AI-Driven Generative Design for Building Systems 

Podcast Transcript 

Steven Rossi-Zalmons: 

Hello there and welcome to the Construction Revolution podcast. My name is Steven Rossi-Zalmons and here on the show we explore the latest trends, technologies, people, and organizations that are revolutionizing and disrupting the construction industry. Today I’m speaking with Francesco Iorio, co-founder and CEO of Augmenta. Augmenta is using AI-driven generative design to automate how building systems are designed, starting with electrical, taking work that used to take weeks and getting it done in a fraction of the time. 

Francesco spent over 15 years in artificial intelligence and computational science at Autodesk and IBM, where he pioneered the concept of generative design, before founding Augmenta to bring that same thinking to the construction industry. 

Steven Rossi-Zalmons: 

Francesco, welcome to the construction revolution podcast. Thank you. It’s great. I’m excited to learn more about you and learn more about Augmenta and sort of a different type of technology than we’ve had before. So yeah, why don’t we jump right in and can you tell me the story of how you founded Augmenta and where you got the idea for the company? 

Francesco Iorio: 

Thank you for having me, actually. It’s a pleasure. 

So, the idea actually from the company came from kind of a long time ago when my co-founder Aaron and I actually met while at Autodesk and we were tasked to solve essentially the problem of advanced manufacturing. At the time, especially with the emergence of 3D printing, the original concept of computer aided design was kind of rapidly becoming obsolete as it was actually difficult for people to imagine. 

The type of structures that a 3D printer could create, right? So out of thin air. And so, we started essentially thinking about how computers could truly become a partner in design rather than just being a repository for design knowledge that people already have in their minds, right? So, but truly being a design partner. 

So, we took actually that concept actually all the way from very early conceptual design and research all the way into products, specifically for the aerospace industry. And then we turned our attention to construction specifically because we realized actually quite quickly that construction is in dire straits, even more than manufacturing. People are spending a lot of time and cycles and energy into solving essentially the wrong problem. 

So, people and technology are being helped to document their thoughts, but they’re not assisted in solving decision-making problems. And that’s really where we thought that the application of artificial intelligence actually could make the largest difference, really, the largest difference. 

Steven Rossi-Zalmons: 

Awesome. So can you tell me specifically on what is it that Augmenta actually does and who are your users and customers? 

Francesco Iorio: 

Certainly. Augmenta is an artificial intelligence startup. We build foundation models for the construction industry. And the aim of these sophisticated agents and models is to help translate a high level design intent that can come from an engineer, from an architect, from a construction specialist into full constructible designs. 

especially for the critical systems inside buildings. So this helps essentially kind of navigate the challenges and the perils and the time that it takes using, let’s say, conventional kind of fully manual workflow to craft these very sophisticated three-dimensional maps, if you will, that represent how heating, cooling, and ventilation happens inside the building or how electricity kind of around. 

The building to power essentially all the systems that make a system, make a building living rather than just a shell. And in particular, today, the first agent that has been released to the public actually so far is in fact actually dedicated to the electrical space, right? So, where artificial intelligence takes high-level power requirements that make up how to functionally describe the building and then translates them into fully actionable three-dimensional designs. 

That can drive manufacturing, procurement, scheduling, right? So, and everything downstream leads to the actual construction and operation of the building. 

Steven Rossi-Zalmons: 

Great, nice. So can you walk me through as a contractor, as someone using Augmenta for electrical planning, what does that process look like and how does it differ from the norm and the traditional process? 

Francesco Iorio: 

For sure. Again, the process of so-called BIM modeling or building information model detailing and modeling is kind of a lengthy and extremely specialized process. So today it requires a lot of expertise, both on the digital side, essentially having the proficiency of using advanced computer-aided design tools that are not easy to learn and to master. 

But on the other side also requires deep experience in the field, right? So, in the actual nuances of construction, there are lots of unwritten rules that people learned essentially during their careers by literally kind of walking, your job sites, by manufacturing buildings, by essentially them operating them. But the combination of these skills is actually very, it’s very hard to find, it’s very hard to train, right? So that actually kind of causes, especially recently, a substantial shortage. 

in the type of people who are expert enough to make these designs. So generally speaking, the designs are created entirely manually. An expert modeler, designer, receives the specifications from an engineer. It receives at the same time, let’s say, the building shell. So, a three-dimensional model of just the building structure, if you will. 

That represents just walls and the ceilings, the floors, columns, but essentially kind of an empty shell that the fundamental critical systems need to be crafted inside of. And so that’s fundamentally the type of information that they receive from an engineer and they have to perform the modeling, so essentially the drawing of all these individual wires and so-called raceways and ducts that kind of spiral around the building actually to make it functional. 

So, the way our AI essentially behaves is that it takes the same information; it takes the same type of, let’s say, multimodal kind of so-called information. So it’s not only text, it’s not just an Excel spreadsheet, it’s not just a drawing, it’s not just a 3D model, but it’s all of the above, And essentially it takes the similar perspective that a BIM model or an expert designer, an electrician in this case, would have when they tackle the problem, So in particular, it’s a complicated… 

system in the sense that it needs to have awareness of a building as a whole. Creating these designs is not a small local problem that involves only the design inside the room, for example, throughout the corridor. Literally, the way these systems are designed and laid out across a building determines their cost, their overall performance, the amount of material and time that are actually used to construct them and operate. 

And so, the AI has to have enough smarts, if you will, right? So, to actually recognize all these characteristics and then perform actually the work on people’s behalf. In essence, if you will, the AI represents an extension of the modeling teams that already exist inside these contractors, where it’s like having a virtual BIM design team behind your screen, So where you feed the same information that you would to a senior design engineer modeler, and then. 

the outcome is a fully implemented BIM design. So, if you think about it, it’s parallel to what cloud code actually does today for software engineers. So, where specifications and intent can go in and the actual functional executable, let’s say code that can be used comes out. 

Steven Rossi-Zalmons: 

Nice. So, I guess I’m wondering in terms of like, as you mentioned, like it’s very complex and requires so much training. How did you actually train the AI model and how long, where did you get all the data? How long did that take? yeah. 

Francesco Iorio: 

No, absolutely. The problem is much more complex than even the state-of-the-art machine learning systems can deal with today. So, if you think about it, the building is a combination of spatial reasoning, thinking about volumes, thinking about space, thinking about geometry, and thinking about physical entities. And at the same time, it is a semantic problem. There are flows of people inside the building. 

The various rooms and spaces are meant for something. So, to contain from like process equipment in a manufacturing plant, so to assembly systems, our guys actually use such a broad space of building typologies that we had to train it on a vast kind of sites. But crucially, the kind of joke that I actually tell people is that these general machine learning models like Claude are trained on the internet. 

They’re trained on an enormous corpus of text, books, and internet resources. But there’s no Wikipedia of buildings. There’s no Reddit of buildings. There’s no YouTube of buildings either, especially at the engineering level of detail. So, you can, in fact, actually today create machine learning models that produce a video. But the level of semantic understanding 

of a world model to produce a video is infinitesimally small relative to the level of detail and precision that engineering applications actually require. So, we had to overcome this colossal problem, which is a lack of data. And the way we actually tried to, or the way we actually did solve it, has been to follow in the footsteps of another field that has been successful at training, which is self-driving vehicles, self-driving cars. 

In self-driving cars, machine learning systems are trained vastly by running virtual cars in virtually simulated environments, right? So not in the real world. Of course, they’re also trained in the real world, but a fraction of the time of the fraction of the virtual, you know, kilometers or miles, so we essentially created the same representation. We created a very large-scale building synthesizer. we created very, very advanced 

patented algorithms, that literally create buildings out of nothing that populate them with expertly crafted building infrastructure and systems, but that take a very long, it takes a very long time, right? So, to operate and execute, something that would not be acceptable for a user to actually kind of use daily, it can take a week to compute like a whole building at the level of detail that an engineer requires, but with that ability, 

We could generate a vast amount of synthetic data, which is the foundation that you can then use to bootstrap the training of large scale, spatial AI machine learning models. And then you feed the machine learning models, the real buildings, let’s say a handful of real buildings at the beginning, to train it further. And then as our system actually got deployed to customers, every project that is actually is using it like, 

Of course, then kind of further teaches the AI on kind of real, from a real-world perspective, if you will. That’s really the strategy and the trajectory that we’ve used. So massive investment in generating synthetic data and then further training of data essentially depends on real world usage. So. 

Steven Rossi-Zalmons: 

Right. Yeah, that’s a really interesting approach. I think, yeah, the comparison to self-driving cars and the spatial AI obviously in general is like an entirely different concept than a lot of things that people may be familiar with. 

Francesco Iorio: 

That’s right. That’s right. It’s a distinction that is actually very stark and it’s not necessarily intuitive. It is actually quite simple actually to test it. I don’t even know yourself. So even the state-of-the-art models have poor spatial reasoning, especially in three dimensions. So, they struggle even in 2D, so the reading drawings make sense of even architectural drawings. 

but like full three-dimensional kind of spatial understanding and reasoning and also the scale, So it really kind of trip up these types of systems. And so that’s the reason why, we decided to have this hybrid system that is in a part, let’s say deterministic algorithmic, so that can make very intentional, extremely precise and deterministic decisions, 

because of rules of code that have to be respected and not violated. But at the same time, give essentially machine learning the power to perform high level understanding and semantic reasoning and spatial reasoning on our building to actually perform, let’s say, command and control operations. Whereas then the fine tuning is done by precise algorithms that finish the job, to place every single screw, every bolt with exacting precision. No hallucinations are allowed there. 

Steven Rossi-Zalmons: 

Sure 

Nice. So, when people hear AI, and especially when we’re talking about AI models that are this advanced, people may assume that it will replace either them or their engineers, designers. How do you frame the relationship between the human expertise and the existing workforce and AI and augment this platform? 

Francesco Iorio: 

That’s a really good question. So that will be a perennial debate in any field that AI touches. And this is kind of particularly true in an industry like construction that historically has been reticent to deploy very advanced technology. So, there’s no question. On the other hand, there is the duality of the information that these professionals actually get, that the reality of deployments in artificial intelligence in enterprises is not stellar. 

of records actually so far, which actually compounds the skepticism, certainly in the construction industry, unfortunately, least for the field, we are in an era of heightened demand for the types of skills that are required to literally construct, to actually design and build these buildings, So organization is on the rise, but the renaissance of manufacturing even. 

the West actually is still kind of booming, then the explosion of demand essentially for mission critical buildings like hospitals and even more so data centers, So we’re living kind of in a really unprecedented moment of demand, which is kind of paired with an unprecedented shortage of skilled personnel, right? These firms are also starting to realize that their inability to design at a faster pace also harms them. 

From several other perspectives, like for example, the concept of prefabrication. Prefabrication is a substantial aspect of cost control. So as in anything that can be repeatably manufactured or assembled inside a fabrication shop, it’s inherently cheaper to produce and then install. So, then something that is completely fabricated directly on a construction site. This also requires, again, extra detailed design work. 

because a construction assembly facility requires extremely precise instructions on how to put together, assemble the components that then are actually shipped to the site. So, there’s compounding factors that are pushing, if you will, this demand that is currently overwhelming, what we see is that people are not looking to supplant engineering and BIM teams like entirely, that’s not the point. But on the other hand, 

They’re seeing the promise of usage of artificial intelligence in making the experts they do have substantially more productive, right? And so, it’s like having each one of these team members have their own team, also overnight, which is an amazing accelerant. And it’s also very much a different type of work; It’s a different type of satisfaction. Most people actually are not ecstatic when they drag the mouse essentially on a screen, you know, 

14 hours a day every day. And that’s also kind of part of the compounding kind of skill shortage. Kids don’t like the idea of actually doing that as a career. this is also, all of this actually makes it slightly easier actually for us to relate of the affordances. And the fact that we even named the company after our desire to extend human abilities rather than replacing them. And really kind of this is where 

the industry kind of seems to be going. There’s like a concerted effort on reducing the waste, actually making things more productive, but not to remove the human touch. So, it’s just to extend the reach that the humans, that the few humans who are still good at it and who still want to do this for a living, so can do. 

Steven Rossi-Zalmons: 

Yeah, absolutely. mean, the shortage of workers in construction is no secret. Anyway, we can augment and accelerate their production. Yeah, absolutely. 

Francesco Iorio: 

It’s also not only an acceleration; the affordance is actually called, let’s put it like this. The affordance of time has secondary consequences that are significant, really significant. Right now, the disproportionate amount of the construction waste and rework and extra excess time that happens in the construction process is due to design errors. 

So, the inability to design at that level of precision and leaving again that level of ambiguity compounds comes downstream. right? material waste, there are some incredible stories of construction projects actually essentially throwing into a landfill 30 % of the construction materials that get on the site, which is insane if you think about it. But this is due to all these variables compounding, accelerating this work of design, of detailed design has impact that get into 

sustainability in terms of all embodied carbon, it has the impact on operational costs, so buildings can actually be designed to be much more optimized in terms of their energy usage and performance than they can do on average. And simply because engineers really never have enough time to spend on every particular project. There’s just not enough time nor budget, And so third, but not least, it’s really the optionality. 

there’s hardly ever enough budget on the 99.9 % of the built environment, to run what if scenarios, to understand kind of macro decisions, and engineers actually always struggle with this, know, owners and developers always struggle with this because again, they can’t wait for, you know, six extra months to run, three, four other scenarios with their engineers and the bare, cost of actually doing that to actually make the most informed decision. 

But design automation and the introduction of AI help tremendously with that. People can actually make decisions in a much more informed way, which has all sorts of downstream benefits. 

Steven Rossi-Zalmons: 

Right. So, I guess on your platform then how long would it take to map out the electro for, mean, obviously it would depend on the size and complexity of the building, but I guess for say for an apartment building or something. Can you run those scenarios like to optimize for materials or efficiency? is that, run those scenarios through. 

Francesco Iorio: 

Yeah, no, absolutely. I’ll give you an example, maybe of a hospital, it’s just a kind of more complex scenario that is interesting from a variety of reasons. I would say, let’s put it like this. So, one of our customers was working on a very large hospital, actually, in the United States. And due to their construction schedule and due to the rush, that they’re under, at some point, they were a month behind on the pre-construction schedule. 

They deployed our technology and they caught up over a weekend. So that’s the level of acceleration, if you will, that people can expect by deploying this type of systems, if you will, this type of affordance, So as in the level of effort required, So it changes actually very drastically and the remaining time can be used for optionary, for better decision-making, for refinement. 

So that’s really kind of the major kind of advantage. And to your point, optionality actually becomes a reality in that circumstance. For example, imagine a high-rise building. They have the ability or the opportunity to distribute mechanical electrical rooms across the floors, for example. They have to host the equipment somewhere between the roof and the basement. Right. So, the foundation is 

But there are infinite possibilities if you think about it; it could be all put in the middle of the building or all on the roof or all on the basement, right? So, all different combinations. There’s hardly ever the opportunity to run studies to understand how manufacturable, how quick, how expensive it is to choose any of these options. Like what people generally do is like, we’ve already built 15 kinds of similar high rises. 

It kind of worked out with this particular design. We’ll just do it again. That’s that’s because there’s never enough time with technology like ours, with our technology, you can in fact run all of them simultaneously. You can actually see the effects of your decision or their options are right. So, in a very stark, kind of visible way, right? So, so all of the designs that we produce by the different trials, if you will, would be fully constructible and costed. Right. So, you can actually get an understanding. 

And you can compare and contrast the different options. You can see the compromises that you’re making. It’s not even necessary that one design is certainly better than all the others from all points of view. But at least if those compromises exist, you will know about them. So, if you pick one design, you’ll maybe be more expensive in terms of actual construction materials. But it could not be. 

to turn out to actually be faster to install, right? So, because it’s easier to reach, for example, in the ceiling’s unit, right? So, there’s, and vice versa, right? So, and then again, you make the informed choice rather than, you like you roll the dice, if you will, right? So that’s really the difference. 

Steven Rossi-Zalmons: 

Yeah, no, that’s for sure. That makes So,lot of sense. And yeah, thoare, those savings on time. That’s very substantial. That’s great. Moving sort of a little bit more to like your background, I know you’ve worked in tech and specifically in computing for advanced computing for a long time and in various different companies and even industry. So I’m wondering how that background has helped you in your journey to starting Augmenta and 

being an entrepreneur for the first time, how has that journey been? 

Francesco Iorio: 

So, it certainly helped to kind of connect the dots, if you will. Especially having been one of the inventors, if you will, of the overall kind of methodology, if you will. So, it certainly helped me to kind of transport the concepts and the ideas and some of the foundational kind of technology approach to a different industry like construction. But having, let’s say, a technical 

background overall, allows you to kind of reason very systematically about problems, right? literally the entire kind of idea behind, you know, generative design and generative systems, is one where, you are trying to solve problems that go beyond the human reasoning, that extent beyond the amount of variables, if you will, that the person is kind of believable to. 

to rally and to match and to reason at any given time. And so that actually transferred extremely, extremely well. And then the journey for me actually to entrepreneurship actually was more one of making sure that this approach would exist in the first place. So that was kind of taken seriously and it would exist. And where it wouldn’t disappear inside a menu, if you would, of an existing kind of software package, if you will. 

This is a very kind of fundamental approach to shifting. So, and myself, my co-founder and our team are very invested into, again, creating this alternative avenue, if you will to design buildings, right? We have to do it ourselves to prove that the methodology can stand alone, and it doesn’t have to be a subset of the way the traditional way has been historically, it’s its own discipline. 

Steven Rossi-Zalmons: 

Yeah, yeah, for sure. Yeah, that’s very interesting. moving a little bit away from augmented for a second, I’m curious in your time and like you mentioned, you’ve been sort of fundamental in developing spatial AI, but even stepping farther back from that, I’m wondering how you’ve seen the industry and construction grow with this adoption of tech throughout your career. 

Francesco Iorio: 

I’ve been in technology for many years. You can probably tell from my beard, but I’m still relatively new to construction. Construction is such a nuanced world, right? So, I don’t even pretend that I’m an expert, nowhere near, what I can say is that 

bringing an outsider perspective into the world of construction, so some constraints and affordances become kind of painfully obvious. And not to say that I would be the first one to understand them or to even see them, far from me to say that. But it’s more that there are scope and opportunity to change systemically some of how these procedures actually work, that they have been ingrained in this way for a reason. Again, there’s no… 

There’s no criticism, but there’s very fundamental reasons actually why the industry actually works the way it works. like today, and I honestly believe that technology adoption being on the rise, certainly in the last decade or so, right? So, it’s not only an economic benefit from the industry, but it’s a global benefit that the entire human race actually has as a consequence. The built environment is such an important component in our lives. 

Technology adoption actually goes beyond the margins of everyone involved in construction. over time, especially the fullness of time, I strongly believe that it will foster, again, a very different perspective, So all the, let’s say the hidden variables, all the affordances will emerge, sustainability, livability, as in concepts that are the fundamental job of an architect, an engineer, et cetera, but that sometime, 

for constraints of economy, so budgets, time scales, it may become secondary, But the adoption of technology can make these prerogatives actually kind of back to being a first-class citizen. 

Steven Rossi-Zalmons: 

Yeah, yeah, for sure. So, I guess as you put it as being an outsider, what is the response that you get when you go and talk to people in the industry and present the solution? 

Francesco Iorio: 

 Well, I mean, we have a good response in the sense that people aren’t learning from me or from us that it’s a problem. So, we’re not, we’re not preaching something that they don’t already know. All we are saying to them is that there is possibly a different way. The technological level that is emerging like from computer science may finally have a very, very stark kind of impact on their, on just their organizations, their businesses and the industry overall. 

So of course it’s a skeptical industry, naturally and understandably so. So, the only way in, on the door, is proof, is building trust, is building proof. And we are, I believe we’re approaching this actually very humbly. We’re not trying to teach anybody, you know, how it’s done. We’re just trying to help and trying to actually show that these affordances actually do exist, and they actually do indeed work. 

at least, you know, in the vast majority of circumstances. And so, we don’t ask people; we’re not trying to shove software down people’s throats, AI down people’s throats. We’re just trying to actually get invested in the problem, showing a possible alternative, a possible solution, partnering very closely with people who are willing to give it a shot, to actually try for real, and then showcasing their success. 

That’s really our approach fundamentally as an organization. So, we’re stepping into the field very humbly, but delivering the actual results, so when it matters. 

Steven Rossi-Zalmons: 

 Yeah, for sure. Yeah, that’s a great approach. So lastly, I was looking on your website, and it mentioned that we talked a lot about electrical planning already, but it’s like you’re going to be getting into things like mechanical and plumbing and even more than that, assume down the road. So, can you tell us a little bit about the things you’re working on and how you sort of see the solution and the company evolving in the future? 

Francesco Iorio: 

certainly. Let me be very clear. Our grand vision and our ultimate goal are for AI to assist in designing whole buildings. everything in a building. The critical infrastructure inside the building is number one, because it’s the number one design pain, the number one design, and construction cost. And so that’s the, and also the hardest to automate, the hardest to teach an AI to work with. So we went in that sequence, if you will, we are going to that sequence. 

because we want to prove that the methodology and technology apply to the hardest problem first so that we gain that trust that then allows us to go downstream, so to reuse the fundamental methodology and technology to move forward. But yes, the next step is mechanical and plumbing, essentially at the same time. So now we service the electrical field, so both contractors and engineers, but the next stage is going to be mechanical and plumbing. We’re working on it already. 

We plan to actually release, we plan to actually complete the work, let’s say in Alpha stage actually in the next few months, because again, the foundations of the platform are already there, it’s an extension and not rewrite. Crucially, that is what is going to move us from our AI, from a pure cost perspective. The tuning, the way it learns, is to actually minimize the construction cost, if you will. In electrical, that’s essential. 

key ingredient, so to take the smallest amount of time and the smallest amount of material to deploy an electrical system. For mechanical, especially, this changes actually the game quite a bit in the sense that mechanical systems are the most linked to performance of a building, the energy usage in buildings is dominated kind of generally by heating, cooling, and ventilation, right? So, the accurate 

and optimal design of HVAC systems, So it is core to essentially the operational performance of a building and it’s carbon emissions, of course, So that will be the first step for us actually to go into the performance realm, So the performance domain we look for, we’re already working very closely with a number of excellent design partners 

Steven Rossi-Zalmons: 

 Yeah, awesome. That’s very exciting. mean, it makes a lot of sense that, I mean, they all need to go throughout the building anyway. So, you might as well plan them all together and make them all work together. So, to wrap us up here, if people are interested in learning more about Augmenta and getting it and using it on their own projects, where should they go and what does that initial process look like? 

Francesco Iorio: 

The easiest steps are to go on our website and fill out the contact form. So, we start essentially processing with an introductory call to understand the nature of the projects that people work on, the nature of their organizations. And then we can follow it up with the demonstration of our technology already being deployed on other projects and other customers. 

Essentially, we got from there. 

Steven Rossi-Zalmons: 

That’s right. Okay, perfect. Well, thank you so much for your time and your insights. It’s been great learning more about you and about Augmenta. 

Francesco Iorio: 

I really appreciate the time. Thank you for having me on your show.