Understanding jobsite progress is time consuming, inaccurate and expensive. Constru AI are using advanced computer vision techniques from autonomous vehicles and other sectors to revolutionize how we collect jobsite data, how it is analyzed, and how it changes engineering and management workflows. Michael Sasson, Constru CEO, explains all of this and where it is going next.
Hugh Seaton: [00:00:00] Welcome to Constructed Futures I'm Hugh Seaton. I'm here with Michael Sasson, co-founder and CEO of Constru AI. They're a breakthrough machine vision company, doing what they do on, on construction job sites. Michael, welcome to the podcast.
Michael Sasson: [00:00:17] Hi, thank you so much for having me.
Hugh Seaton: [00:00:20] Excellent. So let's, let's start the process with, with talking about what you mean by machine vision on the job site. What do you do?
Michael Sasson: [00:00:28] Yeah, that's that's great. So so we are a deep tech company and working on this, algorithmics to understand the visual information coming from the job site. So we, we are actually taking, taking the images on site or using the imagery that is taken by others and and understand them like, like the human does.
So that's, that's pretty much what we do.
Hugh Seaton: [00:00:51] That's there's so much in there. So let's, let's start unpacking. The first thing I want to establish for the listener is that you're actually not providing cameras. You're you're providing the intelligence that works on whether it's a photo or a video. Is that right?
Michael Sasson: [00:01:06] Yeah, absolutely. So, so, you know, if you look at the modern job sites, there are plenty of let's say platforms or, or maybe even the devices that, that take eh, digitize the process. And it could be spherical cameras like Ricos or Insta360s or several other platforms that collect this data and put it against the floor plan.
So we are actually focusing on the brain side of those things. So we, we gathered this images either by gathering ourselves through the, through this, the same devices like that, as in Ricos and the instance. Or maybe we can, we can get this imagery from other platforms. And I, I would also, eh, mention here some robotic platforms, like Spot, which can be used to to, to hold those cameras or maybe stationary cameras that will be, might be in the future on job sites. So we are pretty much agnostic for that. We are working on this amazing computer vision to understand the imagery at the at the level of the human engineer.
Hugh Seaton: [00:02:16] That's exactly the right next place to go is when you say the human engineer or talk to me about what sorts of things that your, your system is helping with right now? Like what specific parts of the process is the system helping with?
Michael Sasson: [00:02:32] Yeah. So if, if you, if you take images frequently onsite, so there is a clear understanding of the jobs what are the jobs are currently in progress? What is already completed? So we, we have this contextual understanding, so we call it the contextual because it, it takes the image and understands what is, what's the job we can see here. Okay. And that there is additional level of identification, is the object identification. So we, we identify every element in the, in the image and try to understand whether this, this element needs to be there.
Why it is there now. And, and then we, we can, we can simply cross correlate this data with the design data. So we have progress tracking. So as we call it and we have discrepancy monitoring. So these two things combined together, provide us with a clear understanding about the project, the project progress, the project risks and And the brings in enormous amount of engineering insights to the teams.
Hugh Seaton: [00:03:39] And that's actually one of the things I want to make sure we cover is that, you know, it isn't that hard to make, you know, a Google API recognize something. I actually did it real. I'm not kidding. I had somebody from from Fiverr, train it on some tools and it, you know, a little demo can make it feel like this is not so hard, but what you guys have done is similar to another, a couple of other companies DADO did this with voice or they didn't do this, but they did an analogous thing where they, they really applied an expert layer to make it work. Can you talk a little bit about a, how hard it is to do what you've done and some of the steps in the process that you guys have gone through to make this work?
Michael Sasson: [00:04:20] Yeah, absolutely. So great question. So first of all, of course there is always a Pareto working, so it, it takes a relatively, a small amount of time to achieve some kind of model, so 20/80 works. But the problem is that we achieve a production level algorithmic sets it's much, much more harder, but the, the problem with construction is actually, that's these elements are, are changing and the scene is changing constantly. So, so that's, that's the hardest part. And eh, to to, to, to get to the, to the, to the level of understanding of the site, you need really to work hard on the various , parameters, let's say, right. And, and I would say it is, it is also not enough to have a visual information only. We are, we are combining this information with with with the information we get from the BIM. And, and from the schedule.
So, so everything combined provides us with an ability to understand the site. And and this is, this is the hardest part.
Hugh Seaton: [00:05:32] It's understanding the site. So in addition to BIM, what else are you able to pull in that helps you to understand the site? What other kinds of information.
Michael Sasson: [00:05:41] Yeah. So, so you know, if you look at the BIM there are, there are different levels of, of data and granularity, and let's say a level of development for the BIM.
So, so it could be different different BIMs with with different level of data. So beyond BIM. And, and if, if we, if we go back to the BIM, so we understand where is the complexity, for example, where are the systems that we need to monitor more frequently and to understand the materials of let's say like mechanical systems or, or mechanical rooms that, that you need really, to, to monitor every element and to understand if this element is correct or not. And, and this, this is, this is like understanding the complexity where the complexity is to understanding what are the discrepancies that, that we really can, can make a difference for the, for the customer and, and like critical things. We, we call them bombs within the wall.
So you, you have those those bombs all over. And my, my first visit to the site. Was around discrepancies really? I was in shock , because , so many small things , that really can, make a huge difference for the, timeline of the project and for the cost are not discovered on time. Okay. So, so this is the first thing.
The second thing is there are, there are lots of other, eh, criteria's for the data to be fused with our output of the, of the computer vision. And we are even talking about weather and if you are on a superstructure and there is there are some engineering points that the temperature is very important.
So, so we even might look at the, at the weather. As a factor. And, and this is something that we are constantly adding to the algorithmics. We are constantly adding it to our brain. So the brain has enough inputs and senses to to understand, again, we are talking about a complete understanding of what, what we see.
Hugh Seaton: [00:07:46] So there's, I want to come back to something you said earlier, just now, is this, this idea of discovering discrepancies early? I mean, you know, everybody in the industry has seen those curves where you know that the sooner you do a thing, the cheaper it is to either rectify or to change You know, are you finding that customers are getting a lot of value from that?
That they're discovering that, that, "Oh, that's not what we thought it was," really early, as opposed to later when things aren't working and they have to fix it.
Michael Sasson: [00:08:16] Yes, absolutely. Of course. So if we take like a typical project in the United States and we are working with a tier one companies and most of them are having VDCs in the process and doing coordination and, et cetera.
So we, we see that many things are actually closed, eh, not within the BIM, but maybe verbally, maybe on paper and then later brought to BIM. So in this case you really need to verify that what was agreed is actually happening.
And we are becoming a verification point.
So this is amazing because, we actually didn't thought about it. And so we are getting those discrepancies sometimes eh, not, not really, not real discrepancies because this is just the deviation from theplans. But this becomes a verification point for the VDCs.
The other things are around things that are not really discovered. And we we got plenty ofcritical discrepancies that arereally not discovered on time. It takes several weeks or, or even more to discover them. And when we raise the flag, let's say, okay, with our discrepancy monitoring system it really makes a difference for the customer.
First of all, itdramatically lowers the risks for the project. Secondly, it dramatically lowers the cost of rework. And in general, we are able to monitor the overall quality of the progress. And and this is, this is something that definitely brings a huge value to the customers.
Hugh Seaton: [00:10:05] This is awesome. And one of the things that it makes me think of that I want to make sure we're clarifying here is, you know, someone doesn't need to have set up a camera. They don't need to have a senior person do a walkthrough. What you're saying is, you know, within limits, whatever camera is there, we can capture that and then send it back to the expert, wherever the expert is.
Is that right?
Michael Sasson: [00:10:28] Yes. So, so in general, we are agnostic to the, to the cameras. There are you know, there are several cameras that are in use now heavily in use now. And I, I mentioned those and we don't care if you know, if we are capturing it or others are capturing it or the customer of course doing that himself.
I think we are focusing now on the spherical images because , it is much easier for us to understand the surroundings when we have a combined image and not taking like six, seven, eight different of the room. Right? So, so that's, that's the only limitation let's say, but overall we can work with any spherical camera, we don't require any additional sensors. Okay. So, so that's, that's the bottom line.
Hugh Seaton: [00:11:18] I guess the reason I keep harping on this is how easy it is and what you're doing is, is, you know, instead of having to send someone who knows what should be being built and understands what the agreements are and having everybody all in one place, sitting in the room, looking at what's there you're able to separate in time and space, the kind of sensing of what's there and the judgment about whether it's right or wrong. I mean, A, your system is doing that, and B the person who's going to do something about it doesn't have to be there. So you're, you're, you're automating what would otherwise be expensive and difficult to continuously schedule?
Michael Sasson: [00:11:53] Exactly. And we, we can see that on several projects and by the way, the the pandemic actually greatly helped us to penetrate very large accounts and, that's the overall you know, tendency of the industry now to go for digitalization and working remotely. If there is no other way. And I can tell you that we clearly see the pattern of increasing the digitization of the site and an increased amount of the data either visual data or a scans like with LiDAR and other means of scanning coming from the site. And we think that customers at the beginning, like if you look at like five years ago it, it was great to have just images and it was already like a huge help.
But now when you need to manually see every image, it's really not solves the problem. If you have this automation of extracted data coming from this imagery and presented to you within a very easy way. And we are building this data platform on top of our computer vision engine. That really provides the customer with ability to slice and dice this data, to search for anything, to get to the bottom line of what he needs to see and get the insights.
And, and this is, this is the next thing. And we, we see it rapidly getting to the ecosystem. And we are definitely there to help the industry.
Hugh Seaton: [00:13:28] Actually you bring up something really cool, just now. And, and so prior to this podcast, I got to look at the product and what you're providing is this kind of critical dashboard where someone who knows the project will look at all of the places where the Constru system has said, there's a discrepancy and we'll be able to apply judgment to say that one's okay, that one's not okay. But the key thing here is you're giving people the opportunity to make that judgment, instead of just hoping everything is going well. Is that how you're seeing people use it?
Michael Sasson: [00:14:02] Yes. Yes. I, I think first of all, we're really getting help from human engineers in teaching the system what is what is correct and what is not correct? And and I can tell you the truth, our first engine was applied on a large project. And we throw thousands of discrepancies on the customers and it wasn't so pleasant for them.
So we tried to understand. How can we really analyze what is, what is the real problem and what is not, and that we implemented this feature. So the first customers we were really, really fortunate to have great customers who helped us. Right? So the first customer really used the system the way you are describing.
So they, they had to, see and and apply some, some input. Is it, is it a real thing? Is it a real problem? Is it, is it really matters to me that the electrical outlet is slightly off. Or not. And now I can, I can tell you that 70% of discrepancies we are sharing with the customers are real discrepancies that are, that are reworked.
Okay. This is, this is an amazing statistic.
Hugh Seaton: [00:15:18] That is amazing. So you're filtering out things that are below some threshold of importance.
Michael Sasson: [00:15:25] Yes. Yes, but this is, this is not, a geometrical threshold of inch or, or several inches. This is, this is completely different thing. We, we have this contextual understanding, so we understand what is really matters and what is not.
We are also applying some some conditional judgment on top of that with a call inspection. So. So we know what is the critical for for this region. For example, this is something that we are starting now, now to do in the US. And we are constantly enhancing our ability to judge things and to understand things again, like a human engineer does.
And we're coming from the autonomous car technology, we were developing different things in the past. And the level of understanding when you are changing the lane or or you are doing something else that involves your judgment on other cars or pedestrians or whatever is happening on the road. It's tremendous. So you see the other car in the mirror and you try to understand if they will, they will give you this opportunity to change the lane or not. And this is the microsecond judgment you are making in your mind.
So, so that's, that's exactly what we're trying to do here, where we are really thinking about all the small things that the engineers are looking at when they see the problem and they really understand the issue, they have in-depth understanding. And that's the difference between Like advanced AI, or very simple like mesh between the scans on side with a BIM or whatever.
Hugh Seaton: [00:17:12] This is a story of construction technology generally is that people often will make things that are a little bit lightweight and surface, and they do the thing they're supposed to do, but they're not really applying deep insight into how the, how the project really works. And that, again was one of the things that was exciting to me is you you've really, as a company spent a lot of time and effort trying to put that expert layer that, additional layer of functionality, whether you call it AI, whether you call it rules or some blend of them, which is probably what you wound up doing is, is just really exciting. Cause I think that's what makes it useful to people.
Michael Sasson: [00:17:47] Right? Right. That's that's that's correct.
Hugh Seaton: [00:17:50] And so tell me about an example of a project that you guys have worked on. Even at kind of at a high level, but what's a, what's a good example to give, to give people a concrete understanding of what you do.
Michael Sasson: [00:18:03] So we are focusing on the high-rise residential now. These are the types of approaches we started from we are not limited to residential we have office spaces. We have data centers now. We have plenty of different types of construction, but I would say well, we, we see that the complexity and the complexity can come from the scale. Or from engineering complexity. This complexity is actually emphasizes the value we provide because it is very challenging for the human engineers to inspect something at a high, very high scale or on the very high complexity.
Right. So typical projects. And I would like to take you to one of the, of the first projects eh, A here in the Israel, in Tel Aviv. We were on the high-rise building. It's like 64 building. And there were, let's say 500,000 square feet or around. And there were brilliant engineers working on that project and really experienced project managers.
And at the beginning, they, they were saying, okay we know everything, we know what's the progress. What could you really bring us with we don't believe that the information will be somehow valuable. And this is like the first time they see that the technology.
And we said, no problem. We will just try to bring you some useful information and you will see if you, if you want to use it. And after a week, we had an additional talk with them and it appeared that it completely changed their way of thinking because they understood that they now can focus on real problems or analytical thinking of the project, and not doing very time consuming work of going, eh, and inspecting actually every floor and every apartment and every corner of the, of this building. And that the real problem is that the human engineers are not so good on repetitive and the highest scale work they're brilliant at at analyzing things like we humans, we, we, we are, we are doing great job with our mind.
So what happened is they totally understood the value. They totally understood that they are not capable to find all the discrepancies because as I said, we brought thousands of discrepancies on them and they, they really understood the, the power of this synergy between the machine and the human engineers.
So, so that's the typical thing. And, and, and they actually called us a new way of construction. I think it came from the point that we somehow predict when the project is going to finish. It's not the, you know, the, brilliant prediction, but it, it roughly gives you the end date and we provide them with all this obstacles along the road. So we show them what's not going right. And what they need the you know, to do actually. So this is, this is the, typical, thing that, that we are seeing on all our projects. The first projects are always like a bit skeptical because they think they know what's going on and then they understand the power.
And we, think that this technology will completely change the way that projects are executed. We really believe that this technology will make a huge impact on the industry.
Hugh Seaton: [00:21:57] Actually, that's a great segue to where I wanted to go next. And that is, I can think of like two and a half ways of thinking about how things will change. One of them is, I mean, for lack of a better word, I'll call it a qualitative difference where you're allowing people to focus on higher value you know, analytic questions and problems instead of, you know, having somebody who's relatively senior walking around, taking notes.
And the second one though, is. We know what data density can do. You know what I'm saying? So if you have people walking around and, and taking notes, once in a while, you get some level of data about what's going on, but what you're talking about is potentially much more data per pass and much more frequent data that is accurate because you're not relying on an expensive human to do it.
Is this the way you're seeing things go.
Michael Sasson: [00:22:49] Yes. Yes. I totally agree with you. This is the, exactly the way we think the and the next solutions we'll go with go. If I can tell you what are the use cases we are discussing now? With the customers, these, use cases are always going to the data density.
And to do an ability to extract things that are already impossible to extract with just human engineers. Because of the scale. And we, we now can, can provide a clear eh, quantities, for example, for the completed work, I can tell exactly in square feet, howmany flooring, how many you know, sheet rock how many things were done completed and , and this is the level of elements.
And the thing, the other things and I I'm taking you to different projects and different geographies. And in some geographies you know, the, the customer wants to understand how many things were broken. Like, like how many tiles were broken, how many windows were broken during this process?
It might sound like irrelevant, but it is possible today. It wasn't possible before, but it is definitely possible now with just one filter in the system, you have this data there. And that you can rate and rank your, your traits. You can, you can understand the patterns.
You can, you can analyze things like we do on the web. You, you can understand the the effectiveness of your decisions as a manager as well. So these, these are the amazing things that really are happening now. As we deploy the system and more and more peoplein the organization starting to use this and explore the power of data.
Hugh Seaton: [00:24:39] So I want to almost put that in a larger context is you could say that construction data in a construction site, went through one round of digitization where there wasn't really much more data. It's just now whether it was in purchase orders or whether it was in RFIs, all that sort of thing, it just became digital. We're still working on making people, enter things in a consistent way. So you can do analytics against them, but that's, that's coming and it's just, you know, kind of normal internal change.
But what I'm seeing you say is there's a new wave of construction data coming that is computer collected is , you know, organized and it has intelligence applied to it, but is also vastly denser and more I don't know, reliable maybe is the right word, but, but certainly accurate that allows entirely new things that we're just starting to think about.
I mean, are you seeing that, that you guys are part of a wave of, of just literally a new, almost a new paradigm of construction data?
Michael Sasson: [00:25:36] Yeah. Yeah. That's, that's how our vision, that's what we are trying to achieve now. And, we want to be a part of the ecosystem. I think there is an amazing ecosystem is built now.
With data documentation as a starting point. And there is a layer of extraction and understanding and like platforms like ours that we are building. And there is an additional layer ofproject management platforms and communicating with all the all the parts of organization and maybe doing many things.
So, I strongly believe that this this ecosystem will bring this amazing change, right? We are not alone. And we might be a great enabler because we, we are getting this data to the next level. We really can extract everything about every element and tell the story about every single element on on-site. Okay. So yep.
Hugh Seaton: [00:26:38] It's an exciting vision. And I'm really, really glad we were able to go into it like this. I think there's just, we're just starting to scratch the surface of what, the capabilities that you're building and, and companies like you are building. Thank you for spending time with me on the podcast.
Michael Sasson: [00:26:52] Thank you so much.
Hugh Seaton: [00:26:54] Last thing before we, before I forget, where should people learn more about you and Constru?
Michael Sasson: [00:26:59] Yeah, you can see more information on our websites. It's constru.ai. And and of course you can , go and write to us: there is a email on the website as well.
And of course the demo, we will be happy to demonstrate the system for all of you guys. And yeah, that's the future is, is amazing for the sector. And I I'm really thinking that the next few years will be amazing for the construction .
Hugh Seaton: [00:27:30] Fantastic, Michael. Thanks again.