In this episode of the "Counting Sand Podcast," host Angelo Kastroulis sits down with Myk Murashko, CEO of Maestro, a construction tech startup. The discussion revolves around the integration of artificial intelligence and other technological advancements in the construction industry, a sector historically resistant to change. Myk Murashko, who comes from an architectural background, shares his journey from studying architecture at Cambridge to founding Maestro. He delves into how his company aims to revolutionize building processes by bridging computer science with construction to enhance sustainability, efficiency, and quality. Maestro's approach focuses on moving the complexity of construction from on-site activities to pre-fabrication processes using CNC machines and digital twins. A significant part of the conversation is dedicated to the potential of AI to optimize the use of materials such as mass timber, reducing waste and emissions. Myk explains how Maestro utilizes advanced technologies like LiDAR and AI optimization to precisely cut timber, aligning the grains for maximum structural integrity and minimal waste. The episode also touches on the broader implications of technological integration in traditional industries. Myk and Angelo discuss the importance of adopting new technologies in a way that respects and incorporates existing practices and the craftsmanship that characterizes the construction industry. This episode not only highlights the innovative strides being made in construction technology but also reflects on the challenges and opportunities of bringing high-tech solutions to traditional fields. Angelo wraps up the conversation by emphasizing the importance of practical applications of technology that meet market needs and improve industry standards.
Host: Angelo Kastroulis
https://www.linkedin.com/in/angelok1/
https://twitter.com/AngeloKastr
www.hermes-nexus.com
www.carrera.io
Guest: Mykola (Myk) Murashko
https://www.linkedin.com/in/mykola-murashko-041b47156/
https://www.maestro-tech.com/
Angelo: [00:00:00] Can artificial intelligence help us with our everyday tasks? Can it help us with things like the building materials we use in our house? That's what we're going to talk about today. I'm your host Angelo Kastroulis and this is the Counting Sand Podcast.
Joining me today is Myk Murashko, CEO of Maestro. Myk, your background though is not business, as one might expect out of a CEO. Your background is architecture. How did you get into this position? What does it have to do with technology and computer science?
Guest: Well, first, Angelo, it's great to be here. It's great to be talking with you. You're right. My background in a technology startup is definitely not standard. Maestro is a construction tech startup and my background is actually in architecture, as you correctly pointed out. What Maestro is, is a construction technology company that tries to bridge computer science, AI, with the world of building. So we try to supercharge the way that we build using the technology of the day. And the way that we got here, the way that I got here, is through a journey that took me from studying architecture in quite a traditional way at Cambridge to looking at architecture on the scale of cities and then looking at how data science can help us understand cities better.[00:01:00] In this way, I met Professor Carlo Ratti, who runs Senseable City Lab at MIT which looks at how data can form the urban design of cities and, you know, after a few years of research with him into where in the built environment can technology help us make it more sustainable, faster, higher quality for the people that live inside, we founded Maestro, which really looks at questions of sustainability and quality and construction, thinking about how technology can help solve that equation.
Angelo: Interesting. So, one of the interesting things I think that you guys do, and in your history specifically, your background, is that it isn't just about any one kind of computer science, and it isn't just computer science, there are other kinds of technology that are involved. Because when you think about construction, it is not merely a theoretical concept inside of a computer that's computations, it has to materialize to the real world, and so that we can do things with it.
Guest: Yeah, and I think this is really the exciting part because if we look back at [00:02:00] kind of the last 50 to 80 years We see that the world of technology, you know, the world of bits around us has changed drastically. If you think about the kind of telegraph operator a hundred years ago versus a person posting on Twitter today, they would have very little in common. Whereas if we took a mason from 1923 and brought them to today, they might even be doing a better job than the masons of today. Very little has changed in the world of construction. And this is really what we are looking at, at Maestro because as much as construction
hasn't changed, our demands for our buildings have. They have evolved massively. And as a result, we end up in this situation where the construction industry accounts for 13 percent of global GDP. We have to build our cities, the places we work, the places we live, our infrastructure. Everything relies on the physical built environment. And this results in immense emissions to the planet. So around a quarter of global greenhouse gas emissions come from the industry of building, and this breaks down mostly as embodied carbon in the structures that we built. So, for [00:03:00] example, the production of cement that goes into concrete is very carbon intensive, as well as the operation of buildings later on in their lifespan, right? So in a way the physical world of building has such a huge impact on our environment that we need to understand how we can use technology to make it more sustainable and faster. And I think this is a huge question to answer. It's one that we are not the first people to pose, but we think that we have something new to bring to the table, which is that because we're talking about the physical built environment, one can not kind of enter this industry and just
bring in technology. Technology is definitely the key to solving some of these problems of emissions and construction. But you need to also come in with a kind of humility about. The fact that you're entering the world of atoms, entering the world of craft, a kind of discipline that has been around since mankind, you know, built the first primitive hut. And so we really believe that the key to making construction more sustainable is using existing technologies from other industries, learning from [00:04:00] those and surgically injecting them into the building process. And there are very different ways and very different technologies that can contribute to this, but really it comes down to a core thesis that we have, which is that if you look at a construction site today, what it is, is effectively a very complex assembly of very simple parts. So the mason putting together bricks or the framing crew putting together two by fours into a balloon frame. And the result of this is that there is this very complex process happening on site. It's very manual, quite difficult to control. And as a result, you get the kind of emissions, cost overruns and delays that plague our industry. Whereas what we believe the future is a transition from this complex assembly of simple parts to a very simple assembly of much more complex components. And what Maestro does is really design and manufacture these kind of components. And we will dive into much more detail about this, but essentially the idea is let's take the complexity of construction away from the site and give it to the CNC [00:05:00] machine, the digital twin, the manufacturing facility, and bring on site these components that can come together and be a building.
Angelo: Okay, yeah, so immediately the computer scientist in me is finding various areas in what you just said. There's opportunities in efficiency, making things happen faster. There's opportunities in cutting waste down, using computer science to do that. And it seems to me that there would also be opportunities in quality, that we could get much higher quality with a little bit of extra planning instead of making decisions in the spur of the moment while something is being constructed. When concrete is being poured it's not the time to make a decision, but oftentimes we're stuck. There's nothing you can do about it. You have to make the decision. And so quality may or may not see its way through, especially when an engineer may have specified the process but they might not be there to see the process happen. So, definitely I can see opportunities. A lot of things you just mentioned hit on those. Any one of those stick out to you as something that you've done any work in?
Guest: Yeah, absolutely. So maybe first quick thought [00:06:00] about where we can make the difference. And then we'll go on to some physical examples of what we've built and sort of proved in the last few years. The first is that, in the end, there are two places where you can really make the difference in the construction process. The first is at the design stage, right? The way that you make the decisions about the building should be informed by a very clear and dynamic picture of what that building is going to be. So that's the first. And that's really to do with engineering and the architecture and that discipline has actually moved quite far forward in the last 20 years.The second place where you can really make the difference is the moment where you translate this into the physical world. When you translate the information about your component parts into the physical component parts. And this second moment is really where we think there's a lot to be done, and that's where CNC computer numerically controlled machinery comes in. Where we can translate code that we can generate through our digital twin environment into the machine instructions that cut our steel, our timber, our glass, but then we bring into our buildings. So, actually where we started is precisely here at this intersection between [00:07:00] design and manufacture and looking at creating a layer on top of the design environment, on top of the 3D modeling environment, to take you to the manufacturing environment to cut component parts. So to give you an idea of what we're doing, at this point we are building three pilot projects across the world. One in Florida and two in Northern Italy, which are constructed out of steel and glass and also primarily mass timber. And all of these component materials are being manufactured using digital fabrication, so using code, and we believe that this way you get huge benefits to the site because the component parts that arrive are tailor made, they're precision manufactured, so all you need to do is bolt them together into your building. Rather than go on site with raw materials that you need to shape and cut. So if we look at what technology that makes that possible, there is a digital record of the final product of the building. And this is something that the design industry has done well for the last 20 years. There's been a huge adoption of something called BIM, building information modeling, and this has changed the way that architects and designers design.[00:08:00] So we went from designing things in 2D to designing things in 3D and storing information about building objects. However, having this perfectly coordinated model of information about the building is not necessarily enough to result in leaps forward in the efficiency of construction, and we believe that this kind of change can be made by layering more information onto the model and that information can be laid in a digital twin. So a digital twin is the representation not only of the static design, but also of the processes that will make that design happen. Specifically, that means going from design to manufacture through computer aided manufacturing. It means tracking the production and delivery of those components on site. And it's about using the design information in the 3D model and translating that to the site using augmented reality so that when components do arrive you've got your AR goggles that help you assemble component parts. So this is kind of the first bucket of what we are doing, which is trying to organize the construction process around [00:09:00] digital fabrication.But there's second also very exciting opportunity that we see is that once we have developed the process in this way it also allows us to do upstream innovation, so to look at the very raw materials from which we built. Because in the end, the process innovation can help you reduce waste and increase efficiency, but if you are to really change the kind of emissions that we do in the construction industry, you also need to look at the raw materials from which we build. And an exciting piece of research that we did this summer is looking at how we can make the mass timber industry more efficient. And maybe this is a good moment for me to give a bit of context about mass timber. So, I think we're all familiar with the timber frame house. Your two-story family house in the suburbs is made with timber structures. That timber frame house uses boards from a tree and uses very many of them to create this light frame. Now, this is a great way of building. It's fast. It's cheap. It's what built most of America, but once you build multi-story buildings, timber frames no longer work. And it no longer works for a fundamental reason, which is that timber planks, [00:10:00] timber boards are quite unpredictable. You know, the grain of the tree means that it is sometimes difficult to engineer big structures using raw timber. And so mass timber is a set of products that was developed in Austria in the nineties which effectively takes timber boards that the industry calls lamellas and glues them together in these layers, and then puts layers of boards on top of other layers at a 90 degree rotation, creating cross laminated timber effectively is a panel that is a few inches thick, which is a mass timber product which can be used in construction as a wall or as a floor slab. And as a result, effectively, can replace concrete for mid and high rise buildings. People build sort of five, seven story structures with this. And we see cross laminated timber as kind of a big step in decarbonizing the industry and it's something we use across our projects. But what we were looking at this summer, is if you actually talk to the people manufacturing this product, you realize that when you consider the volume [00:11:00] of timber in a tree, the mercantile volume that you can use build, and you look at the amount that is wasted during the process of making a CLT, of making mass timber products, it's something in the order of 60 to 80% which surprised us, you know, it's absolutely huge. And so the question that we pose is, well, how could we make this better? Because effectively the question is, there is this natural material that comes from the forest. The forest is shaped in organic ways and we are trying to make from it this perfectly engineered product that we can easily calculate and account for when we design a building. And we believe that this sort of interaction between the natural and artificial can be bridged with a bit of AI optimization and robotic fabrication. So specifically what we did is we've developed a process for manufacturing mass timber panels in a much more efficient way to help our construction process. What it entails is doing a LiDAR scan of the tree trunk of the log before doing the saw milling process to understand this geometry, to understand the different shape of the [00:12:00] tree.
Angelo: Just for listeners, LiDAR is a radar. So you're using a radar to see the log and see how it's oriented and it's structural design. Okay.
Guest: Yeah. Yeah. So the idea is you want to end up with something that's perfectly flat and rectilinear and easy to build with, but you're starting with something that is very organic and shaped in a beautiful way by nature. So the first moment where you need to start is to understand what nature has created. And the best way to do it is to get a 3D scan of our log, of our raw material. And a LiDAR is the way to do it. I mean, most of you might be familiar with LiDAR in the context of self driving, right? It helps us understand our environment around us by scanning the world as we drive through but in our case, we're using it to sort of understand the tree. And the moment that we have a perfect understanding of the geometry of the said trees, we can think about how can one create timber boards that are not rectilinear in shape, but rather that follow the geometry of the tree, that follow the curves, branches, and other sort of irregularities that come in the natural material. And the sort of research we did is combining this imaging [00:13:00] technology with an AI optimization that looked at optimizing or minimizing the amount of volume of tree that you need to cut away in order to create these perfectly tessellating boards that come together to form CLT layers, right? Then you translate your perfect sequence of boards and to machine instructions for a seven axis robotic arm that cuts your log in a very precise way resulting in you basically assembling the same kind of controlled uniform CLT panel, but doing it out of organically shaped boards. That reduces timber waste from the log down to something like 30%. So it's a huge saving compared to what industry does.
Angelo: So in ML when we want to try to settle on an optimal solution, we use a loss function. Loss function essentially will penalize a model that has less desirable characteristics. And so you're free to architect a loss function that makes sense for you. Now in this case what you said was your loss function, I guess, in essence, [00:14:00] penalizes something that requires a little bit more shaping and cutting than something else that would naturally fit. Is that how you designed your loss function?
Guest: Yeah. So the loss function is the volume that you have to cut away. The volume that you have to waste. And effectively the way that we got there is by looking at the log. And the log gets effectively sliced or plain sawn into boards. And you can understand the geometry of those boards since you've got the LiDAR scan of the initial log. You can treat those boards as raster representations, and then you can understand, well, what is the most efficient sequence? What is the most efficient order in which you can take these flat sawn boards, at this point, you've basically lost zero percent of the tree trunk, right, you've just sliced it. What is the most efficient way to order these boards such that they perfectly match one another? Such that they perfectly tessellate. And this is really the optimization part of our process. We are looking at, effectively, the overlap. When you compare two rasters representing two [00:15:00] timber boards, and we are looking to minimize that in order to achieve, kind of, a perfect sequence. Now, naturally, because we are working with a limited set of logs when we try to produce panels this way, you end up having to cut away some timber in the end. But, because we have gone through this optimization process, the amount that you have to cut away, so the volume that you have to lose, is minimized drastically. And you know, it's one of the ways in which we can see how the sort of advances in technology that we encounter every day when we go online or when we browse the Internet can affect the physical world around us. Right? I think it's a really exciting moment because, in a way, construction has always been set back by a degree of inertia, partly because it's a very capital intensive endeavor. Building is expensive and so people try to be as risk free as possible. And partly because it's an industry that is full of very small players, the carpenter, the tradesman, and maybe those players don't necessarily have enough power to inject technology into the process. So what we try to do is we take on the role [00:16:00] of people that do that and sell our service to individual contractors that can then use our materials and use our digital twin to build in a more sustainable, faster way.
Angelo: So, what I love about this is it illustrates one of the things that we data scientists think about and anecdotally we say, that 80 percent of the work is in preparing the data or cleaning the data, wrangling the data, understanding the data. Maybe it's more like 95%. And then the last bit is actually the model work. Figuring out the architecture and optimizing and all that. In this case, this is really interesting because, an AI doesn't do anything here, I mean, it's only as good as the information we give it. So, this process of using a LiDAR, which does nothing unless you do something with the information. It just is going to give us data. And then you have to take the data and make sense of it. And you're rasterizing it, then you're using an AI model to try to figure out the right answer. An AI model is going to give you the best solution it can based on the parameters that you gave [00:17:00] it, your loss function. Then we have to feed a CNC machine, which is going to actually do the work, which is what a human being would have had to do with years and years of experience. But the problem there, is a human being can't see through logs and a human being can't look through 10,000 pieces to see which ones should be the correct matches. It's kind of like the way that a fine wine is put together, or whiskeys in the different barrels and an expert has to mix them. A human being can't do that problem. So it's not even just replacing a human being. It's doing something a human being wouldn't have the time and ability to do because it's just not something that we're good at. So this is, I thought, a really nice metaphor for the process of how technology, you have to look at it as a value chain. It has to be part of this entire chain, right? And one innovation in one thing is not enough. You have to kind of innovate across the chain in order to make it reality.
Guest: Yeah, absolutely. And you know, a strange silver lining of the fact that construction is far behind technologically is that, actually one doesn't need to do hard science to make real [00:18:00] impact in terms of sustainability in terms of quality and speed and construction, because if you think about what we've just described, the component parts are not new. LiDAR has been around for decades. AI might be a new term, but machine learning is probably older than my age. And CNC manufacture was born in the 80s, right? So the individual parts are there. They've been tested, they've been applied in different industries. And I think the role of technologists in the building industry is taking these existing innovations and threading them together to create solutions that work for our industry. And, you know, it requires kind of a degree of craft and a degree of physical understanding of, what is a CLT? What is a timber panel? How is it bolted together? How's it glued together? But also a degree of awareness of the different technologies out there. And then it's a matter of building almost a Lego house where each Lego is an existing piece of tech. They need to sort of put together into your little building and then the little building can help power process of making a physical one.
Angelo: Yeah, I think that there's a couple really important things here for aspiring data scientists and entrepreneurs too. That is the revolution [00:19:00] really comes from taking all these pieces of the value chain and understanding your domain. This is not about being highly theoretical and just, you can apply the same kind of thinking process to anything. You have to really understand the problem space. We call them unicorns. Someone who understands the domain, understands math, and understands computer science to be able to put all these together is really rare. But that's where true value is derived. And if you want to do something really amazing, our first inclination is to go for these moonshots. Like, this is what we have to do. We have to kind of revolutionize everything. That is not where you get the big gains. The biggest gains are made by applying this new thinking to old problems and putting them together and solving the low hanging fruit. Because what you just said about construction, I have heard that said about everything. I have heard that said about healthcare. The nice thing about healthcare is they're so far behind, we're still documenting on paper. But that's true, [00:20:00] I think, with every industry, they're just still doing things the way that they have always been. And now we have an opportunity to bring in some bit of technology to this and a little bit goes a really long way. If you really are invested in it and do it as a business. Yeah.
Guest: Maybe an additional note to add to your advice to the aspiring data scientist is don't expect the solution to be born at once. Right? A lot of us, I think, that are mathematically inclined are always looking for the most elegant way to solve something. And often when it comes to translating kind of technological progress into the physical world, it's about something very different. It's about taking the simplest version of the problem and solving that first. Because solving the simplest version of the problem does two important things. First, it ensures that you don't waste your time. And second, it ensures that once you have solved the simple problem, you can evaluate whether someone in the world actually needs it or not. Because the worst thing one can imagine is sort of spending years trying derive the most efficient [00:21:00] way of scanning a log or optimizing a set of, a sequence of boards, just to understand that, you know, maybe it's ahead of its time, or maybe there is no need for that. Maybe the industry has already solved it in a different way. Right? So putting out these proof of concept level solutions into the world is the surefire way to test whether technology applied to the industry can work or not.
Angelo: So let me ask you this. This is another thing that I think an entrepreneur would need to think about. Just exactly what you said about the market. Think about your market. Is there a need? Yes. Okay. How is the reception of the market to the product? So, clearly I think this is a great idea and it's going to produce a superior product. And as you mentioned, you don't have the data yet as to how superior it is. I'm sure it's structurally better. I'm sure in many ways it's going to be better. We don't have those models to even measure that kind of thing now. So you have to innovate and build those, but how has the industry accepted this? Are they onboard with it or do they kind of see it as technology is still a toy and it's in its [00:22:00] infancy and then, you know, we'll let it mature a bit.
Guest: Yeah. So I think there are two parts to this answer. The first is regarding our core business, which is taking architectural design, developing it into a digital twin, using that digital twin to produce component parts, and then producing those component parts with a network of existing machinery operators. So this model, this kind of high level use of the technology has seen quite good adoption in the last year that we have been testing it. Because the whole idea is that we take a technology layer and then we use existing players in the industry to make it happen, we're using existing materials, we're using existing machinery operators. So we're actually based in northern Italy, where we're in the heart of the European industrial triangle between Switzerland, Germany, Austria and Northern Italy. And we're using, kind of, existing machinery operators that usually work for the automotive sector to produce our steel parts and CNC mills for timber to produce components and then ship them to site where existing contractors can assemble them. So this business model has worked very well to take technology [00:23:00] and inject it into the process. Now, AI Timber is one of these kind of upstream innovations that we're doing. And here, the adoption curve, I think, is a bit longer. Because as much as there is definitely a material saving involved in producing the CLT in the way that we described and there is a structural advantage because the lamellas interlock, thereby increasing their resistance to shear forces. I think we're still about 5 to 10 years out from this being the standard way in which we produce CLT and the reason is that, effectively, every single board needs to be cut in a custom way. Now, this sounds quite scary and discouraging, but the good news is that CNC machines are quite agnostic to the shape that you ask them to cut because in the end it's lines of code and it's movements of a tool bit. However, the cost of CNC processing hasn't gone down far enough for this to make sense for timber, which is quite a cheap product, cheap base material. But you know, it's something that we're using on some high end architectural projects as kind of centerpiece of the design. And I think within my lifetime, part of the industry will transition there. But I think here there is yet another lesson that can be applied to [00:24:00] other industries, which is that, if you want to inject technology into an existing complex process, don't start from a clean slate, instead, try to use that technology to empower the existing players, right? I think it's the only way to do this in construction. The antithesis to this has been proven multiple times in my industry. The most boldly so, by a company called Katerra. And Katerra was founded by some Silicon Valley execs about 10 years ago and they raised 2 billion in funding from SoftBank and blew through it in about two years by having this thesis that they can start from a completely clean slate and build out their own production facilities, their own mounting labor, their own material supply, and layer technology on top of this and create the new way of building. You know, of course it fails because if you speak to any contractor or anyone who has ever held a jackhammer in their hand, they will tell you that people that come from the outside and try to start completely anew are likely to fail. Instead, it's about integrating existing parts. Not only existing [00:25:00] technological parts, like we spoke in the case of combining LiDAR with AI, with CNC, but also existing manual parts, right? In the end, a building will never come together by just code. You need people to bolt component parts. You need people that harvest timber. people that pour concrete in some way, right? As much as we're trying to minimize this, foundations will not go away. So, threading together technology with existing craftsmanship or with existing knowledge in the industries, I think something that probably applicable, like you said, to healthcare or to manufacturing as much as it is to construction.
Angelo: That's true. And I've seen it myself in many industries that an outsider who comes in thinking that they're going to disrupt the industry, doesn't always truly understand it. Sometimes disruptions happen, but usually real innovation comes from working within a system and understanding how to make it better. And construction is a great metaphor because you are not going to succeed if you're going to go to all the contractors of the world and say, forget everything you know, you need to learn how to build things differently. [00:26:00]Construction would halt. It would be impossible. Much easier to say, here is a product. It is stronger and cheaper. Organically, it will win. Do things a little bit differently, it requires a little bit of education, but look, it can happen faster. Now, there is an incentive. People's incentives can be aligned to learn how to do something a little bit differently. Disruption for disruption's sake isn't going to get you where you think you're going to go. And that's true, I've seen it, in many cases. Healthcare is a great example because it's something that we all know is broken so we think, well, let's just do it over. It's way more complicated. So that's, that's fascinating.
Guest: What you're saying is so, so true. Something that I have noticed in the last year is that the thing that has put the biggest smile on my face has never been an optimization. algorithm working or a digital twin showing a precise representation of a process. But rather when this thing hits the ground and you go on site and you see a craftsman or a tradesman that looks at what you have brought and shakes your hand and [00:27:00] looks impressed, is interested in understanding how it works, that's the real sign that you've made something. Right? Because just some technology working is not necessarily a sign that you're making a process more efficient or better. Usually the fact that you've managed to capture the attention of incumbents, that's when you know there's a green light and there's something there that's working.
Angelo: The academic in me, of course, would cheer whenever there's a breakthrough, just for the sake of the breakthrough. But it is true the practical implementation side of me, you know, the commercial side, sees that the true value, it only happens when it's realized. Only happens when something is used. And if it's academic and beautiful, but it doesn't ever get used, then it has no value really. So I agree with that. When we think about construction, we could all benefit from improvements because as we've seen through COVID, material prices went up, cost of construction, time of construction went up. It became so difficult to even manage the supply chain. [00:28:00] And so technology could help in all of that.
Angelo: Thanks again to our listeners. That will conclude this part of it. Join us for our next episode where we take on part two of the same topic. Again, we couldn't do this without you. Please subscribe to the podcast. It's available on your favorite platform and you can follow us on LinkedIn at AngeloK1 or AngeloKastr on X.