Rob Stott:Â All right. We are back on the Connected Design Podcast and excited to sit down to talk about just, you know, a simple subject, the one of AI, right? It’s just a simple topic that we get to dive into. don’t know which way it’s going to go. We’re going to find out and have some fun along the way, but I appreciate Mr. Nick Donahue, the founder of Drafted, an AI platform that allows users to design their dream home on a platform that’s built around what else but AI.
So Nick, I appreciate you taking the time and jumping in and wanting to do a podcast with us.
Nick Donahue: Yeah, thanks for having me.
Rob Stott:Â You bet. So I know we didn’t specifically connect. I want to start to where I was sort of introduced to your brand though, before we dive into it more holistically, but we, I met your brand at the Builder’s Show in Orlando just a few, I say few weeks, it’s almost a few months ago, man time has flown by, but, I, want to give you a chance to talk about, that was kind of, I believe your guys coming out party, right? Being at the Builder’s Show and introducing your brand to the builder channel and all that good stuff. So what was that experience like for you and why was it important for Drafted to be at the Builder Show?
Nick Donahue: Yeah, I guess to give a little bit of overview. So our goal is to make changing the physical world as easy as changing a website. We currently focus a lot on single family home design. So allowing anyone to design a house unique to them. We have a large portion of our user base that are home buyers, smaller portion that’s builders, architects, designers. So it’s really just a way to like expand your imagination to be able to see what’s possible as a home design and have that come to life way more quickly so that you can iterate on your concepts without necessarily having to draw a bunch of lines and model a bunch of houses just to get an idea of which direction you want to go in.
We ended up doing the Builder Show because there’s tons of high quality builders there, people that are just exploring new things, trying to see what’s out there. So I think naturally the environment’s a lot more forward leaning than maybe most of real estate. And so it provides a good opportunity to get in front of these different types of builders that could possibly use the product, maybe meet some designers. Like we actually had some users come by the booth, which was really cool.
So yeah, was our first time doing IBS and really good experience.
Rob Stott: One you’ll go back to, you think?
Nick Donahue: For sure, especially in Vegas.
Rob Stott:Â Yeah, that’s a fair point. Last year in Orlando, I know a lot of people were talking about the excitement around that, you know, getting out to Vegas and seeing it there. It’s been there, but you know, it hadn’t been there in a few years, so it’ll be exciting to get back out there for sure. Any moment from the show kind of stand out to you? Was there one interaction or even like, you know, a brand that you came across that you weren’t expecting or a technology or something like that?
Nick Donahue: Yeah, I mean there were a few moments that really stood out. One was we got to meet the COO of Co-Construct, which is like a bit further downstream, know, project management software. But it was just really cool to be able to show them what we working on and discuss a bit more. We got to meet their like CPO and CTO and their head of like Biz Dev, which was really fun.
There are some other startups in the same area that were doing kind of like tangential things that could easily be like collaborations. Like some would take like an initial plan set and turn it into like a structural plan set. Some would evaluate it on buildability and the build area for a lot that then you could design on so the fact that we sort of plug into the design portion of that process and can take in some of those inputs or you know after the fact sort of post-process that I that was really really interesting really cool and then beyond that just seeing like people actually like search us out like for a very new brand but to have people like man I’ve been looking for you like I finally found you it was really cool, and some of our users coming by as well was really cool. So getting to meet people more in person, you know, we are a digital product, an online service, so I don’t get to interact in person with a ton of these people. So it was just amazing to be able to do that.
Rob Stott: I always find that to be, know, never really been a brand that like is at the show exhibiting per se, but I always find those interactions to be sort of unique and exciting, right? Is to hear like yourself, you were there to potentially meet with and find new users of the platform, but then it’s the other brands that you’re near that it’s like, we could have some really cool, you know, like business development opportunities or collaborations and things like that. So that’s a, the heart of being at a show or the, reason for being there is always more than just the potential clients and things like that. Right. So that’s cool to hear that being, even being so new, you guys experienced that.
Nick Donahue: Yeah, I agree.
Rob Stott: So talk about, you know, the show’s awesome. I look forward to kind of seeing where you guys go as a platform and things like that moving forward. But, tell me a little bit about, for those that don’t know your background, you know, but what got you to where you are today and, interested in the space. Cause you know, it’s a unique industry, for sure. And then obviously being able to launch a, pretty unique tool within it. but just give us a little bit of background on, who Nick Donahue is.
Nick Donahue: Yeah. So I mean, I grew up around the home building space. Both my parents were in it. My dad built neighborhoods for D.R. Horton, Pulte, some of the larger builders. My mom ran the Eastern sales team for Home Depot Supply, so selling flooring and countertops to big box builders. So I’ve been around the space for a long time, but didn’t really get to dive into it a whole lot until I started my last company. So I’ve been, I guess, a serial entrepreneur in that where I did a tech enabled pre-construction service that would take an empty plot of land and a consumer wanting to understand what they could do on that plot of land and help them actually figure out what’s possible, plan that whole process out. working with structural engineers, civil engineers, surveyors, soil testers, and then ultimately just using design as a medium for helping people conceptualize and visualize what was actually possible.
In that process, we were automating away different portions of it from estimating site development work, built a bunch of tech around that. But the core piece that was very unscalable was the design portion. So how do you take this very expansive design space and simplify that down in a way that anyone can interact with and make that a lot faster so that more people can design houses? Because today, there’s 5 to 10 percent of houses in the US get designed by an architect. Most of them are like templatized with big box builders or everyday builders or bought online like a house plan website. And so we’re just trying to bridge that gap, figure out how can we make it cost less, drive down the cost enough and make the time faster that more people can do it. And seeing that problem firsthand in my last company and experiencing that with consumers all the time made it seem like a good opportunity to see as AI started emerging as a new technology that could be used to solve extremely complex creative problems. So now we’re just training our own models, teaching them about the physical world, teaching them about the design space, and trying to help people design houses much faster and much more cheaply.
Rob Stott: No, that’s awesome. I, well, I know lots we can dive into there. before we do that, I want to follow up on the fact that you got to see the home building industry growing up and you weren’t scared away by it. that I can talk about that a little bit. Like you were actually intrigued by it, watching, kind of what your parents did and, decided like that this is, it was interesting enough to you to, to want to pursue it.
Nick Donahue: Yeah, I mean, it’s the most important thing in people’s lives, like beyond your kid or kids and family and maybe your work, like your home is like the representation of your life for most people. So it’s like to be able to have an impact on people in that fundamental way, it just seems like a problem worth solving and something worth working on. Obviously it’s extremely difficult of a space. There’s so many moving parts, extremely fragmented, very slow to adopt new technologies, very old in its ways. So that makes it a lot more difficult, but I’m hoping that we can encourage more people to want to adopt new technologies.
Something that was interesting growing up was seeing my parents go through the 2008 recession, which was very chaotic for a home building family. I’m sure many, many others experienced and I thought that that as well provided an opportunity like right now when markets get really hot builders start to take on more leverage, to speculate more in their builds, which then just gets to a point where as the market turns, things swing so heavily in the home building space. And if we can get more consumers to make those decisions on what a house will be and less speculation happening, then I think that will help in the process of not having such drastic rise and falls in the industry.
Rob Stott: Yeah, no, I love that. And I mean, do you credit mom or dad more one way or the other with that interest? Because usually when I think of parents and wanting to get into an industry or even just growing up around your parents, I was like, I’m doing this so that my kid can have better or avoid the industry that I’m in. Did you experience any of that or were they encouraging of you of wanting to get into this space?
Nick Donahue: Yeah, I mean they were super excited when I started getting into it. It was kind of like a yeah, my kids like doing the same thing Exactly and and more things to talk about you know I think getting into entrepreneurship more broadly like that wasn’t a very familiar thing for them and so you know…
Rob Stott: That’s cool. That’s cool. A little pride for them, right?
Nick Donahue: It felt like as I was getting further in my entrepreneurial experience, they were like, I don’t know your world at all. I don’t know really what’s going on. This whole tech thing is sort of beyond me. So there became that aspect. But just the fact that it was grounded in something they were familiar with makes it like when I do go home for the holidays and stuff, have conversations. I can show them the product. They can literally try it and be like, maybe you could do this. Maybe you could do that, which is really fun.
Rob Stott: Right. That’s, that’s exactly where my head went, was like, you know, before you even have actual customers, you have your parents to kind of throw these things up against and see how they react to it or how they’d use some of these tools. And those have to be some pretty cool conversations.
Nick Donahue: For sure. The irony is my mom’s probably the most critical user there is. yeah, it’s interesting.
Rob Stott: No, that’s awesome. So a lot of, uh, whatever updates are coming down the road, Mom, she played a heavy hand in, that right?
Nick Donahue: Sometimes I’m like, gosh, should I really show it to her now?
Rob Stott: No, that’s awesome. Well, you mentioned it too, kind of in there in that, you know, technology around this space is not something you see often, right? As far as how you’re trying to implement it too, right? You know, it’s a very hands on hand. You think about the history of just architecture and design and, you know, home builders. Yeah, they’ve got printouts and whatnot and blueprints they fought like, but really it’s a hands on business. So the fact that you’re bringing technology into it and it’s something that they are uncomfortable with, is that a hurdle that you’ve, you know, talk about? I, I ask it. Yes, it’s, I know it’s a hurdle that you’re probably overcoming on a regular basis. So what are those conversations like as far as getting people to be comfortable with the idea of adopting a technology, you know, within their business?
Nick Donahue: Yeah, I mean, it’s been extremely surprising, even down to the user interaction level, the way in which you interface with the technology itself.
We end up having to simplify things quite a lot. Something may be very intuitive to me, because I’m familiar with tons of different types of softwares, and especially design softwares. But then to bring it to an everyday user or bring it to a builder, we end up having to really think about, how do we make this as simple as possible? Because there isn’t a familiarity to certain interaction types. So it’s been, from a product perspective, a very useful, unique experience. To me, I didn’t really expect in terms of the actual integration itself. I think when people hear AI, their first first glance is like hesitant, like very hesitant. Partially because there is this mentality or this like propaganda going around like architects are cooked, interior designers are cooked, like everyone is cooked, like everyone’s basically done right now. And it like feeds this sort of aura or mentality where you’re like, well if we’re at war then I’m not adopting anything like like forget that and so I think the more People can share how like it’s actually useful in their day-to-day Outwardly is I think that’s like very very helpful Because you know at the end day like for us We’re just looking to collaborate with people like figure out how we can make the product as useful as possible And so having that sort of pushback because of the narratives that are going on right now because it’s such a new technology and so disruptive to everyone that it’s just very, very scary. So it’s like how do we overcome just that alone, which has nothing to do with the product, has nothing to do with anything else, other than just like how they, the perspective they have going into it.
There are like definitely the early adopter types. I’ve met people that we end up interviewing as they use the product for the first time. And just the way that they view it is very different. They’re like, this is inevitable. It’s going to happen. There’s really no way around it. So I need to get on board. The earlier I get on board, the more beneficial it is for me and for my business. So let me actually work with it and try to figure out how to make it useful. by being a part of the process, think people feel a bit more comfortable.
It’s less like scary because like they’re actually a part of the change and it’s less like this thing’s gonna steamroll me and take my job
Rob Stott: Right. I blame RoboCop personally. You know, people that watched it and they say like, this is just gonna, it’s gonna wipe us out. And then, know, you had Stephen Hawking way back when saying it’s gonna kill us all. Like, no. Yeah, we’re not there. I don’t know if we’ll ever be there. You know, never say never. But they certainly weren’t helping those early conversations around AI and where it’s heading. But do you have an experience of taking someone that is sort of…
They have that feeling of fear around the technology and, you know, exposing them to this and what you guys are doing and converting them.
Nick Donahue: Yes. There’s a design firm owner in Nashville, Tennessee that we worked pretty closely with the first time. And this is the case for like most designers that I onboard. They’ll go through the product for the first time. At first, it’s like super skeptical. They see all the flaws in the model, know, anything it does incorrectly, just like rationalizing why it’s not there yet. Just like, oh, it’s not here yet. I’m good. I’m good. We’re safe. And as they get into more of the interaction types, like we added the ability to like define a shape of a house. And a lot of times when people define shapes, especially designers, there’s usually like moments that make sense for like a garage or a certain room. And you can sort of see it in it if you know what you’re looking at. And a lot of times the model will then like place a garage in that exact position, which they didn’t tell it to do or anything of that sort. And there becomes this just like undeniable, you know, this thing is intelligent, it is working, it is going to keep getting better. And that process then at first, is like pretty fearful like literally on the call it’ll go from them like talking for like five minutes straight to like dead silent like nothing said you can just sort of see the processing happening where they’re like oh my gosh what is going on it’s here oh no…
Rob Stott: Stephen Hawking was right.
Nick Donahue: And then finally there becomes a moment, you know, five minutes later where they start like, okay, how, where do I go from here? Like what’s sort of the next step? And that exact design firm owner I’m talking about ended up wanting, wants to join the company and like actually wants to help us figure out how to make it more useful and build it out and has been extremely, extremely helpful with introductions and user feedback. that it’s, I’ve seen this sort of process happen on nearly every call at this point.
So yeah, it’s happening.
Rob Stott: Incredible. It’s again, it’s just the idea of they need to understand it and put their, you know, misconceptions or preconceptions around it aside for a minute and understand it and learn about it and be open to it before they finally can like understand it and see the opportunities that exist because of it. We’ve kind of talked a little bit about it, you know, here and there, little pieces of it. I want to give you the chance to really explain what Drafted is and what it does and give a little bit more of a solid overview from your seat as far as what the platform is.
Nick Donahue: Yeah, currently it’s focused pretty heavily on like the schematic design phase. So that’s usually the start of any design process where someone is really just trying to figure out the path they want to go down. Oftentimes a designer or someone will present like multiple pathways, but most of that work gets just like completely thrown away. And so they spend hours and hours of time coming up with these different concepts. And so now whether you’re a designer coming in, integrating that into your workflow, you know, creating a—you can give basic inputs like room types, the footprint design. We’re about to add bubbles. You can set lot constraints like boundary constraints that it needs to generate within. And then it’ll generate a spread of nine houses within probably, you know, anywhere from 10 to 10 seconds to a minute. You can regenerate as much as you want. You can change the different input types and then just see the process happen.
It’ll generate both the 2D floor plan plus a 3D model. So you can see it in 3D, spin the house around with all the houses, which just really helps you understand very quickly, like directionally, which pathway do I want to go down? Even if this isn’t perfect, like it just gets the juices flowing. Like there’s plenty of times I’ve noticed something that I wouldn’t have thought of and then that sort of inspires some new ideas out of me. So it’s, we’re trying to turn it into a tool that can go further into the process and help people with more of the actual line work, the different layers, the house plan, like roof plans and electrical and so forth. But right now it’s really focused on that initial concept design. If you find a design you like, you can move walls and doors and windows around you can then if you like where it’s at you can render it out using different materials that we have in our material library And it’ll present to you a rendered house with those materials plus the floor plan that’ll have furniture fixtures and You know at the end of all this process where you finally are like cool I’m ready to like actually move forwards and like start designing my full plan set or like iterate on the design a little bit more in detail, can export it to CAD. gives you a full CAD file. You can use a DXF file, IFC files, or just a PDF if you want to present that to a builder or designer that you’re working with as a homebuyer.
Rob Stott: No, that’s awesome. So without oversimplifying it, I know I say that, but I’m going to do it anyway. It’s kind of like that mood board for your home. Effectively is what it becomes. I, you can do a lot more with it, right? And there’s more potential for like, once you get to a design you like, and you can adjust things and move things and you really get nitty gritty in here before then going onto that next step. But it’s to your point, like at the outset, it’s that, what am I looking for? What am I trying to design? And really as a consumer, and I’m actually going to follow up and ask about this in a second, but as the user, should say, it’s just helping you generate ideas and brainstorm what this home could look like that you eventually will take off of this platform and maybe see built in real life.
Nick Donahue: Yeah, the hope is that we can remove the need for Revit and these heavy design software that honestly a lot of drafters in single family don’t even use. They’ll use like AutoCAD and SketchUp or some other things that are a bit more bare bones. So we are trying to get rid of the need to have to tweak so much outside another software, but that’s a bit further out.
Rob Stott: So who is the intended user? it the consumer? Do you see designers using this? Architects? Homebuilder? Or all of the above, right?
Nick Donahue: Yeah, it is a general design tool for designing single-family houses. A large percentage, know, half our user base are homeowners, home buyers, consumers, but that’s just naturally going to be the case because there’s so many more of them. About 15 % of our users are builders, about 10, 15 % are like drafter architects, and the remaining like 10 % are developers, agents, interior designers, more adjacent user bases.
And everyone’s got the same problem. It’s just how do I get designs quickly? How do I ID quickly? How do I remove some of the process in my workflow, like drawing walls, doors, windows, the roof, 3D models? So people just want to make things faster. That’s very, very clear. And we’re trying to make that really easy for anyone to come in and start their design process wherever they’re at.
Rob Stott: Do you see now those communities that are using it? I not that it’s not a sort of You know elaborate platform because it is but like do they use it in different ways or they you know trying to leverage the the software in different ways depending on the type of background or or you know the specialization that they have?
Nick Donahue: The ended up intent may be a little different like we’re working with a developer in Nashville as well and for that type of user they’re looking more at it as a way to reduce the timeline for financial reasons like for them if they’re developing a community you know taking out money against that community there’s interest payments there’s payback periods and every extra month is like tens of thousands, hundreds of thousands of dollars for some of these massive communities. And so for someone like that, they’re seeing it as just a mechanism to allow them to design quicker and get through the process quicker.
For a designer, like, someone who’s working with a client. Again, there’s a lot of wasted work in that initial ideation period. So for them, it’s the ability to sort of ideate quicker, get ideas or things that they can actually present to a user. So there are a little bit differences in how like the actual reason why they want to use it, but it comes back to like the same thing, which is like reducing the work necessary to get further in the process. And so we’re trying to tailor it to them to make that super simple.
Rob Stott: That’s awesome. And I know even in just a couple of times we’ve chatted and in just a few months that, you know, the platform has been around, you guys have added a ton of new features expanded in different ways. so I want to give you a chance to what’s, what’s kind of the thing you’re working on now that you’re most excited about that you can talk about without giving away too much secret sauce here of what’s going on, but what has you most excited about kind of the, very near future for Drafted.
Nick Donahue: The current product is really focused on, we can only do one story, single family, hip to roof houses. So we’re trying to expand the design space that people can navigate through. So we’re working currently on multi-story, so the ability to have two floors or a basement. We’re currently working on the ability to edit the roof lines. So if you want to change a hip to a gable on certain faces, or change the slope of the roof in its entirety. And then beyond those core expanding the design space side of things, we’re constantly R &D-ing, testing out new ways that people could interact with it and navigate it. A lot of users have asked for the ability to actually move rooms around. But the complication in that becomes the second you move one room, it’s not that simple. You can’t just move a room and it’s just like, oh, it all works. We’re good. You sort of have to start rethinking the architecture, like the exterior aesthetic, how things flow. And for like an everyday person or even a builder, someone who’s not traditionally used to it, like the second you get into that level of detail and that level of understanding, it’s now impossible to actually use the product. So we’ve been working on implementing bubble diagrams. So the ability to take a bubble of a room, like a circle, and say like relationally, where do I want this room related to other rooms? And then the model will use that as an input to then generate a house that’s sort of tries to abide by those relationships as much as possible. But then it’ll draw the polygons of the doors, windows, how everything sort of flows, the shape of the house, which will make it much, much easier.
Rob Stott: Yeah, and even just hearing you talk about it, like the amount of the, mentioned earlier, but the time savings of thinking about that, like someone would have to go through and do that. And now you’re able to just, you know, I say just, but make a few prompts or, or queries or, know, give it a few inputs and it’s done for you, to an extent with obviously customization that can still happen and things like that. So, but I, I get it, you know, kind of what you’re talking about with…
Nick Donahue: Ha!
Rob Stott: …the time savings and the efficiencies that it provides as a platform. so that’s your near term. Like longterm, what are you thinking about? You know, what excites you about, or what’s your intent for where you kind of see this platform going or, how do you see it expanding and just being more for this community?
Nick Donahue: Yeah, I mean, the end state is the same. Like I want to make changing the physical world as easy as changing your website. And the like, you know, five to 10 years side of that, it’s just like, how do we verticalize in the pre-construction process more? How do we take in more context so that the model can design better? Right now, you know, we only focus on rooms, doors, window walls, which is like very basic. So the models have to interpret or intuit like why is this bathroom shaped the way it is? Why did I put the door where I did when you don’t really know the fact that there’s toilets and showers and vanities, these aspects that inform the reasoning for why it ends up designing the space the way it designs the space. in the short term, we’re trying to add more context to the models, like longer short term. So can we get into fixtures and furniture, roof plans, electrical foundation drawings, these other elements are all things that someone would have to consider when they’re designing that house Continuing to expand that further into lot constraints. So What does the site actually look like? What is the topo? How do I navigate the topo to make an efficient house? What type of foundation should I be using on that type of topo or that amount of topo? So there’s just a lot of these elements that come into play when someone’s trying to plan a house out that is like really complex and usually a fragmentation of different parties. And so we’re hoping to consolidate that at least processing to make it easier to plan your design on any plot of land anywhere in the world hopefully over time get to the point where we can bridge the gap into maybe introducing builders on the product into a marketplace-like experience and allowing them to actually interact with and meet potential consumers that are there designing their house.
Rob Stott: Yeah. No, that’s, I mean, lofty goals. No pun intended on saying loft type of residents, right? Ha ha, home building jokes. The thing that I, this is a way we may potentially just sort of shifting gears a little bit. Part of the conversation where I feel like you’re either going to love hearing about it, or this is like the part that scares you is the model, right? You keep talking about having to train the model and what that looks like. And I know we can’t get into that’s that is your secret sauce. So we can’t get into like specifically what you guys are doing. But how would you explain to someone what it means to train the model, to have it be better? And, you know, obviously this is something that is always involved. It’s going to be always evolving because like anything, the more practice it gets, the more inputs that are it’s receiving from a user.
That’s training it right at the end of the day is it’s learning how it performs based on certain prompts and requests and things like that. But talk about that as best you can again without getting too deep into the weeds with it for you guys specifically. But what does that mean to train the model and kind of how are you guys doing that?
Nick Donahue: It’s definitely the secret sauce. I guess in terms of a high level, like the way that I think about it and the framework I look at it through is when these models are just like pattern machines, they’re just trying to interpret a bunch of data, figure out what the patterns are in that data, and then allow you, because there’s now these patterns stored inside of this model, you can use certain interactions to trigger, you know, certain outcomes, it to like process the pattern into a possibility, which for us is like a house plan. So usually I’m thinking about like what is the best medium, like the best way to interpret those patterns. A lot of companies take more like CAD like approaches, which is numerical representations of those patterns, which to me never really intuitively made sense. I’ve gotten to see some of these image models or image model companies develop and we decided to go down a different path. We’d do an image based approach because the way that I view it is that if someone was learning to play chess, for instance, you could in theory blindfold me and tell me about this concept of a—with a graph structure where there’s A1 and A2 and B1 and B2 and tell me where all the pieces lie and where they can sort of move on this board. And I can try to process that and I can definitely get to a state where I can process that, but the rate at which I would process that is much slower than if you just took off my blindfold and had me look at the board. Then suddenly the amount of throughput, the amount of processing that’s there is so much higher. And so we sort of viewed it in a very similar vein.
And so we just took a very similar, very first principles approach to this problem and decided to leverage image models, at least initially, which allowed us to get the model working way quicker than anything else we’ve seen. And so we’ve really, really focused on just like, what is the format of the data? How do we clean that up? How do we make it as year as possible what the patterns are that we’re trying to get it to understand and then what is the medium in which you can interpret those patterns through and that’s sort of what led to where we’re at.
Rob Stott: I appreciate you doing that. And I, um, I’m glad you didn’t get offended by me asking, cause I know you easily could. It’s like, Hey, why are you trying to get me to tell my secrets? I, I, did it in the way that I wanted you to, in that you kind of explained, you know, your process for why you went the route you did. Um, the other way, and maybe this would help too, if you can, again, sort of at a generic high level, um, you know, the difference between the, you guys building your own model versus like, someone being like, why don’t I just take, you know, the same thing that I’m doing here? I’ll go ask chat GPT and see what kind of thing they give me. Like talk about the difference of doing that and why your approach is truthfully, mean, just looking at it. And for those that have been listening, you can go play with it, play around with it. is a free, free to use tool. And I’m sure Nick would appreciate it. you know, the, the, the reason why having your own model that you guys are, you know, educating is really just a more impactful and powerful way of approaching this.
Nick Donahue: Yeah, the way that I view it is, it’s similar to like coding tools. Like a lot of these coding agents, these were, you know, models that were intentionally driven to be really good at coding. obviously they leverage a lot of the LLMs at this point. but it took an active effort from these big model companies to like make it really good at this specific type of task. they do tons of evals on it at this point. And from my conversations with some of these people who run these model companies, they don’t really care about home design. They’re not trying to make that easy. They don’t think it’s that big of a market, which is fine with me. Like feel free to just stay where you’re at. We’re good with that. And so they’re just going to keep scaling the models up, scaling data, scaling compute, and hopefully it’ll get generally better at a lot of things, but slower and slower at different things unless they focus on it. So like we’ve just taken a way more focused approach where it’s like, okay, that gives us a gap that gives us time to make this model really good at this one specific type of task or general task. And then the same goes for the data side of it, which is like these model companies, because they’re not focused on home design, don’t care about structural plan sets and roof plans and electrical drawings and these other elements that provide a lot of context for the model to be intelligent when it’s designing. And so the more we can start to leverage that, our ability to get access to good data, and teach the model things that these models will just never learn with open data that you find online because it just doesn’t exist then like we’ll get further and further away from where what their capabilities are and Generally, they’ll probably be able to do some things But not really deeply understand home design.
Rob Stott: Yeah, I thinking of it in terms of employees, right? Because I think this is maybe a way that someone not familiar with the technology or even AI in general might be able to understand. You know, have your employee, that one employee that’s good at doing a lot of things, but then you have your specialist, right? So, or, you know, if you want to go the athlete approach, it’s like you have your kid growing up that did all the sports and they were like decent at all of them and they got through and maybe they were, you know, a varsity baseball and basketball player in high school, but then you have that, that specialized swimmer that ends up making that run. You got your Michael Phelps that goes on and wins 19 gold medals, uh, that was specialized in that thing that he did, uh, is kind of how, is that a fair way of explaining the difference between like that LLM versus, you know, you guys having this very specifically trained model?
Nick Donahue: Yeah, exactly. The other side of it too is the user interaction side. So like the ability to take in different inputs. Right now, you know, these models are text-based. You can introduce images, but their ability to take, you know, certain structured data forms and turn that into an output is much harder. And that’s something that we have to like directly teach our models to be able to take in as an input. as we’ve been exploring that side of it more, just seeing that there’s a lot of opportunities to diverge from these large players which would again take them caring about this specific space to solve this specific input type if they were going to prioritize themselves and it’s not saying that they can’t it’s just that’s not currently their focus so it’s just the reality of the situation which provides an opportunity for us.
Rob Stott: I feel like it’s something I have to ask just because it’s the question that always gets asked, the data question. Do you ever, do you have people that come to you and they’re like, well we’re feeding this very specific model. We’re concerned. How are you using our data and what, how, how are you addressing that? Or are you having to address that?
Nick Donahue: Not yet. Well, probably just because we’re not big enough for that to be a noticeable problem for people. but I think with a lot of the court cases that have been going on, the precedent is sort of getting set for what that looks like. And generally it seems like what is being agreed upon is as long as the data is not being recreated and the output is user driven, like it’s someone sort of guiding with their own creative process through those patterns, but patterns, not, you know, a recreation of something, then it is truly novel or new in a way that is not, you know, copyright infringement. That’s been our general stance. And from what we’ve seen where it seems like the world’s going for the copyright side of things.
As it relates to the data itself, ironically, people have been very open to sharing data. Some of the users actually have given us their data sets for free. And with the core of it being like, how can they help us make this better for their own use? And again, as long as it’s not actually recreating the data, then we’re not like, or reselling the data, then they feel that it’s okay. That’s been extremely helpful for helping us push the model along faster and the training along faster.
Rob Stott: No, that’s, that’s cool to hear. I think that is a Testament to kind of what we talked about a little earlier in that the people that understand it, they get the importance of it and making the bet, making the platform the best possible. so, you know, their willingness to do that, I think speaks to their trust in you guys, in addition to the, the trust in what you’re building. so no, that’s, that’s really cool.
Lots to, yeah, I feel like we could go a million different ways. And I know we haven’t started publicizing it yet, but we’re going to be picking your brain in a slightly different way at a round table later this year. but you know, we could go on forever and I want to be cognizant of your time, but, but Nick, this was a really cool conversation, man. You guys have a, an awesome platform that, you know, again, is really just in its infancy. So to see it sort of evolve as it has in just a couple of months since the builder show to conversations we’ve had to you talking about it here today. appreciate your openness and, and, know, willingness to jump on and share and, you know, look forward to future conversations for sure, man. Cause this is, like, like everything AI it’s constantly evolving. So I feel like we’ll have something different to talk about in just a few weeks.
Nick Donahue: Yeah, I really appreciate it. Thanks for having me on. Awesome.
Rob Stott: Anytime.
