Michael Fleischman — OpenSpace is Reality Capture Plus AI

We talk with OpenSpace CTO Michael Fleischman about turning job-site photos into spatial data that teams can actually act on, from 360 degree capture to progress tracking and AI agents. We dig into why construction software adoption is so hard, and what changes when your phone can be used to create higher-quality data and automation. 
• Michael’s path from philosophy and psychology to computational linguistics and AI 
• Meeting OpenSpace co-founders at MIT Media Lab and pivoting into construction 
• Reality capture as the foundation and why “agents need eyes” 
• Cameras vs LiDAR on phones and why photos solve most field needs 
• AI autolocation as indoor GPS without beacons 
• Lowering friction as the real key to construction tech adoption 
• Progress tracking against BIM models or 2D drawings and closing the loop with reality 
• Disperse acquisition and why construction-specific spatial understanding matters 
• Flagging potential issues with Spotlight and keeping humans in control 
• Letting customers build customized agents for safety, QA, and workflows 

Transcript

Welcome And Quick Introductions

Roopinder 0:07

Hello and welcome to FoDES, the Future of Design and Engineering Software Podcast. My name is Roopinder Tara. On the show, we will have guests that will discuss tools and technology that engineers will find interesting and useful. Hi, hi, Michael Fleischman.

Michael 0:26

Yes, nice to meet you, Roopinder.

Roopinder 0:28

Hi, nice to meet you, Michael. Thanks for uh coming on the show. Um absolutely I'm uh excited to be here. You are uh I think our paths may have crossed. You were at two, I believe.

Michael 0:38

Um I was. I was uh in Nashville.

Roopinder 0:41

Nashville.

Michael 0:42

Yes.

Roopinder 0:43

Are you in the San Francisco area?

Michael 0:44

Uh yes, I am in the Bay Area, actually about uh an hour north of San Francisco. I live uh up in Sonoma.

Roopinder 0:51

Oh, close to neighbors then. I'm in a little bit less than that north. I'm in Novato, just north north of Bridge. Very close. Yeah.

Michael 1:00

Do you go to the office often or do you just normally work out of the um I you know we're a fully remote company, but we do have an office in San Francisco and headquarters there. And so I'm I'm in the office probably a couple times a week.

Roopinder 1:14

Oh, okay.

Michael 1:14

So I I get stuck in traffic right in your neighborhood, you know, at least a few times a week.

From Philosophy To AI Research

Roopinder 1:20

Sorry, sorry about that. The Novato Narrows is it's infamous. And they've been working on that for uh, I don't know, as long as I've been in Novato, I think they've been working on that issue. It remains to be resolved. Yeah, but that's a heck of a commute to do it every day. We need to talk about what you guys are doing in open space.

Michael 1:39

Yes, happy to talk about open space.

Roopinder 1:41

Looking at it, it sounds very impressive. You've done a lot with the company. It's uh eight, some years old now.

Michael 1:46

That's right, yeah. Um, about eight years, uh, depending on when you start. But yeah.

Roopinder 1:50

I want to go a little bit further back to your history because it's very interesting. Okay, so I'm stuck doing a little bit of research. I don't do a lot of research, but I look at LinkedIn and I'm saying that you okay. Now wait, I gotta understand this. You went to you got your bachelor's in philosophy and psychology.

Michael 2:07

I did, yes.

Roopinder 2:09

Two years later, you did a like a hard turn and went into computer science. So this is the first time I've seen that approach. Yeah, you know what happened there.

Michael 2:19

Yeah, um, you know, it made sense at the time, I'll tell you that. And it wasn't so much a of a hard turn, actually. Uh so I did do my undergrad, was more of a liberal arts uh major of philosophy and psychology. Um, was very interested in uh language. That was as kind of one of the themes. And so the degree I did after I done a master's degree at USC in what was called computational linguistics. Um that was kind of the bridge, and that was an interesting program, kind of a uh half computer science and artificial intelligence and half formal linguistics. So I kind of came into it from that side, but got very just kind of attracted and really uh enrapped by the the AI side of things. Um, and so spent most of my time working on AI problems there in the language space. And then so I went uh ended up going to grad school for a PhD. That actually was kind of started on the cognitive side, science side of the world. I, you know, was interested in learning how children learned language, but that kind of morphed into how machines could learn language and eventually brought in the vision side as well. So that's kind of where computer vision came together because vision and language really, you can't really have one without the other in many ways. And that's actually a theme that continues to come up in in open space as well, but was also part of my uh, you know, I started a company before open space coming out of grad school, which also was a bit of a turn that was uh called a social TV analytics company. Uh so nothing to do with construction, but what we did there was we used computer vision to analyze what was happening on television, like TV shows and advertisements. And then we used natural language processing to mine what was being said about those shows on Twitter and Facebook and trying to uh really kind of create this data set of of what everybody thought about what was on TV, which was not not positive, I will say, uh you know, at the time, probably say the same today. But uh but yeah, so kind of twists and turns, but you know, when you look back on it, you can tell a story.

Roopinder 4:21

It does have a thread. And then you went on to MIT and got your PhD in the U.S.

Michael 4:26

I did, yes. And that's actually uh actually uh so that's where I met my two co-founders at OpenSpace, actually. Um so Jeeven, who's our uh our CEO, and Philip, our chief engineer. Uh, we were all grad school buddies at the uh the Media Lab uh at MIT. We met, uh I I hate to admit it, but uh over 25 years ago, you know, our first years there. And uh, you know, we became friends and kind of tracked each other as we were going through the program. Um and then actually, you know, after we had all left and kind of gone our different directions, Gen had also started a company at the same time that was then bought by a drone company called 3D Robotics, where he became president and was very much focused on the construction use cases for drones. And so kind of around that time, actually separately, Philip and I had started prototyping some kind of uh stuff with 360 cameras. Actually, initially a very different use case, nothing to do with construction. But I was, you know, it's funny, at the time I had a real problem. I was trying, I was moving from uh Massachusetts back to California, and I was trying to find an apartment remotely. And I realized that every time I would look for apartments online, I'd see these tiny images, and then I'd fly across the country, get to the actual apartment, and it would look nothing like the picture that I saw. And uh, and so we we thought, well, what if you had, what if you every realtor had a 360 camera could just take a video and then create a kind of Google Street View for the inside of buildings, and then you wouldn't have that problem. And so uh one day, uh, you know, I got a LinkedIn notification that Jeeven had just left his company and uh called him up and talked to him about this. And he said, Hey, have you ever thought about doing this for construction? Because that that might be a great use case. And you know, really hadn't thought about construction before, but Jeeven and I kept talking. We got on some job sites together. I actually started to do a little internship on a job site, uh uh unofficial internship, and uh and then uh yeah, it kind of took off from there.

Roopinder 6:26

Wow. So okay, that's that is interesting. That your introduction to construction is normally not a something that uh is associated with uh MIT, for example, let's just say. But uh yeah, because I mean but it is an industry. It must it's a smart move as it turned out, right? Because what needs technology more and what is more devoid of technology than construction, right? It is it just uh it's desperate. I see I I see Autodesk like uh so frustrated with the lack of implementation of their own technology in in the field that they're selling into that they offer it and it's like the industries are just scratching the surface with technology, right? So a huge, huge gap that needs to be filled. It's just incredible. So okay, so now let's see what open space actually does, because it seems if you look at a website, it seems like it can do everything, right? But in in essence, it's reality capture, correct? With hey, sorry to put it so succinctly, so there's a lot more to the story, but it's reality capture with intelligence that can help you make business decision or decisions about the project, right? Is that correct?

Agents Need Data And Accountability

Michael 7:42

Yeah, you know, uh absolutely at its foundation, um, and very much where we started was reality capture, and then really have expanded from there into kind of like you said, intelligence on top of the reality, and then even more than that is taking action and how to kind of support our our customers and use that data to take actions. And now, you know, really with the the kind of agentic revolution that's happening, um, more and more it's about agents taking action on that data and and um giving them the information they need to make decisions.

Roopinder 8:13

I've heard uh Jensen Huang, you know, is yes, I do. So he says soon managers will just be managers of AI agents. Yeah. I'm sure he said it much better than that. But yeah, we'll all be leading agents and they'll be like for every engineer, every construction manager, foreman, whatever, they'll have instead of a lot of humans, or they always say in addition to humans, I suspect that's not the case, but in addition to humans, they'll have agents as well to do their bidding. So has open space become that like making agents now?

Cameras LiDAR And Smartphone Capture

Michael 8:48

Uh uh Yeah, absolutely. I mean, agents I think are a big part of the future. And I, you know, I I think I agree with Jensen in terms of that there'll always be a need for humans, uh, for sure, because you know, for various reasons. Uh, but the idea of humans managing a team of agents along with other humans, I think, in general, I think is very much true. And and yeah, open space is is really leaning into that direction. And, you know, we very much believe that just in general, AI is only ever as good as the data you give it. And that's doubly true with agents who are going to be making decisions and and taking actions on your account. Um sometimes kind of we use the idea of you know, what we're doing is giving agents eyes. We're letting the agents actually see what's happening on a job site so that those decisions can be informed, which is you know incredibly important. But the role of the human is is vital in that because in the end of the day, it's the human's responsibility for whatever those agents do.

Roopinder 9:45

Right. I like that idea of the eyes giving a instruction or data eyes, because really you're visual vision-based, right? Cameras. Do you also use LiDAR or radar or any of the other things, or is it all visual based?

Michael 10:00

It's it's primarily vision-based, yeah. Well, the um the platform supports the uploading of LiDAR data. And so you you can do that, but for the most part, um, you know, it's cameras, 360 cameras, drones as well. We introduced capture last year. Uh, but actually, more and more of the focus for us is actually kind of on the phone. Um, you know, the truth is that I think still today the majority of reality that's captured is is on the cell phone and the smartphone. And so we've invested a lot this last year introducing products for the field that are really focused on uh what you can do with your with your smartphone and capturing that reality, contextualizing it spatially so that when data is captured on your phone, we know where it is, you know, and then you know, organizing that uh up in the cloud. And so um, so yeah, I think all of those components for sure.

Roopinder 10:53

Got it, got it. So this is uh so I love the idea that this has LiDAR built in, right? These smartphones that have LIDAR, like the iPhone higher end models. Yeah, they're not very good LiDAR, but it's not bad.

Michael 11:07

You know, it's bad.

Roopinder 11:10

So you have you have LIDAR in your pocket now, and uh, wow, that is so cool. Why isn't everybody using this? And uh I I've used it in the manner that you suggested. I've like gone places where I needed to show somebody something, and rather than capture the video, I capture it with my LIDAR application that uses LIDAR. I've used a couple of these, but uh yeah, capture the whole model and it converts it into a 3D space, 3D picture you call it, and uh it's so much better, right? I've done that with uh places I've stayed at and offices I've been to just to show people that this technology exists and it's in your pocket, right? So I totally get that. That this you know, why have a hundred thousand dollar lidar scanner? Why rely on that? Why make that your choice of technology when everybody's every engineer, every person you've got has a lidar in their pocket. Not great, but but now where do you well where where does open space come in? Do you enhance or make that image more accurate?

Michael 12:13

Yeah, well, we we uh we actually do have a I like what you're describing, a feature that lets you use the the lidar scanner on the phone uh to capture LiDAR. But like you said, it's it's you know it's it's uh not perfect. It takes a while to capture a large space. And so when I actually talk about the phone, it's actually not so much the LiDAR capabilities of the phone, but literally just the photo capabilities.

Roopinder 12:35

Um is the video from there?

AI Autolocation As Indoor GPS

Michael 12:37

Yeah, video and photo, and you know, it's interesting. Like LiDAR certainly has its place, uh, you know, every tool for every job. But what we find is it's not so much that that's not useful, it's just you actually get so much value just out of the image itself that 80, 90% of the time, all you really need when you're documenting an issue on a job site, a punch item, safety item, something like that. Uh the photo is is all you need. So uh one of the technologies that we've introduced last year, which is actually pretty, you know, it's it's kind of probably one of the biggest new breakthroughs we've had is something called AI autolocation. Um and what that does is essentially it's like a GPS for the inside of a building, but it doesn't require satellites, it doesn't require Bluetooth beacons or any hardware to be installed. It's essentially based on the 360 capture technology, but it works just if you have your cell phone on you. And so as you kind of move around the space, the system is essentially estimating where you are on the floor plan on the job site. So when you take photos, um, it can essentially know where you were when you took that photo and localize it for you. This is you know kind of one of the first use cases for that. But really, what it introduces is this idea of the the spatial, the phone having a spatial awareness of you. Right. And what that translates into is the agents now knowing where you are on a job site. So that now when you communicate with an agent through your phone, it can know what unit in the building you're in. It can know, you know, are there issues that are associated with that unit that you should be looking at, give you the context of where you are, and really start to kind of change the game in terms of what these agents can do. More and more like having a little PE on your shoulder that can you know help you with any task that you might have.

Roopinder 14:25

Oh, okay. Hey, so help me understand this because this is a kind of a basic point, but I don't know that much about the technology. It seems to me that if you have a LIDAR, if you have LIDAR in your pocket, like with this phone, or like you said, you can use also the visual, the vision, the count the uh photos and movies. Seems to me that if you have the lidar, why not use that? Because aren't you making the system work harder to take images and convert them into 3D points when you already have 3D points? Aren't you taking an extra step? Or what's the advantage of taking that extra step?

Michael 15:02

Oh, yeah. No, if the LIDAR is available, we'll take advantage of it. That's kind of policy. Every type of sensor that we have access to, we'll take advantage of. And so if we have access to it to do AI autolocation or uh scan, well, the system will definitely use it. Um way that we design these models is essentially to be very robust to uh different sensors we're knocking out, right? So if you don't have access, if you don't have a camera or a phone with LIDAR, which still the kind of majority of our customers don't have LiDAR on their on their cell phones, it's still somewhat of a uh a premium feature on phones. Um didn't take off quite as much as we had thought either. But uh but then the system will still work without it. Right. Obviously, the certain things are going to be required for that if you need really centimeter accuracy or millimeter accuracy for measurement of a door, um LIDAR might be necessary. But again, the vast majority of use cases for us, that type of of kind of millimeter accuracy is not necessary. And so yeah, but we'll we'll use it if if we have it. Oh, we will. Okay, good, good, good to know. So I won't I try not to dwell on this too long, but it but since I made myself a pain in the in the ass to people saying, like, hey, why don't you use a LIDAR in your pocket when instead of taking measurements, I also found it didn't take off like I thought it would.

Why Construction Tech Adoption Is Hard

Roopinder 16:24

I thought, boy, I can get a map of this room in like 10 seconds, yeah, under a minute for sure. Uh and and and I see contractors all the time measuring rooms on their hands and knees, you know, using tape or lasers, right? But usually not. And it never it never took off. I can't understand that. I've been proselytizing this, and then nobody's listening. You must have felt the same way, right? So Apple is probably wondering what did we do that for? Like nobody's using it.

Michael 16:56

Yeah, you know, it's funny. I mean, I think I think Apple probably it has a lot of uses in things like you know, AR and VR. And so I do think there's use cases that are kind of more under the hood that you don't see, you know, although you you could say the same argument about why is AR not taken off more than it has, right? And that's a very different conversation. I think probably part of it is there are just limitations to those Lidar systems, like the range is not as good as you get off of like a dedicated like laser measurement tool. And I think also just usability, like it it's still gonna be to do a measurement, to take out your phone, a number of clicks to get to the right area. I think a mindset where you just have a dedicated tool that you know that is that's not so expensive, right? Um, that you can use, I think is just gonna be a natural fit for folks. So, you know, it's it's funny you bring it up because adoption in general, that's as a software company, it's something we think about a lot and was the really the focus for us uh in particular in our our you know early development of the product. Because getting, you know, folks getting folks in construction really anywhere, but particularly in construction, getting them to adopt something new, you have to both really show the value right away and you really have to lower the barrier of entry. Like it it cannot be um it needs to be something that you can just pick up and use without having to read a manual or or go through a lot of steps. So, you know, I I appreciate the need for the kind of the design to to match the value in the use case. And otherwise people just you keep using what they've been doing before.

Roopinder 18:31

Exactly. Now you're a CTO, you're making all this fantastic technology. I'm covering it best I can. You know, I I fall in love with technology when I see it, and I just want to shake some of these people that, you know, for example, came came to our house and you know, showed me late unrolled blueprints on my kitchen table when all the time I'm expecting, hey, I maybe I should be putting on goggles, you know, VR headsets to look at this rather than blueprint. And what have you guys been doing for the last 30 years if you haven't been embracing any technology whatsoever? If I remodel my house was like an exercise, and it was so frustrating because nobody even showed me a single 3D image of my that they had, and it was and you know, as you I'm sure you could sympathize. It's like, oh my god, why why won't you? What does it take to get you to use this? Do I have to put on a hard hat and pretend talk their talk or you know, carry a hammer on my belt, or what why why aren't you listening to us? Is it because we're we're too techy looking or or we don't speak their language?

Michael 19:36

Or you know, I mean, uh adoption is hard. I I will say, you know, it takes a while for things to kind of filter down, I think, um, you know, through through the ecosystem. But you know, I'll also say that that uh I think the construction industry was really kind of underserved by technology for a very long time. And so, you know, I think there's uh a lot of reasons to focus on the the user, the consumer in terms of adoption. But in general, construction is such a physical industry, right? And so the digital transformation there, the kind of rules of the game are just different than office work, you know, where where kind of I came up, right? Or or maybe you did, where you really and you see this in the kind of history of construction technology adoption is that you really, you know, it was very slow going up until you get new technical breakthroughs. Like in the early days, it was, I think, really the you know, mobile computing and cloud computing that that had the kind of first big unlock that you started to see. So people can now take iPads to the job site. And now the idea of not unrolling a blueprint, but just looking at something makes a lot more sense. Whereas you're not going to run back and forth to the truck to look at something on your PC in 1999. And so I think you you start to see those transitions. And I I believe actually that we are kind of right in the middle of one of those transitions now as well, with the kind of combination of computer vision and AI in particular. I think that really both of those, when you combine them together and you start to have the AIs with this kind of physical understanding of what's happening, which is another, you know, Jensen uh meme there as well. He he points to quite a bit. But once you can combine those two together, I think that's going to be another kind of phase shift in construction. How long that makes its way down into your home remodel, that's like a different question. But you know, uh our customers, a lot of them are kind of the most, you know, tech forward out there. And and yeah, it's it's I think technology has to meet the customer where they are in their field. And it's it's just been harder in in these physical environments like construction.

Progress Tracking Needs A Brain Layer

Roopinder 21:39

About 10 years ago, I was on a job site and I was watching uh, I saw probably the most innovative 10 years ago at the time. The innovation was hey, wow, we can see we can hold an iPad up to what we're making, and an image of the BIM model would show up to compare it to. And I thought, wow, that was 10 years ago. It was on a construction site. And it was done by a little band of programmers in a special section of this company. And they had a similar frustration. They had managed to get it on one job site, but to get it to every job site was a bit of a challenge for them. Oh, what was I going with? It's totally off on a tangent, but I got to get back to what you guys do. So, okay, so now looking at an object is one thing. You said we put the eyes in, but the eyes need a brain to process that information, right? The eyes just recording image light is not enough to tell it. It's not like the brain has to tell it, oh, what I'm seeing there is a dog or a wall or a computer, right? That's where the AI comes in, correct? I'm hoping that I could see and I can say, okay, that's a pipe going overhead, or that's a duct or something. I don't care about points, right? I want to know what that object is.

Closing The Loop With Schedule And QA

Michael 22:54

Yes, yeah, absolutely. I and uh and that's really where it gets it gets complicated, right? And so the AIs that exist, like the language model technology that powers a lot of these agents, they're they're pretty impressive, right? They can do a lot of things um and they can add a lot of value to those workflows, particularly when the images are are spatially located, right? Because you you you give it an image of a you know a uh you know a uh a dent in the drywall, and if it doesn't know where that image was taken, there's only so much you can do with it, even if you can understand what's happening in the image. So there's kind of like a just the foundation of understanding the space itself, which is is really where open space kind of found it. That's our kind of foundational technology. But once you have that, like you're saying, there is this kind of brain element where you need to understand what's happening. And that actually can get very complicated very quickly in construction. And so uh we actually, you know, have been looking at this problem for a long time in the form of kind of progress tracking, where you actually want to take a set of images across multiple spaces and understand the overall progress of the construction kind of broken down at the component level, uh, you know, framing, fire protection, drywall across all of your different trades, um, which can be hundreds, even a thousand different components that you're trying to analyze. And this is where the off-the-shelf language models start to really struggle because again, they don't, they haven't been trained. You know, they're trained on internet data, um text and images, but not spatial data, especially not on construction sites. We actually acquired a company last year called Disperse out of the UK that actually had really specialized in this area that involves bringing this all together so that you can actually get human-verified information about progress pulled out of these images. Um, and then agents can then interact off of that level of data as well, which is now much more easier for them to consume. And then that's really when you get lots of magic happening, when when your agent that's you know pulling the progress tracking data is now working with other agents in other systems, for example, updating the schedule in your schedule system or interacting with your uh you know, ERP system and you know, helping people get paid, that that becomes very powerful. But yeah, having that brain layer is one of the areas that that we have very much focused on and is kind of one of the big challenges in construction.

Roopinder 25:14

Okay. All right, so let me project. Like I said, I did a little bit of research, but only a little bit. But this is how I wish open space worked. Tell me how close I am to this actually happening. Okay, I'm managing a construction site, I'm making a building on it, or fact let's say a factory. And as I'm making the factory, right, I wanted to recognize what is being constructed, like where the I-beams are, the foundation is, where the structural steel is, where the walls are getting built. In each case, I want to take go around taking pictures, I wanted to map it to, I want to record the images for sure, because it gives me a fifth dimension or fourth, yeah, 40, the fourth dimension, time and projection. I want to compare that. Uh, where I wanted to do, what I wanted to do is compare that to the schedule, what the building is supposed to look like. I want it to compare it to the physical, the 3D model of the building structure. I want to be able to take my iPad out at the model and see, okay, where this is where it should be, this is where it is, and also I wanted to see the compare the physical to the to the design anytime, maybe in the even in the future, right?

Michael 26:30

Yeah, I mean that's a pretty good description of progress track, our progress tracking offering, actually. And so um, so yeah, uh, you know, a user will take a 360 caram off the job site to get that base layer, that kind of 4D image data. And they'll do it every every week or even more frequently at some point. That data then goes off, goes through this progress tracking engine. Um, and essentially this is where we kind of see this kind of um, you know, we often talk about it as kind of closing the loop of intent, right? And you have a model, whether it's a BIM model or even just a 2D drawing, you know, a lot of a lot of folks still build off of 2D drawings, particularly in the construction phase, uh where models are harder to keep up to date. But basically you want that reality to kind of loop back at every turn, right? At every at a high frequency. So you're constantly getting up-to-date information about how progress is tracking against that that intent, that model. And then coming back to the job site and now helping the uh team adapt to any challenges that have come up. So making the workflows around RFIs and punch lists and safety issues, everything that's actually happening in the field as well, which is a huge part of making that loop efficient. That's where that comes in. And so this is kind of very much exactly kind of what we're doing with progress tracking, and then linking that in, like I was saying before, to all those other areas, not just observing the schedule, but then you actually want to help update the schedule as well. Um, and an agent can help you do that or help generate uh get generate uh get your workers paid, right, based on what's happening. Like these are all kind of things where agents can now come in and start uh moving the ball forward for you. And uh and I think the other like kind of key piece of that, which is has always been hard in construction, is actually starting to learn from that experience, right? So that the lessons learned on one building can actually be transferred to the next building, um, which right now is very much, I mean, it happens, but it's it's it can often be up to the kind of human experience to kind of transfer that knowledge. Whereas when you can start to really encode what's happening on the job site, then once it's encoded in the system, now you can actually start to use AIs to try and understand how you might be able to improve that in the future. Um, and that's that's really a kind of one of the future areas I think that you get unlocks, um, particularly within an organization. Once you're doing this across your whole portfolio, you can start to kind of do meta-analysis and say, hey, why why did these you know these last three buildings you know take longer than these other four and start to really kind of analyze and compare across that, uh, which is really only something you can do when you when you have that data collected. And so uh yeah, one of one of the exciting things I think for us in the future.

Roopinder 29:12

Yeah, yeah, no, that's that's very important, I I think, because otherwise these teams they have I'm thinking of one team that's building stadiums, it specializes in stadiums. So and boy, every time they make a stadium, it's completely different looking. And I think I think none of their knowledge probably transferred over because they have different geographies, they have different teams, they're not getting the benefit from having building one stadium to the next, right? Whereas it could be done if it's all a common common shared database that would be that would be something. So that's great. Now about about sharing that knowledge, there's also well, tell me what happens then. Is there almost is there also the possibility that you could see how it's differing? Like suppose one part of a large airport is a runway is drifting off the mark from the plan. Is there a way to flag information so you can look at it rather than depending on it being noticed? Is there a way to say, oh, this building's not on a kilter or this runway's way off? It should be. Is that is that kind of checking done?

Michael 30:18

Yeah, there is uh in our progress tracking, we have a feature called spotlight, which is where if the system identifies something that it suspects might be wrong, it kind of flags it as an issue that uh a human can then go in and and uh um you know observe and identify. And there's various other versions of that, um, you know, creating agents that that look at images looking for safety issues, you know, kind of various ways to do that. In in general, though, I think that kind of workflow is important that you described, where it's really the system is raising issues that humans go in and kind of really understand and verify and make um not just obviously correct as well, but also have to kind of take part in that process because the systems can support the human, but really believe that these are designed to make humans do their job better. Um, fully replacing the human is uh, you know, I I I don't see that really as kind of part of the, you know, the part of how these uh agents are really gonna like uh make things work better.

Let Customers Build Their Own Agents

Roopinder 31:18

Yeah. Well, not yet, right? I I'm reminded of uh years ago there was a uh a Belgian company uh that was in the skilled construction, working with the construct skilled construction trades employment service, really, and it hung a banner up on the side of a building and it said, Hey AI, make with it finish this building, AI, finish this building, and it is pointing out that you you need humans, AI. We were talking about agents, is open space now in the business of making agents, and it's been so talked about the last year and a half. I just wonder how difficult that is. Can an ordinary person, like a dummy like me, make an engineer? Do I have to depend on a genius, right, to make that agent, right? If I if I wanted to do something, I wanted to say, okay, find on the site who's not wearing a safety vest, right? That to me seems like a job an agent would do. But could I do that myself?

Michael 32:16

Or you know, it's uh it's really amazing how the AI, the language models have made it so easy. And I think this is actually pretty recent. Like we we've been tracking the technology obviously since the beginning and have always been really impressed by what could be done. But pretty recently, like over the last you know, three, four months, um, you know, a couple big model releases in November kind of kicked it off. Um, where these these the ability to have non-engineers vibe code their own agents, their own uh software, really has just kind of taken off in a way that that is unlike anything I think we've seen in the past. And so uh I think that really is changing the nature of software and and software products in general and opening up the world to a kind of customization that that you we just kind of haven't seen before. And I think that that kind of people talk about agents and generally I think they you know can mean slightly different things. The word is kind of murky, the definition is kind of loose. Um, and so what you kind of extract from it can be can mean different things. But when I think about what it implies for the industry, it's it's gonna it's really gonna change how people kind of interact with software and allow companies and users to be much more in the driver's seat about what is actually built. Um so from OpenSpace's perspective, you know, we are have launched this platform uh that is really designed to enable customers to start to create agents that are really customized for what they want to do. Like you said, if you want to look for safety vests, but maybe you want to look for a particular brand of face of safety vest or um just you know in particular areas of a job site. You might have a very kind of unique thing that you want instead of waiting for open space to develop that feature specifically for that we wouldn't then have to prioritize against every other customer and what they care about. What we want to do is open up an environment to make it easy for you to go and do that yourself. Um and so we have the data, we make it accessible, make it accessible to your to your uh essentially kind of personal builder that can build the software for you. And then you can just kind of run an experiment and do it. And of course, we'll build agents as well that have been kind of tested and more universally applicable for all customers and give you access to that. But in addition to that, this ability to go in and really build something yourself, I think is something that you're gonna see you know over and over again in many different types of uh industries. And I think construction actually can really benefit at that a lot, particularly again when it's built. Again, it's always gonna be only as valuable as the data that you give it. And I think this is why we're so excited about this at open space, because the data that we can provide to these agents are essentially giving it that physical information, that spatial information, that visual information about what's actually happening in the field. Um and I think that's gonna make the difference between the you know valuable agents that actually impact your business uh versus just kind of fun toys that you try and then you kind of forget about in the future. But yeah, that I think it's a big, it's a big shift that's coming that we're kind of in the very early stages of still, um, but is is really gonna change how you know how the world works.

Roopinder 35:39

Yeah, yeah. We talked about barriers to adoption, is actually like let's say it's inertia that people that's too stuck in their ways and they keep doing things the way they want. Too busy to change. You've seen that cartoon, I'm sure, where the wheel is presented to to somebody or a machine gun is represented to an army and they're too busy to adopt this new technology because they're dragging. Uh because they're too busy dragging rafts across the ground or they're too busy firing bows and arrows or something. But anyway, okay, that's one barrier, but there's technical barriers too, right? And to my mind, to my let's say the lay person's mind, is that we've been talking, the whole industry's been talking, the whole world has been talking about loud language models and the language. But to me, it seems like the barrier is physical AI. Like that is still a problem to be solved. Yeah, we can understand language, sure. We could ask it to fill out essay questions and things like that. But boy, the physical thing, like on a job site, you know, the volume of material that's removed, or you know, or this object object recognition, that that depends on 3D information, spatial information, and uh physics, masses, forces, things like that. Do you see that also as a technical technical challenge to the business?

Michael 37:04

Yeah, absolutely. Um, I mean, I think that's one of the big technical challenges in in language models in general for AI. And it's it's very much the focus for us. You know, and there's different ways to think about that challenge. I think there's you know, there's the kind of very deep way with how to actually train these language models using spatial data. Um so instead of thinking about a language model built entirely from data from the internet, which is images and text and you know, uh huge scale, how do you rethink building models that are actually using that 3D information, that spatial information about where images are, what what happens? And uh, you know, I think the data that we have been collecting is is uh very valuable in that effort. And but that's a bit of a longer term project for us. In the more immediate term, there's just, you know, how do we take language model technology that exists today, already very powerful, but how can we augment that with the information that we have extracted from our um our kind of more uh non-language model uh uh AI systems, right? That are pulling in those 360 images, that are identifying where they are on a floor plan, understanding the relationship between those. Um, so there's a lot of information that we can extract and then organize the data in a way that language models today can actually do amazing things that they couldn't do if you had unorganized data or data that you didn't know where it comes from, right? What didn't know what room the image is in. So we can kind of augment the current language model technology, which is what we're doing today and why we're able to create these agents that can do these very impressive things. But then there is that longer-term project of actually the kind of next generation of language models that are actually built on this type of data. And you see companies like uh, you know, out there, NVIDIA talks about this with physical AI, um, Feife Lea is a company that focuses on this. And you know, I think this is very much also part of the future development and something that that we at OpenSpace are contributing to as well, but in addition to what we can do today, just with the off-the- off-the-shelf language models.

Roopinder 39:10

Right. Oh, okay. So you mentioned other companies in the space. Uh Autodesk had a big push into reality capture a few years ago. I want to say 12 years ago. No, more like seven to eight years ago. And then they abandoned it. And uh the people were all let go. Was that the business opportunity that that open space came into? I think because Bentley's still going strong with reality capture, I think, but uh a lot of just kind of abandoned that whole group.

Michael 39:40

And uh yeah, you know, I don't I don't know too much about the history there, um, so I can only kind of speculate. But um, I know reality capture has meant different things at different times throughout the kind of evolution of construction technology. I think when we got into the game, you know, we were really started the company on on a bet that was really about AI taking off and camera technology taking off. And so what was novel when we started the company was the 360 camera had really started to come into play. And I think my my understanding is that that kind of previous reality capture was much more focused on laser scanning and lidar and that kind of three dimensions. And honestly, I I think uh, you know, adoption was just difficult. It was a real barrier to entry there because it just takes a lot of time and expense and and you know, people have other jobs, they have you know their day jobs to do. And so I think one of the reasons that we succeeded there was uh one of the big reasons was because we made it so easy, you know, really banking off of the this the camera technology just continuing to get better over time, and AI technology continuing to get better over time. And the kind of those two coming together is what paid off for open space and allowed us to really make something that was very easy to adopt and kind of broke through that barrier. Uh so yeah, I that that would be my my guess.

Roopinder 41:07

Yeah, I've got to get me one of those 360 cameras.

Global Scale Business Updates And Farewell

Michael 41:10

Oh, yeah, yeah. It'd be fun.

Roopinder 41:11

It'd be a fun operation. The uh I I'm looking at your at your site and I I see you have now locations, customers all over the world. I've got to ask, is your is are you impacted? What's going on in the Middle East? Because some of your locations looks like they're yeah, yeah.

Michael 41:30

Um you know, um, we are uh a global company, um, as you said, yeah. So so we've been fortunate enough. I think we're on about 80,000 projects worldwide, uh 125 countries, and uh and and yeah, definitely some projects in the Middle East. We've been fortunate, you know, first off, in just in terms of, you know, everybody's safe, everybody's okay. And so very fortunate that way. And uh, you know, there there have been some impacts and just because of the uh the war, but um, you know, we're optimistic that, you know, uh that kind of being able to shift things around and and make that go forward. But uh obviously any kind of changes to construction and anything that impacts, you know, construction starts or you know, uh is gonna impact the business to some extent. But you know, the the good news is people are always building. And you know, even when one part of the the kind of global landscape changes and and uh you know residential kind of slows down in one area, then all of a sudden data centers starts popping up in somewhere else. And so it's actually one of the advantages for us having such a wide footprint and and and just being you know kind of the leader in the industry is is uh we really have that kind of diversification, both geographically, but also just from an industry perspective that protects us against that. But yeah, uh, but yeah, everything impacts everything.

Roopinder 42:50

Yeah, we're all hopeful for uh some sort of resolution for sure. Okay, so I'll make this the final question. How so it sounds like business is doing well, and uh you were well funded. I uh well funded several years ago. You got a good round of financing. Is uh how's the company doing?

Michael 43:08

Uh uh doing great. Yeah, no, uh things are going very well. We've kind of continued to kind of hit these milestones and pass these thresholds that I think uh a lot of smaller companies in construction can kind of get stuck in and hit these ceilings and plateaus. You know, construction is a can be a challenging market uh for sure, just in terms of uh how distributed it is and and uh uh kind of non-centralized some of the customers can be. But uh yeah, I think because of our focus on adoption and the strong technology of the team, we've been able to break through that and uh have been growing very nicely and gotten into a very nice size in the construction industry. And uh I will say, you know, we just came off of our kind of global um company meeting. We all get together once a year um somewhere around the world and and uh you know talk about the future. And uh it's really kind of Never been a more exciting time, I think, for open space, for the construction tech industry in general, uh, but just the confluence of technologies that are coming together and how people are adopting them is just uh, you know, I think we're we're really primed for uh a very exciting future. And so uh so yeah, really thrilled about what's coming next. Good luck to you. I think it's great technology. Uh, you know, like I said, I fall in love with technology and certainly fallen in love with what I see an open space. I think it's great. A long time coming. I hope more of it is adopted. Uh I'll I'll do my part, certainly. Absolutely. All right, very good. Thank you very much. Thank you for listening to FoDES, the Future of Design and Engineering Software Show, brought to you by ENGtechnica. I hope you have learned of a new application or technology that will help you with your job. If you have an application you think would be of interest to other engineers, please let me know by emailing ooppoomeoo at repinder at enchtechnica.com or message me on LinkedIn.