google's partners, Premier partners specifically for vision in in state and local government and said, hey, let's go run this and facilitate this workshop. Talk about the technology that ClearObject has in its capabilities, and see if we can come up with something from there. We did a six week POC, and that six week POC was to demonstrate the feasibility that a vision system can detect a vehicle entering a zone, and we can make something go deep
[15:56] and make a horn go boom.
[15:58] Yeah. How hard is it? Turn on LED on that. Yeah. And
[16:02] I will say where I was wrong when we went into this. We said, hey, we'll use as much commercial off the shelf. We knew, okay, camera is not a problem. Compute.
[16:10] No problem. I didn't know a wearable at the time, but I soon hey, some locally connected wearable.
[16:16] Their cell phones. Like why can't we
[16:17] just call yourself a cell phone? Yeah. Yeah, obviously with, you know, and we even tried talking to the Google Pixel team to say, hey, can we just use your watches? And no, nothing was available. That really allowed us to need a local network that could send a very quick signal. Low latency.
[16:34] Right? I don't need any, you know, the lowest latency from the time our system can detect it to the time that worker gets it, more seconds that they get to try to get these safety. So, you know, we initially took, basically one of our developers who's very handy, a little electrical guy, you know, bought some off the shelf type of hardware, I'll say kind of off the shelf.
[16:58] It was the night before the actual demo inside the hotel room, soldering some stuff together, making sure it would work. But I will, you know, happy to say in six weeks you're able to demonstrate the feasibility, right? And so that's where after that came after that project POC was complete, it was really okay. How do this wearable that we used was only to demonstrate you can get a signal and makes up and go beat.
[17:25] It was not going to be what the workers are going to. It's not reliable enough. It demonstrates the feasibility, but it was not reliable enough. And that's where Tim reached out to a couple local contacts that he had to say, hey, who would they recommend? And that's when we got introduced to Glassboard.
[17:41] Yep, I have 720 bucks. I gotta figure out who it was.
[17:44] But no. I think the really neat part of this narrative for everyone listening to trying to how do you go from idea to hardware product? And there's many of these, like lily pads that you jump to along the path and you don't go from, I have an idea to I build this thing in the retail, you test the waters along the way, and you guys did this organically.
[18:02] Some people come to us and I'd like tell them to slow down and let's go do something that seems silly, that, well, we're never going to be able to sell that. I'm like, I know, I know, but you guys did the right thing. You took some off the shelf dev kits and a 3D printer and some duct tape and hot glue, and ran out into the field in the environment it's going to be in and showed someone the vision of this.
[18:19] Working in the base technology, we can send a wireless signal to a device that let someone know our system triggered and that got you buy it right from customers that got you buy it internally to your leadership, and your engineering team got fired up because, look, it's going to work. This is what we have to make real. And I think that's such a powerful tool that most people that are first time founders and hardware miss, because in software, it's so easy these days, especially lovable and your clods and all that.
[18:43] Oh look, here's my clickable prototype. Don't worry about the bugs in the background. We can fix that later with money, but I can show you the vision. In hardware, it's so much harder to have the vision unintended to say, I can do this thing that definitely is in production, but just go with me here. And it's such an important part of the process.
[18:59] And where we got to jump in with you guys is you said, hey, I have a vision. I know that off the shelf can't do it. And for everyone listening like this is where commercial cell phones don't cut it. Yeah, they don't have the range. They don't have the battery life. They don't have the, call it control, like infrastructure control.
[19:15] Have you guys being able to onboard them to a work site? Where is this phone? Who does it belong to? There's all these little things that stack up where customization really becomes quite important. And we get to jump in knowing exactly what you wanted. You held up this 3D print. I want it to be like this when this thing happens, but it has to actually make a construction worker happy or a foreman happy in all these different aspects.
[19:37] So we get to jump in with you and go design thinking again, right? How do we pull this off? What does that technology look like? How was that discovery process for you guys? Is it similar to what you experienced internally for software
[19:49] just with atoms instead of bits? It's been interesting because there's a lot to learn. Basically, our first idea was, hey, we have this thing that already beeps in blinks and we want it prettier.
[20:02] Then you come to realize, okay, we need it. IP67 test, you need to be cleared. We need all these different things that we know nothing about. We know that we can make a dev kit beat and blink. And that's kind of why we're like, tell us what we don't know to start and then we'll start designing. Move from
[20:16] there.
[20:16] Yeah. And it's like, how how do we provision this? How do we charge this effectively? How do we make sure it's available, available on the truck when they need it. And it's all, you know, what if isms. And this is why going slow is so important. Because you have to feel the friction to figure out what you have to add oil to, to Greece it.
[20:33] Because otherwise you can do one of two things. You don't grease enough things and you build a product that's way too much friction. The user will want to use all. You make it way too perfect, and you run out of money, or you get to market in volume. And this is the the fun scale of where we're at.
[20:48] Are you gave us the mission of, hey, we have to make enough of these things. This is like a real amount of volume. And this isn't like we can build two of them and keep them alive, because I can fix them every couple of days, right? And this is where our job gets really fun. So, Elijah, where in your world was this in a balance of like, they'd already done the dev kit trial.
[21:06] Yeah, we had to go back and do a little bit of that, but not to test the product, but to test for core technology. Right. Yeah. It was it was a unique new problem because like you said, the number of devices that we needed to deliver was more than we usually do. Usually it's high volume. Right. And so we're not communicating with a contract manufacturer to go and have, you know, injection molded parts and mass produced PCBs, things like that.
[21:29] But we also need like we need to be able to manufacture this thing in-house, but they need to go out in the field. They need to be reliable because we're not going to be able to touch these things for months. So we need to actually run them through testing. They need to be able to survive the harsh environment, the physical environment.
[21:45] So that was fun for our mechanical. Yeah, it's way harder than inside. So then yeah, they need to be able to survive that, you know, harsh sunlight, you know, being on someone's helmet while they're jack hammering something like that. The vibrations water and rain. Exactly. Dust. And yeah, being thrown into the back of a truck and then having a helmet thrown on top of it, things like that.
[22:06] So yeah. So it it provided some unique problems, but I think it gave us an opportunity to really flex our muscles when it comes to testing. We did some really rigorous testing on the front end and tried to keep you guys in the loop as we were going on that, especially around the wireless solution, that was the big concern.
[22:22] It was hard. It was it was a very hard problem to solve because like you were talking about, your two typical wireless solutions are Bluetooth that everybody knows that does really short range and is really low latency, or is cell service, which is long range. But like you said, your workers aren't going to have cell service out there.
[22:38] So we had to find a technology that was kind of in between. It was a think it was Wi-Fi. Yeah, all things Wi-Fi. Really powerful. Yes. Really far range, really high data rate and is really high data rate. It's great for houses and buildings, even an open like line of sight backyard. It is crazy how not far Wi-Fi goes.
[22:55] Yes, exactly. And so we had to look into some new technologies and we ended up going with Wi-Fi HaLow, which is a longer range version of Wi-Fi, and went through some really rigorous testing on that and ended up being able to deliver super low latency for you guys. It actually ended up being like comparable the like deliverables to what Wi-Fi can do from a latency perspective, which is awesome.
[23:15] I found out I have Wi-Fi HaLow on my lawnmower at home, so technology, it's getting everywhere now. It's a it's a new it's a new product, but it's starting to start on the rise. So and so I think it was really cool to be able to take this vision that you guys had, get a hey, we can build it here.
[23:31] Here's the roadmap from once we prove that this product works at this volume, here's the roadmap to go to high volume. What everyone thinks of I think of the electronics, which is what you guys are building. Right. But we can't jump there. We all wish we could. It would be smaller, it'd be tiny or it would be flashier.
[23:46] But we had this middle ground. And the feedback that you guys are getting on the product is so useful for us both. From does the end product do the solution and how does the wearable affect the user? What do we all need to work on between now and the end, and how has that been for you guys getting that feedback back?
[24:02] Your software team that are usually totally used to, oh, you want me to fix that? All right, let me figure this out. We'll run some internal regression testing and push the software update. Maybe it makes it worse, maybe makes it better. But you can always push a software update. You
[24:15] know, we call those recalls and those are tough.
[24:17] Yeah. It's it's been interesting going out and talking with the clients. I was just in New York last week. We were we you know, we were out there demoing to a few different, few different depots that they have that will be getting these as we roll these out and hearing their questions, hearing their feedback on the hardware.
[24:39] A couple of interesting things. One was I actually looked at the date. It was about a year ago that we actually went out to this site and did a demonstration of the POC. So while we say it was long to them, they were actually quite amazed that like, hey, you guys took something that, you know, you showed us the concept and reality in almost well, it's less than a year, but just about a year.
[25:02] You brought it back out here in a full production state that we're going to be using day to day to help us. So that was that was pretty amazing to hear that feedback. Like you do say though, when it comes to the hardware, right. Like, hey, we wish it was louder, maybe vibrated harder, but other than that, hey, it served its purpose like it's it.
[25:19] I will wear it. It is going to protect me right now. That's awesome. Yeah, that is great to hear. And you know, they've all got some ideas. Some wanted to wear a wearable around their neck that had. Yeah. Joked could also put a little fan in there to keep them cool in the summer and hot warm in the winter you know.
[25:39] So it's really interesting to hear their feedback. But you get there by in because they're like, hey, I understand this. Like what your the purpose of this, you know, hardware is and now that I'm bought into it, great. I just want to thank the different ways I can use
[25:51] it. So and I think that I think it's the great thing about this quantity specifically of doing this kind of middle quantity, not thousands and thousands of devices, but more than just you can kind of show them at the front of a room and say, hey, look at this.
[26:01] This is cool. Yes, this is enough that you can you can hand it to them. You can leave them for weeks, let them use it on their own. And you get the the types of feedback that you would never get from just a demo in front of a, in front of a conference room or something. It's so true.
[26:15] The what's the word? It's the I live with this product first as I saw it. Yes, I saw a demo. That thing was really cool. Like they had to reset it once, but it was really cool versus oh no, guys, I can't stand this thing because every time I charge it this one way, it doesn't charge and I don't know that it's not charging.
[26:31] I'm just giving some examples of other products we've had of like, yeah, if the charge indicator isn't really obvious, people think they're charging even if it's not. And if you'll learn that in this volume that they can take it home, they can live with the pain, right? Because friction and pain is the only way you make a product better.
[26:46] Yes. Very infrequently. Do all of us sit around a table together and just think about how do we make something really cool that never hurt anybody or didn't have a pain point? Yeah. That's the that's the fun part. And so, Tim, in the actual application of like the vision system working in real time, how has that been for you guys to pilot from proof of concept to pilot hardware on like the software hardware integration side?
[27:10] Well, I've spent a lot of time in the parking lots near our office, driving my coworkers Tesla at questionable speeds towards towards cameras. Yep. But it's been a lot of fun. No, it's been a really interesting learning experience for me to think through the hardware and real world side of the implications of our software, the logic that needs to exist to filter out false positives.
[27:36] As we were talking about earlier, the implications of every connected part from our actual software to, you know, the Wi-Fi HaLow access point to how that sends the message out. And really, my goal at the end of the day is kind of getting the the must haves as the system working in as quick of a routine as possible.
[27:56] When we first started this, we were budgeting, just kind of guessing like, hey, we want to get this within about 700 milliseconds or something like that. And just through continuous testing on every single part of the software and hardware line from the hardware that we own, which is the edge devices, the compute that actually runs the inferencing to the software that we own, the logic.
[28:16] What trade offs can we make for tracking that speed up or slow down? How long we need to wait until we know an action is true, to like your guys's side of the WiFi, HaLow and all that stuff. We've gotten things down to, you know, 4/10 of a second. So we calculate or assume that there's a budget of about 2.5 seconds from an event occurring, something needing to alert someone, and then that person needing to get out of the way.
[28:41] Right. So keeping our number as low as possible, that 4/10 a second, is kind of our main goal. We're there now. It feels great to do that. And it's just been a great learning experience for me of, you know, it's not just a website or a dashboard or something that's monitoring a manufacturing line or something like that.
[28:56] This is a whole interconnected system.
[28:58] So. Well, I think the the other part that adds for both of us, both on the hardware side and the software side is there's this magic line when you cross between a convenience product and like a critical product. Yeah, this is can this product hurt somebody or can us not doing our job allow someone to be hurt?
[29:13] That is the whole purpose of preventing harm. This medical device, the safety of the things that the tone in which you look at, like DFM or failures, changes in all of our heads. Right? Hey, how acceptable is this risk? And it just changes. And I think that when Glassboard first started doing medical device 7 or 8 years ago, it was such a culture shock for us, right?
[29:36] Because it was always most clients were startups when we were young. It was moved fast and break things. Shoot from the hip, just go. Yeah. As long as it works, it's good. Like they can reboot it. Yeah, like you can't reboot that pacemaker. That's tough. That's. You can't do that one. Yeah. And it it just gets you to think about product in a different way and think about testing in a different way.
[29:54] It's both the failure mode and how do we test for that failure mode. And I'm just curious on your guys from the software side, have you experienced that throughout your journey, whether this project or another one of like changing console was actually usually been pretty good at self tests and, you know, testing your revisions changes the creation for us, it's usually harder to test.
[30:13] How have you guys approached that with this product? Yeah, I think I'll answer your question. We'll see. You brought up a couple points that I think are interesting around this, but the gravity of this use case, this is people's safety has really, you know, hardened our perspective on a lot of this one from a testing perspective with just the operators themselves.
[30:34] In a lot of our past work, we're working with manufacturing clients. And there's a couple stakeholders, the client manager who just wants this extra data point, and maybe a data analyst who's going to consume our stream of new day. There's something like that. Now we're dealing with tens, hundreds of people who will be using this for their safety.
[30:52] And that just informs the whole testing system. And, and, and really helps us quite quickly and easily say, what are the absolute must haves? This must beeps, blinks at 1300 feet, two miles, something like that. In good conditions. You know, it makes also the operators. I again, we've worked with plant managers who are excited like yay, new data.
[31:18] But having hundreds of people saying, I don't care that maybe this is going to could be a little smaller in the future, it could be a little louder in the future, isn't going to protect me now, and I'm excited about that opportunity. Everyone's been so receptive. I mean, again, I've had great, great experiences with a lot of our clients we've worked with, but I've never had so much reception from, you know, the actual day to day people using this of just this is great.
[31:42] Let's, let's see how we can move this forward as well.
[31:45] So I don't really know if I answered your question, but. Oh, you totally did. You hit what I didn't ask, which is how much more rewarding it is. You get to actually see an end user, not just make more money, be more efficient, or have a cool what is it like?
[31:58] There's fashion, aspirin and medicine. Like the three ways to pitch product, right? You're either fashionable aspirin, like you make my life slightly easier, or this medicine you saved my life. When you're building medicine products, whether it's safety or actual med device, like it is a different experience with your user. I think that is like really rewarding. You and I experienced that and a couple of other things we've worked on.
[32:18] Absolutely. We've had some really cool aspirin products that we love because fun. And they're great. Yeah, but it's different when it's like this level of safety. Yeah. I remember specifically, I think, Derek, you made a LinkedIn post when we first put the demo unit out. And I remember the top comment was like somebody I didn't know, but it was like my cousin got killed while he was on a roadside construction.
[32:37] This would have saved his life. And I remember seeing that and being like, whoa, this is like, I knew that it was a thing. But seeing someone's testimonial in there to say, I wish that that somebody I loved had this product because they would still be here. That's awesome. And it's awesome that the size of team that's able to make this happen with today's technology.
[32:56] And this is at the end of the day, what's amazing is we have a handful of people at Glassboard that work on this, right? This isn't a team of 20, right. This is like what for. Yeah. Like for for really smart people in a room doing electrical and mechanical industrial design and firmware. Yep. You guys have a team of what how many people are actually on this team?
[33:15] It depends on the week 5 or 6. Yeah. Again, I'd probably say a couple pizzas. You could probably feed the whole team between
[33:23] Glassboard and yeah, that's been able to build and develop this entire solution. Right. Yeah. In a timeline. That is really cool to see from concept proof concept where you guys picked it up, where we got to do the first demo of the tech, got to make the first unit.
[33:36] Now we're actually launching build batches for for your team. And I think this is just a really cool testament to how awesome the shoulders of giants have been for all of us to stand on. Right? You guys didn't code the AI algorithms from the ground up. You use technologies that exist. Same with us. We didn't make the microcontroller.
[33:52] We didn't write the course off or library. We imported a bunch of drivers and put them together. And modern engineering is a lot of both software and hardware teams to do things in incredible pace that I just think is made.
[34:05] Yeah, it's it's you know, you mentioned. Yeah, I saw that post. It's amazing when I've gone out and talk to people about the solution.
[34:13] I've heard that story way more than I ever would have thought of. Yeah, I'd probably count count, unfortunately, in two hands, the number of people that have had a very similar story when I've talked to them about this. So, you know, it's really gratifying to hear their feedback, to say, hey, I'm glad somebody is doing something about this.
[34:31] And you said, it's amazing the pace we were able to do this. It wasn't like, hey, ten years, we might have something that helps you. This is hey, we've got folder technology today. Let's put it together and let's roll it out. Yeah,
[34:41] yeah. And I think the the other neat thing about all of this is the reception, not just at the end user, but of the call, the end adopter.
[34:48] Right. So just for everyone listening. It was a good lesson for anyone hard for like your user isn't always your customer, right? At the end of the day, they're very separate stakeholders when we're developing products and companies. And so who is the like the actual customer for you guys? Who is the group that you're having to reach out to?
[35:05] Let them know the technology exist? Like what is this person versus the person who is on the job site on the road probably isn't the one who's writing the check for this device.
[35:13] Yeah, an interesting point on that. I'll let Derek answer on specifics, but you're totally right that the buyers are different from the users in as far as what they actually want as well.
[35:21] We've heard from some of the buyers who kind of run the safety orgs, rather than people on the side of the highway who say things like, you know, actually, we would almost just buy this without the wearables for the data that we get on, how many intrusions we get throughout the day. If our code setups are particularly safe in one way or another, if ghost cop cars reduce the speeds of car driving by, things like that, and then it's a huge added bonus that there's also a wearable that can alert the workers as well, right?
[35:48] So that's the difference really between the purchasers and the users and also the users at HQ. They're using it for a different purpose and whatnot. So just to point out, like it's a pretty wide platform on, as Derek mentioned, even with the manufacturing stuff, we get data around the whole thing so that you can look at what happened, not just if there are events, but, you know, get data for potential policy making around, as I mentioned, you know, ghost cop cars.
[36:11] Hey, we saw that that reduced the speed by ten miles per hour on average when we were able to, you know, rent those out from the station or something like that. But yeah, we can speak to kind of who you think the market has been. Yeah. You know, there's a there's a very broad industry wide concept. It's called Vision Zero.
[36:28] The whole idea of zero deaths on the roadway. Right. And what is everything that everybody can do to try to achieve that? That's you know, there's almost a thousand work zone deaths per year, US raising. Now, not all of those are workers, but due to the fact of, you know, if there's a soda on traffic, people aren't paying attention.
[36:48] Right. So luckily, there's a lot of people doing a lot of really good things to try to make all the roadways safer. You know, we are a piece of that. We are a component of that. And so when we reach out, we talk to clients. We're talking to, you know, roadway authorities, tollway agencies, contractors who are out there.
[37:08] And really, most of the time it is their chief security officer that's looking for new ways to improve the safety for their team. So that is usually the person who's got the budget who can talk to us and, and
[37:21] has has the users underneath that, that also going to support it has the authority and the mission to protect the users.
[37:28] That is their role, that is their job. And I think it's it's always fun. Tim, you nailed my part that the buyer and the user and always the same person and they sometimes want different things. And my my favorite conundrum and building good product is distilling out of both of their desires. What are the needs for each of them?
[37:47] Because it's 50% need and 50% desire usually, oh, I need it to do this, but it'd be really cool if something happened and you can find out that that really cool doesn't move the purchase price or the desire to purchase at all, even if you get a ton of this feedback. Same with the user. They're going to say, man, I wish it, you know, make me bacon and eggs in the morning.
[38:05] And they all wish that. But it doesn't affect whether they actually use it, whether it actually makes their lives better, what actually makes them safer here. And I think that's the hard part in being like the taste makers in product development is how do you discern even desire from actual purchase price want.
[38:20] Right? I feel like that's one of the biggest things I've learned.
[38:24] Just working on a lot of our consulting side generally is just people have a lot of ideas, and we kind of have to lead them to, okay, which one is actually makes sense
[38:32] to help you out. And what's interesting is when we always do that, like you and I have this role, it's almost always math. Hey, you have all this emotional want.
[38:41] Let's go try and put some numbers to each want. How much does it cost to develop? How much does it make the bomb cost or the product change? How many users actually want it of your end users and people from your customer base, not just in your company? What are they willing to pay differently about it? Right. This is effective purchase price.
[38:57] Does it affect all these things? And it's the nerds in the room having to use math to go emotion math into an equation to make a spreadsheet. That's all I like. Does this return on investment either in time or money or or advertising? Yeah,
[39:11] right. In our world it's a lot of the time, like pay for accuracy, kind of like not really, but kind of it's like if you bought a better camera sensor that is inherently more expensive, we would,
[39:20] you know, get more pixels to look at or better lighting or anything like that, or it would have more frames per second output for you to interpolate between or run inference on or whatnot.
[39:28] I need to run this on a tiny GPU. Or can we get the one that makes sense for this scenario? And. But at the end of the day, it's, you know, there can be an issue sometimes with people wanting to over science project something, and it's totally reasonable for a client to say, hey, we want 80% accuracy and we're willing to pay this much for it, and we can kind of get them there specifically on that.
[39:49] So like you're saying, reaching that middle ground of what's possible, what does it take to get there is,
[39:54] I guess, what all this is, it's the art. Yeah. Right. In all the science that we all do, it's the actual art of figuring out how good is good enough, because there's always V2. If you make v1 good enough to get to market, you always have enough time, money and effort to get to V2.
[40:07] And I think when I started in, in this field, you know, I thought a lot of my time would be spent coming up with cool ideas and convincing clients that they need them. A lot of it's actually spent convincing them they don't need the cool ideas they came up with. And coming up with this, this what is the I mean, we use the term minimum viable product that you can actually get out to, to your customers, especially at this stage you guys are at what's the minimum thing we can get in front of the people who are going to use it so that we can get
[40:32] feedback?
[40:33] Yeah, that's one thing. Working with Glassboard throughout this whole process, specifically myself. I know I've had to make decisions. When it does time come, time for cost and money to get to where we are today, right? There was times we made those kind of asks and you had to tell us, hey, that's we
[40:49] should probably not do that.
[40:50] Yeah, I do that. And there was other things that you said, hey, you should think about these things. I mean,
[40:55] I know specifically one was, hey, what about again somebody relying on this for for them to save their life. The last thing I want them to do is be relying on it, and then they're outside the range and they don't know that.
[41:07] Yeah. So one of the you know, I think that, you know, as through the process, one of the things that brought up by your team is we need to make sure that, hey, when batteries low or their upside of range, this thing starts beeping in a way that's not like the alert, but in a way cut. I think we used it as the smoke detector.
[41:23] Simply start giving you a little
[41:25] chirp to let you know. Bring your attention to it, to say, hey, you're no longer connected. I'm unhappy. You can't listen to me. I'm not going to be useful. Yes, please get back and range or know that you're going to have to watch for cars. I can't do it for you. Yeah, that's so much fun.
[41:39] Well, I think the the neat thing about all of this is we're building a product that people are currently like today using, which is really neat. But you guys have built a technology platform that has other applications, like this is one of those moments where, yes, someone built a vision system in the in the early days and discovered computer vision and AI vision, and then you guys would applying it to all these use cases.
[42:00] This is one of those use cases where you had to build a complex version of this that detects this kind of motion, these kind of like cars and this kind of space. And you're starting to see other applications, right. So I think that our construction safety is the first place you guys want to tackle this. What are the ones that you see coming down the pipeline.
[42:19] That would be the V2 or the V3 of this kind of a
[42:22] platform. I mean, even in the same vein, on construction safety, we were out in the field with some guys recently who it was really gratifying to see how quickly they intuited how the system works. And they're like, oh, well, we could put one of these cameras on a backhoe with a swinging arm and make sure a person doesn't walk into the zone.
[42:39] I was like, yeah, you're exactly right. We're detecting people as well. We can config swap and say alert if a person enters or anything like that. So first off, like that's a separate zone that this works in the same space. Yeah it's in customer base just different applications, same customer base. But also it was really exciting for me to just see like like I work on the side of the highways like they're talking about.
[43:00] We don't understand this stuff at all. And your AI vision stuff, you know, the kind of joking with us.
[43:04] Yeah. Well, who was the wizard in the box actually pushing the alarm button? Yeah, but also, they understand it quickly enough to be like, oh, I could just
[43:10] draw this zone on whatever camera I want. And then the obvious kind of fallen from there is service workers, like traffic stops in a police car or something.
[43:20] Same idea. You have a camera pointed backwards. Potentially. They already have rear facing cameras on. There may or may not need the wearable device on the officer. If you can tap into
[43:28] the radio like they're wearing a radio live right now, they have lights. They have sirens already up there. So that vision system, it's the same
[43:35] logic. We've already built same platform from the edge, compute all the way up to where we need to send the message, and then unplug the Wi-Fi HaLow for the wearable device.
[43:43] Potentially. Yeah. Love it. Which I mean, I think we might have mentioned, but like, you know, we started with the trucks specifically with a siren plugged in. Then we added the wearables and whatnot. So we already have the siren component add the wearables. It's really just much a Lego pieces. You can kind of plug
[43:58] and play in different combinations.
[44:00] That's just engineering. If you zoom all the way in like a microprocessor, it's just a bunch of light switches together like Legos in a very complex way that we, you know, we we arrange the rocks in a way when we hit with lightning, it things amazing. But when you take engineering all the way down, it is just a bunch of small Lego bricks that all of us by pre-assembled in some level of block.
[44:20] And then we snap those together into a more complex block that we then pass on to the next person, and someone ends up being the end user, right? Yeah, but that's the coolest thing about modern engineering is very little is ground up. Like truly, truly, from the ground up, everything is on the shoulders of giants. Even if we build a product that is fully custom.
[44:37] It's designed with software that was built by someone smarter than us, and all this humanity collective group think is what makes cool products half of the day, especially with a couple of pizzas feeding the entire team. Yeah, yeah.
[44:49] I mean, I'll say like from my perspective, like vision is super cool. It's really exciting software space to work in.
[44:54] But at this point a lot of vision models are to this point, you know, commoditized, where pretty much anyone can go train a YOLO model. Every person I interviewed for an internship has trained a little detection model or something like that. But to your point, it's how you connect all these pieces and come up with a final solution that ties everything together.
[45:10] That's kind of where the sauce is. And I mean for the wearable specifically in this scenario, that's like a in our view, a huge differentiator, right? Like if the vision is commoditized, you know, anyone can go train a computer vision model to go watch a car potentially enter a zone. Now we do, I would say, have expertise around what does doesn't work, speeding things up.
[45:29] And, you know, maybe I've just been doing it too long and I'm used to it. So but being able to work with you guys on the hardware side is really opened my eyes on
[45:36] the iteration side from that. Well, I think the, the what, you were actually just eliciting that you feel like what we do is black magic.
[45:42] We feel the same way about how you guys manipulate software. It is the awesome world we live in. That 0 to 80% is really easy today. The tools will work with makes getting 0 to 80% of a thing just awesomely fast, right? I think your first demo, your first proof concept, both the software and the hardware that took to get there.
[46:02] But the difference between that and a product is the last 20% of functionality is 90% of the effort and the expertise, and everyone having their 10,000 hours in their craft. That makes it a wonderful product that actually wants to use every day and makes it a business.
[46:16] And to thinking about, I don't know, just thinking about the users, think about the product as well.
[46:21] Like, I came into this world as just a nerd who likes solving technical problems. And I've learned a lot from Derek, specifically around, you know, not everyone just wants something because it's clever or unique or interesting. You have to think about who's using this, what it does. I know you came up with, you know, Devin jokes about breathing LEDs or whatever.
[46:40] Yeah, silly things like that, where it's like we can do clever stuff as nerds with our software and hardware.
[46:45] But what does the actual person actually want at the end of the day? Yeah, well, it takes all kinds, right? We need the business people to make a business case. So there's enough money to have enough time to go to the creative case, and it takes both of those things together to actually make a product that solves a problem or gets to survive into the world.
[47:04] And then eventually, it takes the marketing people that put lipstick on it and make sure everything's pretty and gets it out in the world and goes from there. But I this has been such a really fun conversation, going about how do hardware and software teams work together? How do all of us make something from nothing? I think the what I want listeners to take away from the episode is there are all these small steps, all this attention to detail.
[47:26] That is what's differentiator between here's Michael Demo and what is a valuable product or a valuable company at the end of the day. And that last step is those valuable products or companies end
[47:36] up making users happy great or safe. Right. So that's I think that's an
[47:42] amazing episode for us. Derek and Tim, thanks for joining. Thank you for having us.
[47:47] Thank you and everyone listening. We will have not to bury the lead, but to bury the lead. We'll have much more cool content coming out about this product and these teams. Shout out to Altium is working with us to make some really cool content outside of just these four walls, this podcast studio. So stay tuned for more. And if you ever need help working on these kind of problems, let let us know.
[48:06] We can hook you up with your
[48:07] object team. Take care everyone.
ClearObject - Four Tenths of a Second to Save a Life
Guests: Derek Bleyle, VP of Product, ClearObject
Tim Hagan, Principal Data Scientist, ClearObject
Hosts: Grant Chapman, CEO, Glassboard
Elijah Clarke, Hardware Engineer, Glassboard (guest host)
================================================================================
NOTE: This is a machine-generated transcript, lightly cleaned for proper nouns
(ClearObject, Glassboard, Wi-Fi HaLow, guest names). Speaker attribution was not
reliable in the source file, so lines are timestamped but not labeled by speaker.
Verify any quote against the audio before publishing it.
================================================================================
[00:00] Welcome to the podcast.
[00:08] Everyone, welcome back to the Hardtech Podcast. I have not the usual suspect with me. Oh, Aisha, thanks for joining. Hey, how's it going? So we are here with some fun friends from ClearObject.
[00:18] Derek and Tim and our team have been working together on some really neat tech. Part of the reason we had them on today is the journey that they've been on from a software centric product and going through the crawl, walk, run of hardware. I think it'll be really good for our listeners to hear. This is truly how you get an idea out, get it tested, and then start getting user feedback before you hit volume production.
[00:39] So that's the story that I think school for the listeners, but I think for everyone that is building a hardware product today, it almost always has a software feature or focus in it. And you guys can finally tell that narrative because we haven't had that many software folks on. So thanks for joining us.
[00:53] Thank you for having us.
[00:54] Thanks. Thank you. So Derek, I'm going to pick you first. All right. Derek Bleyle, senior VP of product and operations have ClearObject. So got the responsibility of the engineering and delivery across the organization. Condolences. Yes.
[01:12] Excited to talk about you know, one of our you know, I'd say probably the project that we're all the most proud of at the moment today here. So looking forward to the conversation. Awesome. Tim. Yeah. So Tim Hagan, principal data scientist at ClearObject. Yeah, that I kind of split that in two parts principal and data scientists.
[01:31] So, principal, I kind of run a lot of our slide opportunities and a lot of our vision stuff. And then data science responsible for data science team.
[01:40] Yeah. Like Derek said, really excited to talk about the product we've been building alongside you guys. Awesome. And then you're you've been a guest host before. But what is what is your role been playing with these guys on this cool product.
[01:50] Yeah. So Elijah Clarke I'm a hardware engineer here at Glassboard, and I was brought onto the project with them as a hardware designer. So designed all the circuit boards in the product and then also just worked on the larger side of the project, doing a lot of experimenting, playing around with different technologies and helping them to achieve the goal here.
[02:09] So let's start with ClearObject. What is ClearObject? You guys are in software, which is really broad word, you know, what do you guys do as a as a core business. How does this fit into that. The product that we're going to talking about.
[02:22] Yeah. So I've been with ClearObject eight years. So I'll talk a little bit about the journey we've been on, because it's changed a lot in those eight years about what we have done.
[02:32] So we started as an IoT technology company, meaning, you know, eight years ago, everybody was wanting to say, I do have hardware. Now I've built it. How do I connect it? How do I make use of the data coming off of that software? So started with a lot of connecting to those products. We weren't we weren't developing them.
[02:52] Right? We were more of a we were helping them with their products, build the digital solution to that in the software side of that solution, from that, you know, we returned as an organization, continued to focus a lot on not just what sensors, but how could probably the most powerful sensor out there as a camera, right. So it's multimodal, what you can extract from that.
[03:17] So we really started to lean into the space around vision AI. So extracting information from cameras, making real time intelligence out of that as a connected product, turning that into whether that's insights for manufacturing or food and beverage retail, all of the above. So that's really been our focus over the last I'd say 4 or 5 years. And this broad project that we're going to talk about today started about a year and a half ago, specifically looking at how can we use vision, a technology, to solve a problem specific to it had a lot of similarities of what manufacturing has.
[03:58] So manufacturing you need real time intelligence, meaning, you know, I don't have any latency talking, you know, 30 frames per second type of inferencing speed. All because I need to be able to understand what's happening, make a decision, infer, communicate that
[04:17] to something to take action in real time. Also, you know, if the
[04:23] network goes, there's no Wi-Fi connection.
[04:27] Manufacturing facilities still need to manage. So we'll talk a lot about the the work zone solution here today. But I think it ultimately combined a lot of our specialties, from connected devices to software to how we use the connect, the data coming off of connected devices to
[04:43] provide intelligence. Yeah. Well, I think it's always fascinating to see the evolution of a tech like dispersion in the ecosystem.
[04:51] Right. When we all got started in connectivity in the 2010s. Right. Like the 2010s, we all thought it was going to be industry 4.0, because B2B is usually how most technologies break out and find the first fit. And we thought industry and manufacturing and connectivity and it's big business was going to take off. And we were all wrong because everyone got a smart door lock and a smart thermostat and smart toaster first.
[05:13] And then now we're starting to see industry pick it up and medicine is now going into it. But the data that is coming out of each one of these sensors keeps getting more and more rich is kind of like the narrative. I keep seeing the industry. It's a very simple thermostat or on off switch. Is my door locked or not locked?
[05:30] And all these very binary or individual number metrics that I'm sure you guys were building on of iOS apps and dashboards and databases to collect graph over time, and alert when number went above or below a certain number. And over time these started to get more and more integrated and connected with multiple sensors connecting to make if then statements right or automations happen.
[05:51] And then in the industrial automation world, speed started to matter. And I think this is where you guys got some of your really cool first applications from hearing about your history is you started to be able to take really fast action on a fast moving assembly line that isn't a mechanical operator anymore, right? A lot of times there's mechanical light beam, brake sensors and things like that.
[06:09] That would trigger a PLC, but you guys are adding way more intelligence for some of your clients behind the scenes, stealing what you took from IoT and putting it into a fast paced factor that has to run locally, not a large distributed network.
[06:21] That's exactly it. And I use the example of, you know, this is the simplest example of what is the type of intelligence today.
[06:29] You might have somebody on the line projecting, rejecting a component because they see a defect. Well, how can I augment that? I can automatically do that. But when I do that automatically with an AI system, I can know not only there was a defect where the defect, where was it, where it was the timeliness of that. And so I can track that information and start to do real time process control, plus tie that back to variables upstream to say, okay, what are the actual variables impacting that
[06:56] that are causing and find the root cause, be able to correct those.
[07:00] And if you put it into a big enough database with disparate enough data, you can do the really neat thing that we're all learning how to do, which is find the needle in the haystack. Hey, we see that the upper right hand corner is get bent in this part when it's really hot outside. And we haven't serviced this one, you know, in a while on one of the presses.
[07:19] And you can put these disparate pieces of data together if it all is in one place. And with modern compute and inference and AI and other algorithms, we're starting to find needles and haystack in this industrial in medicine and things like that. But you guys have taken large subset of data to a whole new level with the product we're talking about.
[07:36] So we moved on to vision of what you guys are building in this. And can you guys tell us what the product is so I can sink my fangs in and talk about the neat tech you guys
[07:44] build? Yeah, yeah. So I'll talk about it. We call it Clear Work Zone. But ultimately from the simplest perspective, it is built for field crews that are out working day to day doing operations along the highway.
[07:59] So we're all familiar. We're all driving every day. We see the cones on the side of the road. You know, you might, may or may not pay attention and see that crew there. Tim and I can both tell you from being out in the field, how scary that really is to be, alongside traffic flowing that supposedly 45 miles an hour, right?
[08:18] We all know that's not the case. And so we were approached by a client and what they asked was, hey, we've seen what you guys could do with vision and other use cases. They they were forward thinking New York State Thruway Authority. They've been great partners, really, really leaders in this space to say we want to do something different to protect our workers.
[08:42] So we did a design thinking workshop with them. You go back to the thermostat, funny enough, right? We always started those workshops with tell me a new way to design a thermostat. Then we'd rephrase the question of tell me a better way to control the temperature of the room, right? So taking that back, we said, hey, what is the what is the best way that we can protect a work zone for those workers?
[09:05] And when we did that, we looked at other, other, other people who have tried this in the past. What were the limitations of what they've done? And really the feedback that we heard from workers is, I want something that is simple to set up. I don't want to spend any more time in the works on than I need to, because I got to set up the safety barrier.
[09:21] It has to be local. We work in areas where there's no cell connectivity, there's nothing else. Right. Which we'll come back to that when we talk about the wearable technology and why we selected those pieces and working with you all.
[09:36] Excuse me and be reliable. So ultimately, what we developed is a vision solution that gets put on work trucks and monitors those zones. So we could define protective zones to think of an imaginary boundary that's protecting that zone. And anytime a vehicle intrudes in that area, we in real time send notifications by our wearable, which is where we worked with you all to help develop that capability, where the workers received then a multimodal type of notification.
[10:07] So it's got haptic, visual, inaudible, knowing that they might be performing many different jobs. And we need something that no matter what they are doing, there's a form factor that can try to get their attention, given them every second and or what we say every millisecond
[10:22] of time that they can to get to a safe area. Yeah. Well, I think it's it's so neat what you guys have been able to do with the UX of the vision system.
[10:30] So for everyone listening, I, we all have bloodstream knowing exactly how this thing works. So when they say on the back of a truck, you guys have your camera system that is facing backwards, looking into traffic that's coming into the work zone. And you can use on a like a tablet, see that screen. And the operator can just draw.
[10:45] These lanes are open. These lanes are my, you know, orange caution area. And these lanes are completely closed alert if someone's going, you know, fast in the orange lanes, alert of someone's even getting close to coming into the out of bounds lanes. And that is the vision system that we've been talking about. How cool it is, how fast you guys can process that data.
[11:04] Because every millisecond here matters. You guys then send that down to your computer that does that locally. That alerts out you have some alerting on the truck. I think this was like the very first version of what you guys built is it wasn't actually on the user yet. You had an alarm system that could set off an alarm.
[11:19] We had danger. Will Robinson. Something bad is happening. And you realize that sometimes these workers are jack hammering into the concrete. Sometimes they've got their on a piece of machinery or they're too far away from the work truck. These work zones
[11:32] aren't tiny. How long is your lungs work? So we walked to the end of one once and made it 1300 feet.
[11:41] Couldn't see the trucks anymore. It was close, you know, the two miles back from where we started. So two miles back to, you know, another quarter of a mile
[11:50] forward. So yeah, you're talking two and a half mile closures that could be even larger than that. Right. Which means sound is out. Right. Just an alarm isn't going to get everyone that you're trying to capture here or cover up.
[12:00] And and you had to go reach them where they're at. And this is part of the design thinking how do we alert the user. Not only how do we detect, how do we even reach them in all the weird ways that they're working? Because your your use case user is so different than everyone else in everyday life, right?
[12:16] It's hot outside, it's bright, it's loud. It's all these things that aren't normal for even an industrial zone sometimes. So you guys have to tackle that. And that was a huge amount of fun. So how did you guys get started in the journey? Right. So we're design thinking. We're fever dreaming. Really cool ways to make people safe. We detect fast cars or cars going the wrong way.
[12:36] I mean let's start there. What are all the things you guys tend to detect? Like what are what are the modalities you guys can pick up on on call it drivers being bad behavior.
[12:45] Really we're just tracking the cars. So we run a couple models as the system running and we're looking for cars, people, cones, lanes, all that good type of stuff.
[12:56] And frame by frame. We're just monitoring the behavior of those vehicles with pedestrians and whatnot. So really the action that we care to find is that a vehicle starts outside of the zone and enters the zone, but particularly at speed, and then we send an alert for that. The specifics around that is important because there's plenty of trucks already in the zone.
[13:18] Right? There's the foreman's jeep that's already parked on site. There's all the work trucks and all that type of stuff. So, the system, while simple and intuitive for people to use and understand, does need to be somewhat complex to filter out day to day things of lots of cars will be in this. We need to not alert when those things are happening, but do alert when
[13:39] you know the bad thing does happen.
[13:40] And false positives are tough, right? Because this kind of worker, if they're experiencing a false positive once a day, they're going to get through this thing overnight. Yeah.
[13:47] That was that was the biggest feedback as well during the during that workshop. Right. They said, hey, the systems who tried to pass we got alert fatigue on day one.
[13:57] Yeah. More notifications than they were ever going to respond to and say we would never get our job done if we had to take action every time we got one of these alerts. So Tim alluded to it. We don't try to predict that someone's going to enter a zone, and that's on purpose. If you look at this data, we looked at it right.
[14:14] The fraction of second that you might see a car going in a straight line first suddenly veer into into that work zone. Even if you could predict there's maybe a fraction of a second that you can see that difference in trying to tell if they're going to impede in the zone for they're going to correct. So the cost of that many false positives versus saying, hey, when they
[14:36] impact the zone, that's where we want to send.
[14:38] And relative to the human reaction time that's going to be, you know, your computer is gonna be able to react way faster and it's not actually going to make a real difference
[14:44] for the people. Yes, exactly. So in that that is one of the benefits of a vision system like this as well. Every time one of those alerts goes off, we have the ability to capture that information and record it and show the user.
[14:57] So if they're like, hey, why did that one go off? Well, here, here's the F2 50 that just came and knocked down two barrels
[15:02] before it corrected itself. Or watch this guy's mirror get folded in as he tapped the barrel. Right. Yeah, exactly.
[15:09] Exactly. So you did ask, how did we get started with this? So we worked very closely with Google, has a rapid innovation team that's focused specifically in the sled area, trying to say, hey, where are the problems that are industry wide, where we could take some of our best partners?
[15:26] Clear object being one of Google's partners, Premier partners specifically for vision in in state and local government and said, hey, let's go run this and facilitate this workshop. Talk about the technology that ClearObject has in its capabilities, and see if we can come up with something from there. We did a six week POC, and that six week POC was to demonstrate the feasibility that a vision system can detect a vehicle entering a zone, and we can make something go deep
[15:56] and make a horn go boom.
[15:58] Yeah. How hard is it? Turn on LED on that. Yeah. And
[16:02] I will say where I was wrong when we went into this. We said, hey, we'll use as much commercial off the shelf. We knew, okay, camera is not a problem. Compute.
[16:10] No problem. I didn't know a wearable at the time, but I soon hey, some locally connected wearable.
[16:16] Their cell phones. Like why can't we
[16:17] just call yourself a cell phone? Yeah. Yeah, obviously with, you know, and we even tried talking to the Google Pixel team to say, hey, can we just use your watches? And no, nothing was available. That really allowed us to need a local network that could send a very quick signal. Low latency.
[16:34] Right? I don't need any, you know, the lowest latency from the time our system can detect it to the time that worker gets it, more seconds that they get to try to get these safety. So, you know, we initially took, basically one of our developers who's very handy, a little electrical guy, you know, bought some off the shelf type of hardware, I'll say kind of off the shelf.
[16:58] It was the night before the actual demo inside the hotel room, soldering some stuff together, making sure it would work. But I will, you know, happy to say in six weeks you're able to demonstrate the feasibility, right? And so that's where after that came after that project POC was complete, it was really okay. How do this wearable that we used was only to demonstrate you can get a signal and makes up and go beat.
[17:25] It was not going to be what the workers are going to. It's not reliable enough. It demonstrates the feasibility, but it was not reliable enough. And that's where Tim reached out to a couple local contacts that he had to say, hey, who would they recommend? And that's when we got introduced to Glassboard.
[17:41] Yep, I have 720 bucks. I gotta figure out who it was.
[17:44] But no. I think the really neat part of this narrative for everyone listening to trying to how do you go from idea to hardware product? And there's many of these, like lily pads that you jump to along the path and you don't go from, I have an idea to I build this thing in the retail, you test the waters along the way, and you guys did this organically.
[18:02] Some people come to us and I'd like tell them to slow down and let's go do something that seems silly, that, well, we're never going to be able to sell that. I'm like, I know, I know, but you guys did the right thing. You took some off the shelf dev kits and a 3D printer and some duct tape and hot glue, and ran out into the field in the environment it's going to be in and showed someone the vision of this.
[18:19] Working in the base technology, we can send a wireless signal to a device that let someone know our system triggered and that got you buy it right from customers that got you buy it internally to your leadership, and your engineering team got fired up because, look, it's going to work. This is what we have to make real. And I think that's such a powerful tool that most people that are first time founders and hardware miss, because in software, it's so easy these days, especially lovable and your clods and all that.
[18:43] Oh look, here's my clickable prototype. Don't worry about the bugs in the background. We can fix that later with money, but I can show you the vision. In hardware, it's so much harder to have the vision unintended to say, I can do this thing that definitely is in production, but just go with me here. And it's such an important part of the process.
[18:59] And where we got to jump in with you guys is you said, hey, I have a vision. I know that off the shelf can't do it. And for everyone listening like this is where commercial cell phones don't cut it. Yeah, they don't have the range. They don't have the battery life. They don't have the, call it control, like infrastructure control.
[19:15] Have you guys being able to onboard them to a work site? Where is this phone? Who does it belong to? There's all these little things that stack up where customization really becomes quite important. And we get to jump in knowing exactly what you wanted. You held up this 3D print. I want it to be like this when this thing happens, but it has to actually make a construction worker happy or a foreman happy in all these different aspects.
[19:37] So we get to jump in with you and go design thinking again, right? How do we pull this off? What does that technology look like? How was that discovery process for you guys? Is it similar to what you experienced internally for software
[19:49] just with atoms instead of bits? It's been interesting because there's a lot to learn. Basically, our first idea was, hey, we have this thing that already beeps in blinks and we want it prettier.
[20:02] Then you come to realize, okay, we need it. IP67 test, you need to be cleared. We need all these different things that we know nothing about. We know that we can make a dev kit beat and blink. And that's kind of why we're like, tell us what we don't know to start and then we'll start designing. Move from
[20:16] there.
[20:16] Yeah. And it's like, how how do we provision this? How do we charge this effectively? How do we make sure it's available, available on the truck when they need it. And it's all, you know, what if isms. And this is why going slow is so important. Because you have to feel the friction to figure out what you have to add oil to, to Greece it.
[20:33] Because otherwise you can do one of two things. You don't grease enough things and you build a product that's way too much friction. The user will want to use all. You make it way too perfect, and you run out of money, or you get to market in volume. And this is the the fun scale of where we're at.
[20:48] Are you gave us the mission of, hey, we have to make enough of these things. This is like a real amount of volume. And this isn't like we can build two of them and keep them alive, because I can fix them every couple of days, right? And this is where our job gets really fun. So, Elijah, where in your world was this in a balance of like, they'd already done the dev kit trial.
[21:06] Yeah, we had to go back and do a little bit of that, but not to test the product, but to test for core technology. Right. Yeah. It was it was a unique new problem because like you said, the number of devices that we needed to deliver was more than we usually do. Usually it's high volume. Right. And so we're not communicating with a contract manufacturer to go and have, you know, injection molded parts and mass produced PCBs, things like that.
[21:29] But we also need like we need to be able to manufacture this thing in-house, but they need to go out in the field. They need to be reliable because we're not going to be able to touch these things for months. So we need to actually run them through testing. They need to be able to survive the harsh environment, the physical environment.
[21:45] So that was fun for our mechanical. Yeah, it's way harder than inside. So then yeah, they need to be able to survive that, you know, harsh sunlight, you know, being on someone's helmet while they're jack hammering something like that. The vibrations water and rain. Exactly. Dust. And yeah, being thrown into the back of a truck and then having a helmet thrown on top of it, things like that.
[22:06] So yeah. So it it provided some unique problems, but I think it gave us an opportunity to really flex our muscles when it comes to testing. We did some really rigorous testing on the front end and tried to keep you guys in the loop as we were going on that, especially around the wireless solution, that was the big concern.
[22:22] It was hard. It was it was a very hard problem to solve because like you were talking about, your two typical wireless solutions are Bluetooth that everybody knows that does really short range and is really low latency, or is cell service, which is long range. But like you said, your workers aren't going to have cell service out there.
[22:38] So we had to find a technology that was kind of in between. It was a think it was Wi-Fi. Yeah, all things Wi-Fi. Really powerful. Yes. Really far range, really high data rate and is really high data rate. It's great for houses and buildings, even an open like line of sight backyard. It is crazy how not far Wi-Fi goes.
[22:55] Yes, exactly. And so we had to look into some new technologies and we ended up going with Wi-Fi HaLow, which is a longer range version of Wi-Fi, and went through some really rigorous testing on that and ended up being able to deliver super low latency for you guys. It actually ended up being like comparable the like deliverables to what Wi-Fi can do from a latency perspective, which is awesome.
[23:15] I found out I have Wi-Fi HaLow on my lawnmower at home, so technology, it's getting everywhere now. It's a it's a new it's a new product, but it's starting to start on the rise. So and so I think it was really cool to be able to take this vision that you guys had, get a hey, we can build it here.
[23:31] Here's the roadmap from once we prove that this product works at this volume, here's the roadmap to go to high volume. What everyone thinks of I think of the electronics, which is what you guys are building. Right. But we can't jump there. We all wish we could. It would be smaller, it'd be tiny or it would be flashier.
[23:46] But we had this middle ground. And the feedback that you guys are getting on the product is so useful for us both. From does the end product do the solution and how does the wearable affect the user? What do we all need to work on between now and the end, and how has that been for you guys getting that feedback back?
[24:02] Your software team that are usually totally used to, oh, you want me to fix that? All right, let me figure this out. We'll run some internal regression testing and push the software update. Maybe it makes it worse, maybe makes it better. But you can always push a software update. You
[24:15] know, we call those recalls and those are tough.
[24:17] Yeah. It's it's been interesting going out and talking with the clients. I was just in New York last week. We were we you know, we were out there demoing to a few different, few different depots that they have that will be getting these as we roll these out and hearing their questions, hearing their feedback on the hardware.
[24:39] A couple of interesting things. One was I actually looked at the date. It was about a year ago that we actually went out to this site and did a demonstration of the POC. So while we say it was long to them, they were actually quite amazed that like, hey, you guys took something that, you know, you showed us the concept and reality in almost well, it's less than a year, but just about a year.
[25:02] You brought it back out here in a full production state that we're going to be using day to day to help us. So that was that was pretty amazing to hear that feedback. Like you do say though, when it comes to the hardware, right. Like, hey, we wish it was louder, maybe vibrated harder, but other than that, hey, it served its purpose like it's it.
[25:19] I will wear it. It is going to protect me right now. That's awesome. Yeah, that is great to hear. And you know, they've all got some ideas. Some wanted to wear a wearable around their neck that had. Yeah. Joked could also put a little fan in there to keep them cool in the summer and hot warm in the winter you know.
[25:39] So it's really interesting to hear their feedback. But you get there by in because they're like, hey, I understand this. Like what your the purpose of this, you know, hardware is and now that I'm bought into it, great. I just want to thank the different ways I can use
[25:51] it. So and I think that I think it's the great thing about this quantity specifically of doing this kind of middle quantity, not thousands and thousands of devices, but more than just you can kind of show them at the front of a room and say, hey, look at this.
[26:01] This is cool. Yes, this is enough that you can you can hand it to them. You can leave them for weeks, let them use it on their own. And you get the the types of feedback that you would never get from just a demo in front of a, in front of a conference room or something. It's so true.
[26:15] The what's the word? It's the I live with this product first as I saw it. Yes, I saw a demo. That thing was really cool. Like they had to reset it once, but it was really cool versus oh no, guys, I can't stand this thing because every time I charge it this one way, it doesn't charge and I don't know that it's not charging.
[26:31] I'm just giving some examples of other products we've had of like, yeah, if the charge indicator isn't really obvious, people think they're charging even if it's not. And if you'll learn that in this volume that they can take it home, they can live with the pain, right? Because friction and pain is the only way you make a product better.
[26:46] Yes. Very infrequently. Do all of us sit around a table together and just think about how do we make something really cool that never hurt anybody or didn't have a pain point? Yeah. That's the that's the fun part. And so, Tim, in the actual application of like the vision system working in real time, how has that been for you guys to pilot from proof of concept to pilot hardware on like the software hardware integration side?
[27:10] Well, I've spent a lot of time in the parking lots near our office, driving my coworkers Tesla at questionable speeds towards towards cameras. Yep. But it's been a lot of fun. No, it's been a really interesting learning experience for me to think through the hardware and real world side of the implications of our software, the logic that needs to exist to filter out false positives.
[27:36] As we were talking about earlier, the implications of every connected part from our actual software to, you know, the Wi-Fi HaLow access point to how that sends the message out. And really, my goal at the end of the day is kind of getting the the must haves as the system working in as quick of a routine as possible.
[27:56] When we first started this, we were budgeting, just kind of guessing like, hey, we want to get this within about 700 milliseconds or something like that. And just through continuous testing on every single part of the software and hardware line from the hardware that we own, which is the edge devices, the compute that actually runs the inferencing to the software that we own, the logic.
[28:16] What trade offs can we make for tracking that speed up or slow down? How long we need to wait until we know an action is true, to like your guys's side of the WiFi, HaLow and all that stuff. We've gotten things down to, you know, 4/10 of a second. So we calculate or assume that there's a budget of about 2.5 seconds from an event occurring, something needing to alert someone, and then that person needing to get out of the way.
[28:41] Right. So keeping our number as low as possible, that 4/10 a second, is kind of our main goal. We're there now. It feels great to do that. And it's just been a great learning experience for me of, you know, it's not just a website or a dashboard or something that's monitoring a manufacturing line or something like that.
[28:56] This is a whole interconnected system.
[28:58] So. Well, I think the the other part that adds for both of us, both on the hardware side and the software side is there's this magic line when you cross between a convenience product and like a critical product. Yeah, this is can this product hurt somebody or can us not doing our job allow someone to be hurt?
[29:13] That is the whole purpose of preventing harm. This medical device, the safety of the things that the tone in which you look at, like DFM or failures, changes in all of our heads. Right? Hey, how acceptable is this risk? And it just changes. And I think that when Glassboard first started doing medical device 7 or 8 years ago, it was such a culture shock for us, right?
[29:36] Because it was always most clients were startups when we were young. It was moved fast and break things. Shoot from the hip, just go. Yeah. As long as it works, it's good. Like they can reboot it. Yeah, like you can't reboot that pacemaker. That's tough. That's. You can't do that one. Yeah. And it it just gets you to think about product in a different way and think about testing in a different way.
[29:54] It's both the failure mode and how do we test for that failure mode. And I'm just curious on your guys from the software side, have you experienced that throughout your journey, whether this project or another one of like changing console was actually usually been pretty good at self tests and, you know, testing your revisions changes the creation for us, it's usually harder to test.
[30:13] How have you guys approached that with this product? Yeah, I think I'll answer your question. We'll see. You brought up a couple points that I think are interesting around this, but the gravity of this use case, this is people's safety has really, you know, hardened our perspective on a lot of this one from a testing perspective with just the operators themselves.
[30:34] In a lot of our past work, we're working with manufacturing clients. And there's a couple stakeholders, the client manager who just wants this extra data point, and maybe a data analyst who's going to consume our stream of new day. There's something like that. Now we're dealing with tens, hundreds of people who will be using this for their safety.
[30:52] And that just informs the whole testing system. And, and, and really helps us quite quickly and easily say, what are the absolute must haves? This must beeps, blinks at 1300 feet, two miles, something like that. In good conditions. You know, it makes also the operators. I again, we've worked with plant managers who are excited like yay, new data.
[31:18] But having hundreds of people saying, I don't care that maybe this is going to could be a little smaller in the future, it could be a little louder in the future, isn't going to protect me now, and I'm excited about that opportunity. Everyone's been so receptive. I mean, again, I've had great, great experiences with a lot of our clients we've worked with, but I've never had so much reception from, you know, the actual day to day people using this of just this is great.
[31:42] Let's, let's see how we can move this forward as well.
[31:45] So I don't really know if I answered your question, but. Oh, you totally did. You hit what I didn't ask, which is how much more rewarding it is. You get to actually see an end user, not just make more money, be more efficient, or have a cool what is it like?
[31:58] There's fashion, aspirin and medicine. Like the three ways to pitch product, right? You're either fashionable aspirin, like you make my life slightly easier, or this medicine you saved my life. When you're building medicine products, whether it's safety or actual med device, like it is a different experience with your user. I think that is like really rewarding. You and I experienced that and a couple of other things we've worked on.
[32:18] Absolutely. We've had some really cool aspirin products that we love because fun. And they're great. Yeah, but it's different when it's like this level of safety. Yeah. I remember specifically, I think, Derek, you made a LinkedIn post when we first put the demo unit out. And I remember the top comment was like somebody I didn't know, but it was like my cousin got killed while he was on a roadside construction.
[32:37] This would have saved his life. And I remember seeing that and being like, whoa, this is like, I knew that it was a thing. But seeing someone's testimonial in there to say, I wish that that somebody I loved had this product because they would still be here. That's awesome. And it's awesome that the size of team that's able to make this happen with today's technology.
[32:56] And this is at the end of the day, what's amazing is we have a handful of people at Glassboard that work on this, right? This isn't a team of 20, right. This is like what for. Yeah. Like for for really smart people in a room doing electrical and mechanical industrial design and firmware. Yep. You guys have a team of what how many people are actually on this team?
[33:15] It depends on the week 5 or 6. Yeah. Again, I'd probably say a couple pizzas. You could probably feed the whole team between
[33:23] Glassboard and yeah, that's been able to build and develop this entire solution. Right. Yeah. In a timeline. That is really cool to see from concept proof concept where you guys picked it up, where we got to do the first demo of the tech, got to make the first unit.
[33:36] Now we're actually launching build batches for for your team. And I think this is just a really cool testament to how awesome the shoulders of giants have been for all of us to stand on. Right? You guys didn't code the AI algorithms from the ground up. You use technologies that exist. Same with us. We didn't make the microcontroller.
[33:52] We didn't write the course off or library. We imported a bunch of drivers and put them together. And modern engineering is a lot of both software and hardware teams to do things in incredible pace that I just think is made.
[34:05] Yeah, it's it's you know, you mentioned. Yeah, I saw that post. It's amazing when I've gone out and talk to people about the solution.
[34:13] I've heard that story way more than I ever would have thought of. Yeah, I'd probably count count, unfortunately, in two hands, the number of people that have had a very similar story when I've talked to them about this. So, you know, it's really gratifying to hear their feedback, to say, hey, I'm glad somebody is doing something about this.
[34:31] And you said, it's amazing the pace we were able to do this. It wasn't like, hey, ten years, we might have something that helps you. This is hey, we've got folder technology today. Let's put it together and let's roll it out. Yeah,
[34:41] yeah. And I think the the other neat thing about all of this is the reception, not just at the end user, but of the call, the end adopter.
[34:48] Right. So just for everyone listening. It was a good lesson for anyone hard for like your user isn't always your customer, right? At the end of the day, they're very separate stakeholders when we're developing products and companies. And so who is the like the actual customer for you guys? Who is the group that you're having to reach out to?
[35:05] Let them know the technology exist? Like what is this person versus the person who is on the job site on the road probably isn't the one who's writing the check for this device.
[35:13] Yeah, an interesting point on that. I'll let Derek answer on specifics, but you're totally right that the buyers are different from the users in as far as what they actually want as well.
[35:21] We've heard from some of the buyers who kind of run the safety orgs, rather than people on the side of the highway who say things like, you know, actually, we would almost just buy this without the wearables for the data that we get on, how many intrusions we get throughout the day. If our code setups are particularly safe in one way or another, if ghost cop cars reduce the speeds of car driving by, things like that, and then it's a huge added bonus that there's also a wearable that can alert the workers as well, right?
[35:48] So that's the difference really between the purchasers and the users and also the users at HQ. They're using it for a different purpose and whatnot. So just to point out, like it's a pretty wide platform on, as Derek mentioned, even with the manufacturing stuff, we get data around the whole thing so that you can look at what happened, not just if there are events, but, you know, get data for potential policy making around, as I mentioned, you know, ghost cop cars.
[36:11] Hey, we saw that that reduced the speed by ten miles per hour on average when we were able to, you know, rent those out from the station or something like that. But yeah, we can speak to kind of who you think the market has been. Yeah. You know, there's a there's a very broad industry wide concept. It's called Vision Zero.
[36:28] The whole idea of zero deaths on the roadway. Right. And what is everything that everybody can do to try to achieve that? That's you know, there's almost a thousand work zone deaths per year, US raising. Now, not all of those are workers, but due to the fact of, you know, if there's a soda on traffic, people aren't paying attention.
[36:48] Right. So luckily, there's a lot of people doing a lot of really good things to try to make all the roadways safer. You know, we are a piece of that. We are a component of that. And so when we reach out, we talk to clients. We're talking to, you know, roadway authorities, tollway agencies, contractors who are out there.
[37:08] And really, most of the time it is their chief security officer that's looking for new ways to improve the safety for their team. So that is usually the person who's got the budget who can talk to us and, and
[37:21] has has the users underneath that, that also going to support it has the authority and the mission to protect the users.
[37:28] That is their role, that is their job. And I think it's it's always fun. Tim, you nailed my part that the buyer and the user and always the same person and they sometimes want different things. And my my favorite conundrum and building good product is distilling out of both of their desires. What are the needs for each of them?
[37:47] Because it's 50% need and 50% desire usually, oh, I need it to do this, but it'd be really cool if something happened and you can find out that that really cool doesn't move the purchase price or the desire to purchase at all, even if you get a ton of this feedback. Same with the user. They're going to say, man, I wish it, you know, make me bacon and eggs in the morning.
[38:05] And they all wish that. But it doesn't affect whether they actually use it, whether it actually makes their lives better, what actually makes them safer here. And I think that's the hard part in being like the taste makers in product development is how do you discern even desire from actual purchase price want.
[38:20] Right? I feel like that's one of the biggest things I've learned.
[38:24] Just working on a lot of our consulting side generally is just people have a lot of ideas, and we kind of have to lead them to, okay, which one is actually makes sense
[38:32] to help you out. And what's interesting is when we always do that, like you and I have this role, it's almost always math. Hey, you have all this emotional want.
[38:41] Let's go try and put some numbers to each want. How much does it cost to develop? How much does it make the bomb cost or the product change? How many users actually want it of your end users and people from your customer base, not just in your company? What are they willing to pay differently about it? Right. This is effective purchase price.
[38:57] Does it affect all these things? And it's the nerds in the room having to use math to go emotion math into an equation to make a spreadsheet. That's all I like. Does this return on investment either in time or money or or advertising? Yeah,
[39:11] right. In our world it's a lot of the time, like pay for accuracy, kind of like not really, but kind of it's like if you bought a better camera sensor that is inherently more expensive, we would,
[39:20] you know, get more pixels to look at or better lighting or anything like that, or it would have more frames per second output for you to interpolate between or run inference on or whatnot.
[39:28] I need to run this on a tiny GPU. Or can we get the one that makes sense for this scenario? And. But at the end of the day, it's, you know, there can be an issue sometimes with people wanting to over science project something, and it's totally reasonable for a client to say, hey, we want 80% accuracy and we're willing to pay this much for it, and we can kind of get them there specifically on that.
[39:49] So like you're saying, reaching that middle ground of what's possible, what does it take to get there is,
[39:54] I guess, what all this is, it's the art. Yeah. Right. In all the science that we all do, it's the actual art of figuring out how good is good enough, because there's always V2. If you make v1 good enough to get to market, you always have enough time, money and effort to get to V2.
[40:07] And I think when I started in, in this field, you know, I thought a lot of my time would be spent coming up with cool ideas and convincing clients that they need them. A lot of it's actually spent convincing them they don't need the cool ideas they came up with. And coming up with this, this what is the I mean, we use the term minimum viable product that you can actually get out to, to your customers, especially at this stage you guys are at what's the minimum thing we can get in front of the people who are going to use it so that we can get
[40:32] feedback?
[40:33] Yeah, that's one thing. Working with Glassboard throughout this whole process, specifically myself. I know I've had to make decisions. When it does time come, time for cost and money to get to where we are today, right? There was times we made those kind of asks and you had to tell us, hey, that's we
[40:49] should probably not do that.
[40:50] Yeah, I do that. And there was other things that you said, hey, you should think about these things. I mean,
[40:55] I know specifically one was, hey, what about again somebody relying on this for for them to save their life. The last thing I want them to do is be relying on it, and then they're outside the range and they don't know that.
[41:07] Yeah. So one of the you know, I think that, you know, as through the process, one of the things that brought up by your team is we need to make sure that, hey, when batteries low or their upside of range, this thing starts beeping in a way that's not like the alert, but in a way cut. I think we used it as the smoke detector.
[41:23] Simply start giving you a little
[41:25] chirp to let you know. Bring your attention to it, to say, hey, you're no longer connected. I'm unhappy. You can't listen to me. I'm not going to be useful. Yes, please get back and range or know that you're going to have to watch for cars. I can't do it for you. Yeah, that's so much fun.
[41:39] Well, I think the the neat thing about all of this is we're building a product that people are currently like today using, which is really neat. But you guys have built a technology platform that has other applications, like this is one of those moments where, yes, someone built a vision system in the in the early days and discovered computer vision and AI vision, and then you guys would applying it to all these use cases.
[42:00] This is one of those use cases where you had to build a complex version of this that detects this kind of motion, these kind of like cars and this kind of space. And you're starting to see other applications, right. So I think that our construction safety is the first place you guys want to tackle this. What are the ones that you see coming down the pipeline.
[42:19] That would be the V2 or the V3 of this kind of a
[42:22] platform. I mean, even in the same vein, on construction safety, we were out in the field with some guys recently who it was really gratifying to see how quickly they intuited how the system works. And they're like, oh, well, we could put one of these cameras on a backhoe with a swinging arm and make sure a person doesn't walk into the zone.
[42:39] I was like, yeah, you're exactly right. We're detecting people as well. We can config swap and say alert if a person enters or anything like that. So first off, like that's a separate zone that this works in the same space. Yeah it's in customer base just different applications, same customer base. But also it was really exciting for me to just see like like I work on the side of the highways like they're talking about.
[43:00] We don't understand this stuff at all. And your AI vision stuff, you know, the kind of joking with us.
[43:04] Yeah. Well, who was the wizard in the box actually pushing the alarm button? Yeah, but also, they understand it quickly enough to be like, oh, I could just
[43:10] draw this zone on whatever camera I want. And then the obvious kind of fallen from there is service workers, like traffic stops in a police car or something.
[43:20] Same idea. You have a camera pointed backwards. Potentially. They already have rear facing cameras on. There may or may not need the wearable device on the officer. If you can tap into
[43:28] the radio like they're wearing a radio live right now, they have lights. They have sirens already up there. So that vision system, it's the same
[43:35] logic. We've already built same platform from the edge, compute all the way up to where we need to send the message, and then unplug the Wi-Fi HaLow for the wearable device.
[43:43] Potentially. Yeah. Love it. Which I mean, I think we might have mentioned, but like, you know, we started with the trucks specifically with a siren plugged in. Then we added the wearables and whatnot. So we already have the siren component add the wearables. It's really just much a Lego pieces. You can kind of plug
[43:58] and play in different combinations.
[44:00] That's just engineering. If you zoom all the way in like a microprocessor, it's just a bunch of light switches together like Legos in a very complex way that we, you know, we we arrange the rocks in a way when we hit with lightning, it things amazing. But when you take engineering all the way down, it is just a bunch of small Lego bricks that all of us by pre-assembled in some level of block.
[44:20] And then we snap those together into a more complex block that we then pass on to the next person, and someone ends up being the end user, right? Yeah, but that's the coolest thing about modern engineering is very little is ground up. Like truly, truly, from the ground up, everything is on the shoulders of giants. Even if we build a product that is fully custom.
[44:37] It's designed with software that was built by someone smarter than us, and all this humanity collective group think is what makes cool products half of the day, especially with a couple of pizzas feeding the entire team. Yeah, yeah.
[44:49] I mean, I'll say like from my perspective, like vision is super cool. It's really exciting software space to work in.
[44:54] But at this point a lot of vision models are to this point, you know, commoditized, where pretty much anyone can go train a YOLO model. Every person I interviewed for an internship has trained a little detection model or something like that. But to your point, it's how you connect all these pieces and come up with a final solution that ties everything together.
[45:10] That's kind of where the sauce is. And I mean for the wearable specifically in this scenario, that's like a in our view, a huge differentiator, right? Like if the vision is commoditized, you know, anyone can go train a computer vision model to go watch a car potentially enter a zone. Now we do, I would say, have expertise around what does doesn't work, speeding things up.
[45:29] And, you know, maybe I've just been doing it too long and I'm used to it. So but being able to work with you guys on the hardware side is really opened my eyes on
[45:36] the iteration side from that. Well, I think the, the what, you were actually just eliciting that you feel like what we do is black magic.
[45:42] We feel the same way about how you guys manipulate software. It is the awesome world we live in. That 0 to 80% is really easy today. The tools will work with makes getting 0 to 80% of a thing just awesomely fast, right? I think your first demo, your first proof concept, both the software and the hardware that took to get there.
[46:02] But the difference between that and a product is the last 20% of functionality is 90% of the effort and the expertise, and everyone having their 10,000 hours in their craft. That makes it a wonderful product that actually wants to use every day and makes it a business.
[46:16] And to thinking about, I don't know, just thinking about the users, think about the product as well.
[46:21] Like, I came into this world as just a nerd who likes solving technical problems. And I've learned a lot from Derek, specifically around, you know, not everyone just wants something because it's clever or unique or interesting. You have to think about who's using this, what it does. I know you came up with, you know, Devin jokes about breathing LEDs or whatever.
[46:40] Yeah, silly things like that, where it's like we can do clever stuff as nerds with our software and hardware.
[46:45] But what does the actual person actually want at the end of the day? Yeah, well, it takes all kinds, right? We need the business people to make a business case. So there's enough money to have enough time to go to the creative case, and it takes both of those things together to actually make a product that solves a problem or gets to survive into the world.
[47:04] And then eventually, it takes the marketing people that put lipstick on it and make sure everything's pretty and gets it out in the world and goes from there. But I this has been such a really fun conversation, going about how do hardware and software teams work together? How do all of us make something from nothing? I think the what I want listeners to take away from the episode is there are all these small steps, all this attention to detail.
[47:26] That is what's differentiator between here's Michael Demo and what is a valuable product or a valuable company at the end of the day. And that last step is those valuable products or companies end
[47:36] up making users happy great or safe. Right. So that's I think that's an
[47:42] amazing episode for us. Derek and Tim, thanks for joining. Thank you for having us.
[47:47] Thank you and everyone listening. We will have not to bury the lead, but to bury the lead. We'll have much more cool content coming out about this product and these teams. Shout out to Altium is working with us to make some really cool content outside of just these four walls, this podcast studio. So stay tuned for more. And if you ever need help working on these kind of problems, let let us know.
[48:06] We can hook you up with your
[48:07] object team. Take care everyone.
