AI Blueprints And As Built Mysteries

Robert: Welcome everybody to Contact Exec. I'm here with Raymond Zhao of let's see, what is Structured AI? Raymond, welcome. I'm glad you're here with me. We met at ENR Future Tech, and now we're here chatting again on the podcast. How things been going for you?

Raymond Zhao: Yeah, firstly, thanks for having me. really excited to have this chat. yeah, things are busy. we've we've been back from a big conference road trip again where we met. and

Robert: Mm-hmm.

Raymond Zhao: now we're just going through all the all the con connections we made through that. So yeah, things are busy, things are picking up. It's exciting times.

Robert: That's good. That's good. Now I suppose everybody's gonna wanna know. Raymond, w give me a little bit your background and then give me a little bit about how you kinda got into this crazy business of construction technology. What what triggered it?

Raymond Zhao: Yeah. so it's my background, but equally my co-founder's background. So I'll try to do a quick summary of both. I started by studying architecture in so I was I'm from Australia, so I studied architecture in Australia. my family has a very strong more on the A and E side of the sort of the industry. So my mom worked at an interior design company, and she was like, Hey, like maybe you should really Like do an internship early on to see what it's like in the in the industry. very interestingly, at my internship at an architecture firm in Australia, my buddy sort of warned me, like, you know, so somewhat convinced me to not pursue architecture as a as as a profession, which

Robert: Yeah.

Raymond Zhao: I think it reflects more broadly of just I think a systematic problem of the industry, which I'll get into a little bit, which I think then perpetrates into the construction sites as well. Again, things end up as RFIs, things end up as change orders and project delays, all those. and then I did a hard pivot into math and studying statistics. so I studied at Oxford, I worked as a quant, I worked as a software engineer, I got really got into the sort of AI side of things. So that's my that's my backstory. quickly going to my co founders. My co founder also I met him at Oxford, so he's my good friend from university. He his background is in like very much like AI research, machine learning research. however, when he was 17, he was building like AIs which can play like video games. So he's training different AI agents to be really good at playing different 3D simulation. have you heard of Unity 3D? It's like Yeah.

Robert: Yeah, not me. I'm probably

Raymond Zhao: Okay.

Robert: not the guy for that one.

Raymond Zhao: Right. it's essentially he trained a fighter jet to trying to dodge NTM missiles trying to shoot it down. So then

Robert: Okay.

Raymond Zhao: completely just using like reinforcement learning. So the fighter jet would you know, it will crash, it will get hit by the bomb, but then eventually it learns to dodge the bomb because that's w what it's trying to the game is trying to play. And it got

Robert: Sure.

Raymond Zhao: so good at it that it started bobbing and weaving, it started like skimming the ground so then the bomb would hit the ground. And it was really it was really cool. simulation that he did, which she posted on YouTube and Reddit. I think this was back in like I wanna say 2017, or maybe 20, maybe 2019. But this was before I think AI really took over as like, you know, the AI it is today. so that was a really interesting thing he posted. And this one GC construction guy in the state reached out to him on Reddit because he posted on Reddit, it blew up. He essentially framed you can turn any. You can turn anything into a game and then you can have an AI trying to play the game and get good at it. So this

Robert: Mm-hmm.

Raymond Zhao: one construction guy who was like, hey, like we have a lot of MEP pipes. Can you try to get the AI to arrange the pipes to push up the ceiling? Because that's a problem with we're facing a lot. Like, you know, how do you optimize the pipe arrangement in these buildings? So that was his first like gig in the construction space. He spent the next seven years just like being a software engineer, building lots of cool like software automation projects and like research like machine learning research stuff for the construction space. And more recently he was working for a design engineering firm called Cisco Hennessy group based in New York actually where he was he was the AI innovations lead. So really me and Brandon and we have a lot a third co-founder called Izzy we came together we've realized the technology is getting to a point where it's starting to solve a lot of the problems that was really difficult to solve in construction beforehand. because it's a very visual, it's a very visual field. you

Robert: Mm-hmm.

Raymond Zhao: can't just have text, you know, it you have to really have agents to understand drawing sets, understand a lot of the engineering of of these plants. so yeah, that was really what kicked off the company. I always wanted to I guess come back to the sort of industry that scared me off as a as a kid. and then he's got a lot of hands-on experience building software for the like for the construction AEC industry.

Robert: Well, that's interesting 'cause you kinda did kind of come back full circle here a little bit, when you think about trying to design software that basically looks for the mer the errors, the misses in the drawings themselves. Obviously it's a big deal. What What happens when AI gets it wrong?

Raymond Zhao: Yeah. I guess taking back a step on the sort of liability, our company is like I guess me personally as well, like way of the view that that should always be in the engineers, like whoever is professionally licensed to do it. they should always have the responsibility. I don't think I can't really see a world where everything is just gone through so the way we've actually set up our platform, it's it's an assistant to the human. It helps you to find errors faster, but a human should always be there. Again, like it beyond like even if the technology keeps getting better, like there's a legal liability, there's like a, you know, life safety question here. And that is like a lot a lot more serious than just Hallucinations and like all the all this kind of stuff. So yeah.

Robert: Yeah, and this I'm gonna put you on the spot a little bit, but you know, this is always the question that you're gonna get, right? It's it's always are you are you sure that that's right? And then who checks it? And and are we sure that that person's right? Because again, it gets it comes back to it comes full circle. Right now

Raymond Zhao: Yeah.

Robert: we're not getting it right. So where does that land?

Raymond Zhao: I I think yeah, I think from what we've seen the industry, really the the picture is right now most construction projects Because of the tight schedules and I think a lot of the structural problems within the industry, QAQC is something you'll s like the senior engineers like really rush at the end. So the status quo is engineers are always complaining, I don't have enough time to quality check everything thoroughly. So it's really I I don't think from like again the way I'm seeing it, it's not can AI completely replace this whole field. It's It's a pretty big problem t the way it is today. If we can help make that twenty percent better, if we can help make that fifty percent better, that's already I think a big win to be had. so like like I think a really like a simple like use case is a lot of times Like a lot of the errors that actually end up being quite serious on the field, are not like very big structural. I mean, some of that also has happened in recent news, but it's it's not really the big you know, structural beam, that support beam that's been you know, like not that that's breaking. It's really like, hey, like it's

Robert: Yeah, it wasn't engineered correctly. Right. No, I get it. It's it's not always the big things, but it's the little things, right?

Raymond Zhao: Yeah. And a lot of things are really hard to catch. and I think, you know, Revit and BIM have been around for a while now, but it's also introduced so much information to modern day projects that I don't think humans are like designed to thoroughly catch all of them. But yeah.

Robert: That's so you've you've just had kind of something that you're starting to fix errors inside Revit. So kind of get so there's so much going on, right, inside of just

Raymond Zhao: Yeah.

Robert: a set of drawings. Kind of take me through your thought process on like how do you approach something like that? Do you do you start small? Do you start like obviously I would think it would have to be incremental approvem improvement across the board. What do you start with?

Raymond Zhao: start really, really small. again I d I do think this is one of the cases of I think the technology is getting there, but it's really like, hey, this is a pretty serious thing if it does go wrong. So like the sort of good example of what we do right now. So We've internally have tested our agents are pretty capable of like doing some pretty complicated fixes of the model. However, we've deliberately sort of gated people like we we don't let people do that. what people can do today is like if you have a spelling error, it will automatically fix the spelling error. If you have, you know, like the schedule is not up to date or like or this, you know, the tag in system is like one-off. You can do these kind of stuff, but then we don't actually allow people to do the like more complicated fix. It's I I think it's, you know, like if if the user isn't using it correctly, it's you know, it can do great good and it can do great harm. So it you know, we I think

Robert: Yeah. So maybe give you the I here's an example. So I was we were doing a very detailed kind of a whole brass I used to build a lot of restaurants, so we did this

Raymond Zhao: Mm-hmm.

Robert: big giant brass bar. The designers had put together all these drawings and then they they got into the detail of of the drawing and how to put things together and I start looking at the detail because I'm leading to generate shop drawings of for this big, you know, thing. And, you know, come to find out that, you know, the channels that they were using and the the mirror and the glass it was supposed to get in, it it was it's it wasn't really to scale. It wasn't exact,

Raymond Zhao: Yeah.

Robert: right? And so then it kind of once you start to see that that was a bit of a fabrication and it it it kind of as they say, look good on paper, but, you know, in reality, like it had a the dimensions weren't correct and so therefore so is that the kind of thing that you're looking for is that you kinda start with the details sections and go, Okay, these need to be here because like you said, you start small, so you go to the detail sections first, or you know, when you start talking about span of an I beam or something like that, well that again we're talking about an engineering you know, problem is the span of the I beam, if I increase it six inches, does that change you know, obviously it changes a bit of the load, you know, and and how much load is on it. Again, th the ratings, th that all seems to change. So like I said, it's so complex for me. I I can't figure out where like where you start.

Raymond Zhao: Yeah. are are you talking about editing or detecting errors? Like

Robert: Well, both because quite honestly, if you detect an error or s you see something, it's kinda like see something, say something, right? You hey, there's something going on here. You want to put it in front of somebody that then says, No, that's correct or can go research and go, you're correct. You know, that's a lot of detail, a lot of work.

Raymond Zhao: Yeah. we actually started I think from the opposite direction. So we started with code compliance and like cross-disciplinary inconsistencies. so I I I I can I can sort of break this down more into the details. but it's ri it's quite cool. So for example, cross-disciplinary inconsistencies, we physically we so you you know how in Bluebeam you like anchor three points and you can overlay different views, and that helps the human to like see what's inconsistent and what what's not. we've built a very similar algorithm where the AI agent automatically overlays different views and it will actually scale like a like a detailed version. like a detail, it will scale like a you know subsection to the correct scale to be able to do that like matching. And then almost then again through like another algorithm, it almost has an X-ray view of like, hey, here's two disciplines on the same physical space. Do I see if there's a like you know, if there's an air conditioned unit in one of them, do I see the same unit in the other one? Or if this unit has this dimension, is that the same size that would fit into the other thing? so that's more on the like spatial side of things, but also on code detection, really we're trying to break the codes into very small like checks that can be run over like the relevant sections of the drawing set to then see like, hey, like is this. up to the code, a lot of times it would be like a parameter that it has to be within and it will go over. And then we then flag that as an issue. So we we have we've actually started from the I think the big things and then more recently we've gone to into the shop drawing review and making sure that's then aligned with like the architectural plan.

Robert: Yeah. So are you finding I I'm assuming you're running this as a as a as a pilot of sort with some companies, is that what I'm finding? So are you finding that they're challenging your models t fairly well and and are they meeting or exceeding your their expectations?

Raymond Zhao: Yeah. we've so we have some people in pilot, but we've actually have qu like quite an increasing amount of our clients have now finished the pilots and they've converted to like annual contracts. Well like you know, multi-year. So they've we've we've seen the completion of like like half a dozen pilots at this point.

Robert: Mm-hmm.

Raymond Zhao: I I think we're doing something right. Every person who Adam Parler has converted into an annual contract at this point. So I do think they're happy with the results. yeah,

Robert: It's good. It's a good conversion rate. Yeah.

Raymond Zhao: it's really good conversion. We worked really hard to make sure we solve their problems. I think really Like the pilots that go successful. there's a strong champion and they're using the tool and then they're then rolling it out to other people. sometimes I think it's more a like I think the tool is powerful enough to change like impact a lot of the work they do. really I think it's in the change management that's where like a lot of the setbacks are. So we've had a pilot where they try to roll out just everything. and again our platform has evolved to like do a lot of like things now. So there Then this one firm try to roll out everything to everyone at once. And then all the engineers are, whoa, like I have no idea what's going on. It's lagging so many issues. I don't want more issues to look at, like I'm being overwhelmed. And then we were like, okay, let's, you know, let's not overwhelm them. Let's think about what the lowest hanging food we can solve. And then once they get familiarized with this tool, we can then introduce more capabilities one at a time. So that's what That's what they do. And then I think because we do work closely with our clients, we also set expectations of like, hey, this is what we want to achieve in this pilot. so yeah, so we we do we we do that pretty well, but I do think like it's mostly like how do we get people who are used to Bluebeam Revit and nothing else, or sometimes not even Bluebeam, just you know, to go into a new tool and for it to feel not like it's they're spending more time trying to get used to it.

Robert: So you bring up a good point. you know, with this with this tool, as with anything, you start to You start to create a lot more work for people. and it now hear me out though.

Raymond Zhao: Yeah.

Robert: You you you create a lot more work for people, but it's work they should have been doing possibly in the first place. But what happens is is they get this it's almost like information overload, right? Where they're like now I've got instead of having this list of, you know, five things to do, they've got a list of a hundred things to do and like For me, I just shut down. I mean, after ten things somebody gives me a list of ten things to do, I'm like, just please don't give me a list, you know. I need the pick your best three.

Raymond Zhao: Yeah.

Robert: Right. So when and I would suppose that you're getting there like priority wise, does this are you able to scale this, you know, as far as you know, pick it out, say, Hey, you know, pick my top three kind of thing and then and then work on those? Or we just go with like a page at a time, I mean, I don't know.

Raymond Zhao: Yeah, definitely. I think different firms have different preferences on this. So we've seen clients who are like, Give me everything, like quality, like it's really important. Like we we get bunch I mean, it's I don't know, it's it's I think it's a very interesting question you're asking because it is a lot of task to be done now, but it's like if you don't do it You remember what happened last time where you got a massive like you know, like R you got twenty RFIs and you so

Robert: Yeah, right. You could either have a list of of you know, a hundred things to fix or you could have a list of a thousand RFIs. Which I mean, which one do you want?

Raymond Zhao: Yeah. but we we we do we do customize this to each client of like so I guess okay, to maybe say another like direction we care about a lot is not just is this issue correct or not. It's is this issue relevant to this person. So we've spent a lot of time to work with our clients like, hey, like what would you like to see? So some firms really care about the design, side like the the d design standard, cat standard, beam standard, whatever. And some firms like I don't I don't care. Like our clients are not gonna get that pissed if it has a spelling over here now. If it has an engineering issue, that's a big problem. So we try and like figure out you know what really matters to different

Robert: Yeah. Priority.

Raymond Zhao: Mm-hmm.

Robert: Right. Yeah. So one of the things I always thought about was the translation is I think it's a it's a it's not really a technology problem, it's a human problem. and the human problem is this. so you can get it done faster because you're utilizing said technology, so you know, get it done faster and I only have to pay this much. do you find that that's gonna be a a problem, you know, where Is it is it just going to shorten shorten the length of assumptions that are going to be made, like, well, you're using that AI product on this thing, it should be able to find all your issues, just fix right? I but it's that's it's not that simple, right?

Raymond Zhao: Yeah, really, really good question. I think beyond that, there's an even bigger one, which is a lot of these consultants gets built hourly. So if they use this tool and they finish a project in half the time, do I now just have to build less? And then it's like, well, am I really saving money or am I just making less money? so I I I do think there's a lot of ways. like, you know, there's a lot of like more s industry, how money flows, how like client relationship works that is at play here. the way the way I see it it's like It might eventually get to a point where projects have been delivered a half the time, but then I think you can like as a consultant you can then be like, Hey, I want to get built a little bit more because I'm saving you a lot more money than like how much I d I do think it's it's sort of a conversation where the the contractor whatever has to almost like fight back and but you're getting better quality work and your real work cost is being halved. So, you know, even though we use AI

Robert: And I always so and I always thought i and it could be my error, but billing by the hour, but they didn't have a budget, right? There is a there's a okay, we're gonna give you sixty hours on this, it's gonna be this much, you know, and then I mean, is that still a thing, with these AE industri you know, the the architects and the engineers? Is that still they still or are they just billing by the hour saying we're just we're We're billing by the hour and we think it's gonna take this much.

Raymond Zhao: Yeah, I mean yeah, so a lot of times there is a budget. so yeah, a lot of times it is like, you know, and I mean yeah, there's many ways like consultants bill nowadays. I think it's becoming more and more just like diversified. but yeah, I think yeah. I mean it's that's one thing, but another thing it's like great, like Maybe like, you know, this project has not gone that well. We're behind schedule. and there's lots of change orders that's come in, like, and we've used up our like, you know, fixing, like free fixing time that we had in the contract.

Robert: Sure. Yeah.

Raymond Zhao: Now who's paying for it? And then that's when a lot of the like I think ugliness comes out and everyone's trying to like push it to the next person. but yeah.

Robert: Yeah, well see the the ugliness comes out with it the guys in the field are like, These drawings suck. and you've you know, we've spent hours, you know, reviewing your drawings and we've spent hours in RFIs, we've spent hours in cha having to make changes, having to, you know, maybe remobilized to certain areas because these these weren't thought through properly or whatever. Every set of drawing I ever got was probably maybe eighty percent complete.

Raymond Zhao: Yeah.

Robert: and there was a lot of lot of fixes to do. But your product has man, it's got a lot of upside to it. but it it has the it has the structure to it feels like to me to kind of redefine, you know, how we we look at the drawing. and and what it means, right? To say, okay, may maybe now the drawing set is ninety-five percent, right? That drawing set's ninety-five percent. We've gone through it, we've used structured AI, we've, you know, figured out where most of the problems are, and then all of a sudden now we got a 95% and then there's a problem in the field. Right. Then what happens? Right. Well that did we really solve it? Or is it something that went wrong in the field? it it cha it kinda changes the game a a bit, doesn't it?

Raymond Zhao: Yeah, that's I think that's the exciting I think future where a ninety five percent means the same across everything and then that's you know that's then the standard and it's it becomes a bit more clear who's responsible in sort of everyone pointing fingers at each other every time something goes wrong.

Robert: Yeah. So do we think that this Are we is it gonna do we think that this product, I guess, as well, will be able to accept like feedback from the field, like, okay, the drawing went this way, we did captured reality here,

Raymond Zhao: Mm-hmm.

Robert: you know, like a three sixty. Will it be able to accept, you know, kind of historical information about the build as to, you know, was this right? Was it again be it's kind of a self improvement feature?

Raymond Zhao: Yeah. Yeah. that's actually something we've been like looking into a lot. That's sort of like an exciting direction on the roadmap, which is again, like yeah, it's on the field or like a lot of the construction projects nowadays are like renovation and rebuild project. It's not from scratch. So it's like, hey, here's the old building, here's the plan from fifty years ago, which is not at all up to any code to standards because the standards have changed.

Robert: Yeah, yeah.

Raymond Zhao: But here's the real physical building and we've reality captured, we've transformed those into like a sort of like a CAD file. And then can you then compare that to the new stuff you're trying to do to see if stuff matches up? I do think that's like a really interesting direction. and another another one which one of our clients we potentially going to partner on, which is really cool, is so we're working with this very like AI native luxury home builder in Australia.

Robert: Uh-huh.

Raymond Zhao: and they're like they're like very like fully on it like they're using claw chat gpt like google everything and they've set up again this is a little bit maybe like black mirror but they've set up cameras in each of the construction sites to see like issues on the construction side real time and they're streaming that in and they're like hey like we have all this data can you somehow like connect this to your platform's intelligence to be like, hey, like we feel like the builders are building it wrong because we have the plans and we're seeing we're tracking what they're doing and then we're st we'll we feel like they're not building it to what the plan is drawing. And then that should be flagged to like the GC or that should be flagged to someone to then just like troubleshoot in real time. So hey yeah that's you know these are really exciting directions that like we we're we're excited to explore.

Robert: Yeah, that's so I I guess the way that I look at it is we we collect we collect a ton of data. We've been collecting a ton of data all this time. and and we've created the feedback loops, you know, with the with the RFIs and the and the and

Raymond Zhao: Mm-hmm.

Robert: the you know, those kind of things. And it seems like still nothing changes. I mean, still there's an issue. And I don't know, do you think it's a is it is it a technology issue or is it just a human behavior issue a lot of times, which I kind of feel like that's what it is. no matter how much technology we put in front of somebody, we still have some real bad habits.

Raymond Zhao: Yeah. I I think the human aspect is massive. like your average project, I think each subconsultant has some somewhat of their own timeline and their standards. and very often It's I think it also comes down to the way that industry's been paid. that different stakeholders just care about they have different risk liability levels, so they care about stuff differently. we'd very often like we would find an issue on a plan set and then they would be like, that's like the that's the MEP team's part, so not off our hands. so really like, you know, then it's like, great, like, you know, it's you're you you found an issue but then you're not really you don't really prioritize it because it's not your part of the issue and then it I think yeah like that is like a pretty big problem I'd say and then I think some somehow it is because the way liability has been spread around is it's it's just not very productive for like the completion of projects, I'd say.

Robert: Yeah, I think that's an interesting point because at at some level, depending on you know, who's using the tool, that becomes you know, so like the MEP portion. Let's just say you've got your mechanical drawings. you've utilized the tools and we found out that there are specific errors. I I l may May three three years ago, three two years ago, I was at I was at the an actual mechanical it was kind of a Revit convention, or

Raymond Zhao: Mm-hmm.

Robert: blue beam conveni you know, the lots of lots of V D C folks there. And they all kind of said the same thing. kinda to your point, it was, well, we can find a lot of stuff, right? And we already know that there's gonna find certain amounts of problems. But as soon as we get it out in the field, th the It didn't matter if we knew the problem was coming or not. the the the field just totally did it their own way or it it you know, it it it it we had the wrong information from the field, and so it didn't work out that way. It's this weird communication layer that we're missing. I think it was, you know, kinda to your point before if you got like this remodel project. One of the biggest problem with the remodel project is is that you don't really know existing conditions. there is no such thing as an as built, you know, of some fifty year old drawing that nobody ever updated or you know, whatever. So I don't know. I the the complexity layers here just astound me. and but what all what what always seems to come down to is that human interaction with the technology that makes it worthwhile.

Raymond Zhao: Yeah. No definitely. I I somewhat of a tangential point. but I think what's like I don't you know, maybe a a small part of the reason was up until again this might be like a pro AI talk, but up until really like a gentic capabilities have been around recently. Every software tool was another tool that a human had to sit down and click buttons on To help them somewhat automate a workflow. And the extent of things you can do is very limited because like Like you know, like take our example, QQC. Every city, every l like there's so many different codes that before you had agents who can like intelligently understand that, hey, like this two codes are different, but then really the fire safety principles fundamentally are the same, but then just they've just been upheld to different levels of like strictness. it was really hard to really build a lot of tools that because construction like each project is different, there's so much like heterose like you know d it's so Not homogeneous, like you know, like it's like it's so many different variations. It was quite a difficult industry to build software for, traditionally, because you will build it for this one workflow for this one way, and then the next company's like, we do it differently because our clients want it this way. And then that you know, that software isn't useful anymore. so really was for the first time we've seen software actually being able to accommodate to the way the industry works. And I yeah, that's why I think like some breakthroughs will happen.

Robert: I think you're right there. I think one of the biggest things is the adaptability of the adaptability of a large language model or a or an AI product of some sort where as before you needed somebody or a a person of some sort to be able to translate to a developer, by the way, this is what we need. And then that developer then had to take that translation and turn it into some sort of product. And then that iteration comes out and is like, hopefully that helps. Is this what you were talking about? And then the the loop is very slow in that, right? Because it takes a while for that person to translate and fix that problem and they get it out to the and then by the way I forgot about this part. It's almost like the product manager in the in the you know in the in that loop just that it never quite got translated correctly and then it everything lost in translation then comes back as a as a problem. So

Raymond Zhao: Yeah.

Robert: that that's how we ended up with like legacy software, right? and I think now what we're looking at is is these AI products now going to challenge that very process in itself of it's not going to be the developer that needs to translate. it's going to be the foreman in the field or the the person out there that says, it needs to do this, this, and this, and then the AI translates it and says, this is what they're looking for. This is what part of the drawing that's incorrect. It's almost like at reverse, instead of AI trying to figure out where the problem is, it more of more of the the product trying to listen to the input of a builder, right?

Raymond Zhao: Yeah.

Robert: And then understand that and extrapolate that into a an answer or this is how this should go together.

Raymond Zhao: Yeah. yeah, I d a hundred percent like so one of our clients, they have a really cool use case where they just talk to the so they don't They also use sort of the QA, but they're like really they like, hey, like we really like the ability to talk to the plants and just like ask, you know, ask, are all the water tables in the schedule? Like are you know, are any of them missing? And then just that ability to really quickly query stuff and then get an answer back is so valuable to them. And yeah, and exactly to your point, it's that ability to just translate what they're saying in like plain language into like again, like software, you know.

Robert: Right. Well, okay, so I just had this weird just thought, but almost as if the agent itself that you create is let's just say it is the building. Right? The agent

Raymond Zhao: Yeah. Yeah.

Robert: knows everything about the building. It has the drawing sets in there, it has all the codes in there, it has everything it needs, right? And then you're asking the building questions, by the way, I want to put another story on. How do I again

Raymond Zhao: Yeah. Can I do that?

Robert: How are we going about that? What does that drawing look like? Right? Give me the instructions back rather than it becoming a thought of you're you're giving you're building the building or whatever, you're designing the building. No, the build the building is the agent. And the agent just says it looks like this and this is how it goes together.

Raymond Zhao: Yeah, that's a great analogy.

Robert: I mean i 'cause when you started to say talk to the you know, talk to the the the plan or whatever, well that's really what you're trying to do. Like I wanna know, let's see, we're putting in six doors on, you know, floor number, you know, five, right?

Raymond Zhao: Mm-hmm.

Robert: Where are the drawings, where are those and you know, it looks like we got the wrong hinges or whatever.

Raymond Zhao: Yeah.

Robert: It's again, it's like it would have all that information.

Raymond Zhao: Yeah.

Robert: so the agent itself actually becomes the I I would say almost the the mind of the asset.

Raymond Zhao: Yeah, I think this is such a cool an analogy made. Because the industry sort of has evolved from two D plans, like handron plans, two D to now, you know, three D, BIM, like digital twins. But that I think the problem with that is like it's really hard for people on the field to access like these three D models and like just get it to like work in a nice way on an iPad or on your phone. But now it's really like, hey, like the agent captures all of this information and you really don't have to click around all these like tiny buttons trying to like figure stuff out. Like you can just ask it and then the agent Again, as the asset will just reply to you with you know, the information that's useful.

Robert: Right. Yeah, that's I mean that's how I see it because again I've when I work with the stuff that I do with with my AI agents and then the the things that I research, is like I try not to think of it as it's it's a I'm doing that's another person that's doing this. As it's a totally different entity. It doesn't work like a person works, it doesn't think like a person thinks if you if you want to say that it's thinking, right? Doesn't

Raymond Zhao: Mm-hmm.

Robert: It's not that's not it. It captures the information and then truly your your whole job as as the human element of it is to prompt it in the correct direction to give you the answers that you're looking for. It it the answers are in there. It's like a set of drawings. A paper set of drawings has all the answers in it, right? Even if they're the wrong answers, there's still has a set of answers, but you've

Raymond Zhao: Yeah.

Robert: got to go searching and looking for Now we're at that point where I ask you a question about yourself, you could tell me, right? I say, what's ailing you, Raymond? What's what's wrong with you when you know I've got a cold, you know me, or I

Raymond Zhao: Yeah.

Robert: you know what I mean, or I'm feeling great today, whatever. We're almost at that thing to give we're at that point where we give our buildings, our creations, a mind of their own a little bit.

Raymond Zhao: Mhm. And I think that would be really powerful 'cause you mentioned communication. this is off almost offloading the communication burden from the humans of like like this asset itself can now communicate in a very like consistent way. So then you don't need to rely every time on I have to call up an architect to figure out where he put this thing. You can just ask the building and then it will come back and tell you. I mean obviously I think there's still cases where you wanna call up your architect and ask like I I'm still confused. I think something's wrong like majorly. But yeah like it skips a lot of the reliance to have to go to another party to figure out stuff like if you can just have this entity that everyone can just ask questions to and then it will give the correct answers back.

Robert: Yeah, well it's a it's a little froofy out there. It was just it just kinda popped into my brain from what you said, like they want to talk to the set of plans. I'm like, okay, well that seems it seems logical to me where you put all the plans and you put all of the the codes and you put all the specs in it. and then it just i I think what your what your program does now, it takes that, it takes all that information and looks for where they don't match up.

Raymond Zhao: Yeah, exactly.

Robert: Right? And then they don't match up. Why? Do we fix it? Do we not fix it? You know, is it a you know? So you know, we just keep the you know, keep the inspectors out of the way, we'd probably be all right. So that's one of the things

Raymond Zhao: Yeah, exactly.

Robert: that I think your product may have a great space for is helping the inspectors and the people that are looking at drawings, helping them answer questions for like these design reviews, and those types of things. one of the bidd biggest bottlenecks we have in construction is getting approvals. Right? Getting a plan set approved, getting somebody that has to go through the documents. Well, why do we need to go through that? You know, we're in the age of we really shouldn't have a human being going through documents putting stamps on, you know. when quite honestly we should be able to take a a product like yours and say, Here you go, we're gonna find out, answer these questions. Again, talk to the asset. It's like

Raymond Zhao: Yeah. Exactly. we yeah, so we I mean it's not public here, but we we are working with some like more government like it's I g we we have like this project with a I guess on the government level, like then

Robert: Uh-huh.

Raymond Zhao: sort of the building authority level of like, hey like like We want to start automating like a like a like a checklist of hey, if it doesn't pass this, it should then just get flagged back. It shouldn't have us going through each one of them manually. And yeah,

Robert: Right.

Raymond Zhao: so it's I think yeah.

Robert: Yeah, I think that that does of that's of huge value. And and I'll I'll and I'll tell you the reason why. We are in the age of AI and we're going to need AI to help build more AI.

Raymond Zhao: Yeah.

Robert: and if we can't get it built fast enough, we're in real trouble. and it and it's simply it's it's more of an economic thing than it is, I would say, a a security thing or I think. I was watching this what some documentary the other day we were talking about the the race is not towards you know solving AI. The r you know, the race is towards agentic AI, you know, AGI, I guess, artificial general intelligence, true AGI. And then and then it is expands, expands from there. But it's right now we've got an economic spend of, you know, some seven hundred billion dollars kind of hanging out there. But we don't have the we don't have the ability to power any of those data centers. We we don't have the ability to to get those things built. One, because we can't get it permitted fast enough. Two, because we don't have quite enough people to build it all. And and three, is the materials are on a long, long, long wait list. Like products like this and ideas like this where we can speed up the areas where we don't necessarily need to be involved. so much anymore, would really kind of benefit us in a long term way because those hold costs on seven billion dollars are really, really expensive. And we don't want people dropping off, right? We we want people to keep spending that money.

Raymond Zhao: Yeah, definitely. we yeah, so we we we have a few clients who build data centers and yeah, like I think this is like a really interesting, like you need this platform to help build more data centers quicker so then like more AI can be trained and then it's like almost like each like again, it's each layer is depending on the other one. And I do think increasingly it's I w people are seeing this as like an infrastructure problem of like again materials and then like construction. so yeah.

Robert: I s I'm still you kinda just off the wall question here. As as you know, I'm just

Raymond Zhao: Yeah.

Robert: what do you think okay, what do you think's worse? the the the economy without AI or the economy with AI. because you know there's a lot of pushback, right? You see it all the time. I mean New York just put a ban on data centers over a certain what is that 10 gigawatt or you know 10 10 whatever I can't remember what it was, but they've just it's actually a ban on natural gas, but you can't even power anything up. But you got some people totally against it and the other people that are like absolutely for it. And they're they're both button heads and either way, the economy's gonna be affected one way or the other.

Raymond Zhao: Yeah. not too sure on the like economic I'm not a e e ecologist, but on the just like I think a really interesting analogy I always like to make is when like the industrial revolution came around, someone was like, like everyone will be out of jobs in the next fifty years because machines will just take over like Like machines will just take over all the need for human labor and then everyone just can just like stay at home and there's no more jobs. and I I do believe we've reached at a point where if society was structured in that way, then yes, like we could have that world. However, it's not how sort of the world has evolved, right? Like Like I don't know, like sports as example has become a massive industry now. Sports was not as big back then when it wasn't like an important aspect of life. So I do think like with AI, and it's always hard to imagine what that would look like into the future when you haven't really experienced it. But I do feel like given how we've evolved with like industrial revolution, with the internet and now with this new technology, new new jobs get invented, new stuff becomes, you know, relevant. Maybe some old jobs will become obsolete over time, but then like I think economy as a whole will probably boom because of this on the long term. And I have no idea what that will look like, but I I have a I have a feeling like we will have this new industry in 20 years that seems impot like just crazy today. But then

Robert: Yeah.

Raymond Zhao: in 20 years that's like a massive industry.

Robert: Yeah. Well, I agree with you there. And I do agree, I I do think that if you're going to be in construction, I just had this same conversation with the guys from from CavTech, or Cavterra. but I do think if you're gonna be in construction, you're you've got to be technically savvy. You're you're going to have to be engaging with technology construction companies. that if you're gonna be a if you're gonna be a construction company, you're going to have to embrace technology, use it as part of your vertical. it has to be you kind of hand in hand. I don't think that I don't think the old ways are gonna gonna work anymore. you you just not you can't be efficient enough in doing it. it could be a nice hobby if you wanted to do some construction on the side or, you know, whatever that's a nice little hobby. But if you really want to be successful and you want to be competitive, it you're going to have to

Raymond Zhao: Yeah.

Robert: engage in utilizing AI you know, straight up. I don't care what business you're gonna you're gonna have to engage with utilizing it to its to its fullest. meanwhile still keeping your human element to to to keep it to keep it grounded, I think.

Raymond Zhao: Yeah. No, definitely. I mean I think yeah, you know, it's the same like innovate or die, like I think it's becoming quite transparent that AI is a technology that has like immense value. And then I mean like d I I you know, I don't know how you feel about this, but twenty twenty six this year, the construction industry's perspective on AI has changed so drastically, so quickly. I remember like even way back in twenty twenty five, a a lot of people were like, I'm skeptical, I don't like you know, I I don't I don't wanna do this, I don't wanna touch it. but now everyone's like, like AI is here to stay, like how do we get on top of this? How do we best like utilize AI? What does AI actually

Robert: Yeah.

Raymond Zhao: mean for us as a firm? So I I'm really it's it's so cool to see the industry like picking up in the sort of adoption cycle and everyone starting to like actively talk about how will AI change construction instead of

Robert: Yeah. Yeah. So more important in into that, as we kind of wrap up here, you know, it is important that we embrace AI cautiously, I think. keeping the human element involved. But we've gotta get it done quickly. we have got to get these and not all the data centers built, but we gotta get these ones built where we're actually advancing these models so that they are getting better and we're not we're not over what is it w we're not giving compute too much room to be of such high demand that the cost balloons on us. because that's where we're headed. We're y we're getting to a point where compute becomes so expensive because we just can't power it that it becomes you know the end of tanium of of the future is like we so we have to find a balance. but I love products like yours, Raymond. I love guys like you who are thinking, you know, kind of way out there and and and engaging. I appreciate your time with me today. Thank you so much.

Raymond Zhao: Thanks so much for having me. This was such a cool chat. All right.

Robert: All right, folks. That's this episode of Contech Exec Podcast. If you guys get a chance, check out Structured AI. check out Raymond, he's on LinkedIn as well. You could probably reach out to him there. I don't know if he'll answer, but he might. You never know. But

Raymond Zhao: I'll I'll I'll ask. I'll try my best.

Robert: All right. Well, until next time, everybody be safe out there.