May 31, 2020

Can AI Prevent an Over-Supply of Clothing? – with Nick Clayton

Show Notes

Humans produce 13 Million Tons of Textile waste annually!

Today's guest aims to do something about it, by reimagining the creative process of creating clothing.
Nick Clayton is the CTO & Co-Founder of Savitude, a fashion tech company that aims to use its cutting edge Artificial Intelligence platform to bring your excess clothing inventory to zero.

Savitude was created by industry stalwarts such as Camilla Olson & Jungah Lee, who with the technological backing of Nick Clayton realized they could create a solution to fix the fashion industry problem of designing clothes for one body shape.

Currently, there is a huge problem linking fit, body shape, returns and excess inventory.

Savitude solves this by changing the way clothes are designed and sold from the ground up. Their curation technology solution enables fashion designers deliver the latest styles for all body shapes, with a quicker time to market, resulting in lesser returns and reducing inventory.
Taking ideas and parsing data to automate the sketching process thereby allowing the designers to quickly iterating the design process and get to market faster

Additionally, Savitude's intelligent software products augments the creative ability of a fashion designer, & refashions clothes with unique design details to resonate with a brand’s customer base.

Check out the Things Have Changed Newsletter for the latest on the economy, tech and the the pioneers moving the needle!


Nick Clayton [00:00:00] When designers are taught to design, they're taught to design for a single body shape, they are taught on essentially the mannequin that they have in their design studio, and most of those mannequins look the same and they don't really factor in the real diversity of body shape that real people have. So our goal has always been to address that issue. 

Adrian Grobelny [00:00:24] We create 13 million tons of textile waste each year. Ninety five percent of it, which can be reused or recycled. With all of the tech advances we see going on today, the fashion designers are still using single sized mannequins to design clothing. Today, we speak with Nick Clayton, CTO and co-founder of Savitude, an A.I. based fashion company that is aiming to bring excess clothing inventory to zero while changing the way designers create clothing altogether on THC. We're going to talk about Savitz origins, the tech behind A.I. and fashion, and how Nick got a cease and desist order in his early engineering days. 

Shikher Bhandary [00:01:41] Welcome to things have changed, I guess. We've had a few episodes now. We've had a guy with self-driving trucks. We've had a guy with medical device and neuroscience, and we've had ivied finance and big data. Yeah, seems to be like the buzzword of the century. It has found applications in so many fields from like cybersecurity supply chain chat bots. And I remember this very clearly when I was graduating from, I guess, grad school, the commencement speaker was this a kingpin, if you can call him? And he was like, oh, the thing that you can do to save your careers, literally, it was such such a depressing commencement. But I was like, all your jobs are going to be rendered useless because A.I. is coming. And that's great. Yeah. The only thing he said was, OK, you got to do some creative stuff. Creative work is always going to stand the test of time when A.I. and Tech comes into this. And today we have Nick Clayton, CTO and co-founder of Savitude and a high powered fashion tech company. So that's basically using A.I. in art and fashion, which all this while I thought was creative work. So what's about that? 

Nick Clayton [00:03:15] You can't get away from it. One of the things I'd say about that is we are not looking to replace fashion designers at any point in the process. What we're looking to do is really scale their creativity and allow those individual fashion designers to do the types of work that are more interesting to them while taking some of the sort of analytical side of fashion design and making that process easier for them, that we look at the types of things that a fashion designer might get traditionally from their marketing departments or their merchandizing departments where they're saying, you know, we need this kind of neckline and this kind of silhouette to be on trend for the next season. We're taking that and we're passing that into that sort of creative space, the illustrative space, and sort of automating that sketching process so that the designer can take ideas and really quickly iterate on them. But we're not. We're looking to keep the designer in the loop throughout the whole process. I don't think we can ever have a truly creative process that is 100 percent replaced by, hey, there's some interesting things that you can do to augment creativity. Replacing it entirely is definitely not here yet, at least. 

Shikher Bhandary [00:04:40] OK, so we're good for all of you for now. 

Nick Clayton [00:04:43] Your creative jobs are OK. 

Shikher Bhandary [00:04:46] That's good to know. So, yeah, you touched a bit on that with regards to augmenting the designer, serving as a means through which the designer and I can create more, I guess, original designs. When you guys started this company, was it always just a number of renderings that I can generate with the data points? It could create so many more designs and then the design, actually, the human designer comes in and says that's that's interesting. It developed that a bit more. 

Nick Clayton [00:05:18] Yeah. So we've actually gone through several pivots. The common thread throughout Savitude has been that when designers are taught to design, they're taught to design for a single body shape. They are taught on essentially the mannequin that they have in their design studio. And most of those mannequins look the same and they don't really factor in the real diversity of body shape that real people have. So our goal has always been to address that issue. We started out very early on as a marketplace that would have some logic around shape for recommendations. We quickly realized that a better business model for that was creating that recommendation engine and selling it B2B. But beyond that, we realized that really. Making recommendations on a flawed set of clothes can only take you so far. So one of the things we saw a lot is we would look at a particular company's assortment and we would see there's really nothing that we can recommend for this particular body shape. The best thing that they have is mediocre at best. And so we would, you know, sort of grudgingly recommend those things. But we realized we can take that process upstream and actually feed that data into the design process itself and correct that problem at the source. So that's how we got to the design aspect that we're at now. And it is true that the the generating algorithm can run through trillions of possibilities, that it's just not feasible for a designer to do. 

Shikher Bhandary [00:07:00] You mentioned how designers first work on mannequins, which isn't like a true representation of what I guess the population looks like. Right. Has this point of you not been really employed by other by the big fashion companies? Has it always been this case? 

Nick Clayton [00:07:21] So to an extent it has. There are a number of designers that have made attempts within their fashion designs to be conscious of the complete suite of body shapes. In fact, my co-founder Carmilla, in her label, part of what led to Savitude, is that she did create a collection with the mind body shapes in mind and trying to serve all nine body shapes with that collection, that sort of analytics and that sort of keeping nine different body shapes in mind alongside all of the key proportions that you need to keep in mind as well. It's sort of too much to keep it all in your head at once. And it's very difficult to design a collection, keeping in mind all body shapes. And so people sort of default back to their their training on in women's clothing. It's the hourglass body shape. 

Speaker 5 [00:08:13] So we mentioned nine different body types. How did you guys come up with that number nine? Because I'm sure you can come up with trillions and trillions of different variations of a body. Everyone's unique and their bodies personal in some way. So what was the process to getting to that sweet spot of nine different body shapes? 

Nick Clayton [00:08:36] So we look at nine body shapes and four key proportions, of which each have three options, which are essentially like a sort of band around the average, the band above the average and the band below the average for that proportion. So it would be like wide shoulders, average shoulders, narrow shoulders. And then we look at nine body shapes that is based on 3D capture data. And there is a a set of nine basic types of proportions that encapsulate that 3D capture data based on some some scientific studies. 

Shikher Bhandary [00:09:12] And just to add to that, would these serve as your data inputs as well in the sense you mentioned, I guess, the different body stents? Right. They they might have a different like the neck, the arms, the waist, the body. So what did the main data inputs that go into the model that that provides this rendering? And has that remained the same from your from the start to now, or is it more like now adding some more variations that give you better outputs? 

Nick Clayton [00:09:48] Yeah. So the main data inputs into the system are those mind body shapes and for key proportions on the body shape side, we are also looking at clothing and we have an extensive taxonomy for clothing where we look at the shape of the clothing, the design details things like the shape of the neckline, the type of shoulder, the type of rise in the pants. All of that is incorporated into our taxonomy and where we extract that information from images that we don't have to so that we can essentially normalize all garments into our taxonomy that allows us to look at any given clothing and understand how it is going to interact with those body shape and proportion. And then when we're looking at it from a design perspective, we're looking at what types of design details or combinations of design details are on trend for the trends that the designer is targeting. Or maybe the designer is targeting a particular demographic or a particular area. And we can look at the data for what body shapes look like in that area. What's the distribution of body shapes in, say, the United States as a whole if they're targeting the US market and we can adjust both the number of styles and. A number of garments of that style they should manufacture, 

Jed Tabernero [00:11:13] you mentioned that the program uses vision to kind of figure out what those what those data points are. Where do you get the wealth of data to use in the beginning of the input process 

Nick Clayton [00:11:26] through our co-founders, our domain experts and fashion? One of our co-founders, John Hall, is our fashion architect, and she both sets that initial taxonomy and seeds. The data, one of the things that we can do, because we know that we're looking for clothes. There's a lot of Pryors that provides to the algorithm as the whole. So we can essentially we know, for example, that there's a high probability that clothes and people are correlated positionally in the image, which is to say that people are probably wearing clothes and there's a ton of open source data on identifying people in images so we can leverage things like that to sort of bootstrap the data. And then we built our own data sets of millions of images. 

Shikher Bhandary [00:12:17] Well, and when I am kind of like shopping, right, I'm more like a fairly techy kind of guy. So I like to see how it looks. I like to wear it. And then I'm like, OK, so does having customer consumer input into your design process, is that a data input as well to to kind of improve on a design or. OK, he likes this aspect of it. He or she likes this aspect of form. The more I guess. Let me just focus a bit more on that. 

Nick Clayton [00:12:48] There are kind of two aspects to that. One is the the actual individual consumer preference beyond what their body shape and proportion would suggest look good on them. We can also infer what they like. This is probably blatantly false from a fashion perspective, but anyone can wear red. I don't know that. But certainly the body shape and proportion doesn't necessarily impact preference for red clothing. And so we can take those those attributes and take the feedback from customers and incorporate that into a model specific to a particular customer. But then we can also aggregate that data and use it to improve the model as a whole or use it to improve our understanding of a particular demographic. So one of the things that we can do is we can look at data from social media sources or from event sources that are relevant to millennials or GenZE or some some particular demographic. And we can aggregate what sorts of commonalities we're seeing across those inputs and get an understanding of what that trend is. And using some predictive modeling, we can actually look forward for what that trend might be in the future. 

Shikher Bhandary [00:14:04] Is there any limitation to data? 

Nick Clayton [00:14:06] There is. OK, if you have a really intricate what's referred to as computationally complex analysis, you're not going to be able to run really large quantities of data. The other limiter is how you get that data input. Right? So we have around 40 or 50 dimensions that we look at with regards to clothing that all comes from our visual recognition system and that's all built out in our taxonomy. And we have classifiers specific to identifying the neckline and a garment. So you can't fall out of thin air. Right. And that's to come from somewhere and you have to have some way to get it. There are obviously privacy concerns that we don't run into as much. We're very intentional about not dealing with personal privacy data. We don't necessarily care who you are. We don't necessarily care what ethnicity you are. We care about your body shape and proportion and we're able to make recommendations based on that. And that avoids some of the privacy concerns, which I think is a really major issue in AI as a whole is how do you get that sort of big data analysis while still respecting privacy 

Shikher Bhandary [00:15:24] on your website? You have how like a Porsche could influence the design, how certain images could add to a design, because ultimately, like designers, you draw it from inspiration. So this image is inspiring the model to create a better design. Right. I'm looking at how personalization reduces wasted. So just trying to understand that goal of using all this data within the fashion industry. 

Nick Clayton [00:15:56] Yeah. So the the main goal is that the first element to get everyone dressed and get clothes on the market that are the right size for everyone. And what you get from that is you. A, you know, people who are happier, people who aren't uncomfortable in their body. They aren't whatever mold the fashion industry has created and you get if you create your assortment intentionally, you end up with a lot less waste at the end of the day and you end up with less clothing going into landfills. And one of the things that we can actually do with our technology is we can take that clothing that would be going into landfills and we can suggest tweaks to either where the company is selling it or how it is actually designed. So maybe it needs to have the sleeves shortened or lengthened or something to that effect and then it would be on trend for the next season. So we can actually reduce the end goal would be for there to be no clothes going into landfills. At the end of the day, 

Shikher Bhandary [00:17:03] it's like A, B to B and B to C where or I guess catering to the consumers as well as the retailers by helping them create better. 

Nick Clayton [00:17:14] Yeah, we do have a sort of direct to consumer aspect to our company. The primary focus is on helping designers and businesses and we have, you know, aspects of our technology that our consumer facing. There's some really interesting applications of the technology in mass customization where you can actually take the design intelligence and you can expose it to consumers directly and allow them to customize with the design intelligence assisting them. But our our focus right now is on helping designers and retailers better, better serve their customer population. We do also make recommendations directly to customers at the point of sale. So we have a questionnaire. It's six questions. Ninety one percent of people who land on it complete it and it takes them about 40 seconds. And then we're able to make recommendations based on their body shape and proportion from the assortment that is available on the retailers e-commerce site. 

Jed Tabernero [00:18:20] Well, this is game changing, dude. This is definitely going to affect, like the way we're going to see fashion shows and everything is going to be executed going forward. 

Nick Clayton [00:18:29] Yeah, certainly more diverse. So. Yeah, that's right. 

Speaker 5 [00:18:33] So so you mentioned how you guys work with different trends and being able to predict these different trends and know where they're going. And maybe a current clothing that retailer isn't doing well with might be better suited to introduce it to a different market or in a different region or in a different at a different price point, like selling it to Russ or other stores. There's so many different ways to deal with excess inventory. So how do you model your engines to be able to predict fashion trends and and what is it what goes into being able to kind of know what is going to be in next year beyond just kind of seeing what models are wearing on these runway shows or at these big, major events? 

Nick Clayton [00:19:32] Yeah. So to to an extent it is that what models are wearing and runway shows, there's a cycle in trend in fashion. And we can look at social media influencers, we can look at retail sources, we can look at major events, magazines, runways, and we can see how trends are propagating through the market. And so we can infer that because certain events or influencers are experiencing certain trends, that other portions of the market will be experiencing those trends down the line. And then we can also look at elements like seasonality, elements like trajectory, and we can take all of that in combination to to make a prediction of what's going to happen for this particular section of the market in the future. 

Speaker 5 [00:20:28] That sounds like overwhelming, like where do you start to try to figure out what's what's going to be in and like which which which trends do you think would you'd weighed more in on or which ones are kind of just a phase that people go through. So that's that's wild that you guys are working with so much data. 

Nick Clayton [00:20:48] Yeah. You kind of can't start anywhere per say. It's really it is really cyclical. So you really have to look at the whole the cycle holistically before you can make any sort of meaningful insights on it. 

Speaker 5 [00:21:01] And do you see crocs getting back into fashion, 

Nick Clayton [00:21:07] maybe doing analysis for you? 

Shikher Bhandary [00:21:10] That's selling like crazy, doing covid-19. That's what I read somewhere. Yeah, no comfy. 

Nick Clayton [00:21:17] They are company 

Jed Tabernero [00:21:19] to get into fashion a little bit. I mean, the people that that you work with right now are phenomenal people who have had so much experience in the fashion industry. And one mentioning Camilla Olsen, she's been in the fashion industry for a while, but she's also she also has two bands, three. And she also is a serial entrepreneur. She's also like all this other things that relate to Silicon Valley and fashion. I'm just trying to think like the team that you needed to generate or needed to have to complete your bold mission and to be able to do what you are doing today, especially with models. What kind of people did you have and kind of how important was it that Camilla was was in your team leading the strategy towards. 

Nick Clayton [00:22:03] So definitely Camilla has a wonderful, rich background. She's brilliant in so many ways. That combination of the predictive modeling that she's done. She had two successful predictive modeling companies in the biotech space and she ran her own label for five years after getting an MFA in fashion. That sort of marriage of those two worlds is where the inception of the idea for Sabatier came from. And it really is. It's about the really close relationship between our designers and our computer science. So we didn't go off and build a machine learning system in a vacuum and then hand it to the fashion designers to see data into which you see a lot in. You can sort of take a cookie cutter machine learning approach and apply it to sort of anything and it'll work OK, but we built the the design intelligence with. You know, computer science and design working hand in hand and really understanding the domain, and rather than just having that domain expertize built into the data that's feeding into the model, we have that domain expertize built into the construction of the model itself and that that deep domain expertize is really important. 

Shikher Bhandary [00:23:32] I don't know if you noticed, Nick, but all of us literally before the call, we would like to how to dress up. Right. We got to have a good shot. What's the most fashionable thing you own? And we writing and I put on like a beach. Should I have pineapples on my shirt? Same with Adrian and even the same budget. And we were like, yeah, we are going to the beach. Right. This is how we do it. 

Nick Clayton [00:24:00] It's nice to have you guys all look good to me. I am a computer science expertize in computer science and design marriage there. So you looking to me for what 

Jed Tabernero [00:24:14] we definitely didn't use to produce our outfits. 

Shikher Bhandary [00:24:18] But if you know any new trends that would help us stand out next fall, 

Nick Clayton [00:24:24] please let us know how to dress them. 

Jed Tabernero [00:24:27] Yeah, that's actually I think I'm curious about that. You're getting insights into what the industry is going towards and kind of like what what are the latest trends based on the Instagram models and a collection of data points? Right. Is that something of interest you presents as a sort of product as well? The insights that you get from this, like, oh, we know what's fashionable. We're not only going to provide it to these designers, we're going to have it as an open source like newsletter for people who are interested in fashion. 

Nick Clayton [00:24:58] Yeah, so we look at the design intelligence as having sort of multiple modules. Right. So when we talk to companies that are more fast fashion oriented, they are super interested in that trend modeling aspect. And that is in and of itself an independent capsule that we can provide as sort of a trend analysis, a trend letter like you were talking about. And then, of course, when we talk to smaller designers, they tend to care less about the sort of trend analytics and more about things like automating sketching and that sort of rapid idea iteration, because they're much more into that design aspect where the larger companies tend to be a little more data focused. 

Shikher Bhandary [00:25:49] I keep looking at this, Porscha, that that influences it is an image that you threw in into a model which totally, like, jumbled everything up, like was there a limiting factor with, I don't know, the resolution or something that the image had that totally messed up the outputs that you got? 

Nick Clayton [00:26:13] Yeah. So what's really interesting with particularly in images where we're not actually looking at clothing and we're still having the the model make inferences about what types of design details it's seeing there. It's picking up on features in those images that resemble design details in clothing. And it's not necessarily looking for a V-neck T, but it's looking for a, you know, a set of edges or a shape that looks like of inequity. A lot of our visual recognition systems are learning convolutional where on networks which don't necessarily give you an explanation of why they are coming to the decisions that they're coming to. When you look at statistical machine learning methods, you can say, you know, this happened because, you know, X, Y and Z were true, which is not necessarily something we have in the the neural network world. It's a field of some really interesting research, actually. So we do occasionally put in images that just like don't have any strong visual indicators for any sort of clothing. You know, obviously, if you put in a blank white image, for example, you're not going to get strong indications for anything. Yeah. So the thing is something like a Jackson, I was about to give a Jackson Pollock painting as an example of something that might not give anything, but also there might be some really interesting shapes in there that you could actually use as design details. And one of the the really cool things that we can do is take those sections that have been identified as a particular design detail. And we can actually. Those those patterns directly from the inspirational images and apply them to the sketches, so, you know, there might be some lines on a building that are reminiscent of a neckline and we can apply that to make a really cool sort of new, completely novel neckline come out of our system. 

Shikher Bhandary [00:28:31] Yeah, that's that's insane. I'm just wondering what an image of. I guess so. Just a white picture. What happens when we have, like, someone like Jay-Z in a white shirt? What does that do to the model? 

Jed Tabernero [00:28:44] Probably just makes it better. 

Shikher Bhandary [00:28:46] I probably self-destruct, Stewart. 

Jed Tabernero [00:28:48] I think it increases the validity of the fashion in the model. 

Nick Clayton [00:28:55] So if you throw a bunch of people with white shirts at the model, it's going to probably it'll it'll pick up white shirt as a trend that 

Shikher Bhandary [00:29:08] might lead to any one white shirt. 

Jed Tabernero [00:29:10] Everyone I'm just thinking about, like, you have to identify influencers and what's on their feeds for Instagram. 

Nick Clayton [00:29:19] Yes. So we look at, you know, we look at influencers and we look at the types of what they prefer, what their preferences are, and then we can look at the sort of map of of influencers and how influencers are influencing each other and influencing the market as a whole. And that identification of. Essentially, causal correlations between influencers is super interesting and that some some really interesting insights into how trends are propagating through, even just like a single social media platform, let alone the market as a whole. 

Jed Tabernero [00:30:00] I'm curious about which social media platforms you leverage. So Instagram, probably Pinterest, Pinterest, 

Nick Clayton [00:30:09] Instagram, Pinterest, any primarily visual social media? Because we do. We're looking at visual recognition. 

Shikher Bhandary [00:30:17] Imagine if you guys consult with fashion over that creates a design that an Instagram model wears, which ultimately goes into the model that you guys are working on. So it creates like this infinite loop. 

Nick Clayton [00:30:32] Yeah, I don't know. Yeah. The black aspect is really interesting. One thing you can do with the feedback loop is actually consult a retailer can actually consult their customers dynamically and you can have the design. Intelligence is able to create a wide variety of iterations on designs. You can have sort of a core concept for a collection and you can run that by your consumers and see what it is that they are liking from that collection and what it is they're not liking from that collection and dynamically adjust that feedback as you are getting or dynamically adjust what you're asking the customers about as you are asking them. 

Jed Tabernero [00:31:18] Basically, it's interesting because I'm seeing now these companies before kind of what you would see in these big companies websites, something sexy, something sleek, whatever it is, these new companies that are coming out who work with a lot of big data every time, like it's an optimization problem, they're trying to get to zero. They're trying to reduce this. They're trying to work with sustainable. Yeah, it's always something super duper interesting. Like I went on your website and would immediately come. I was a sustainability piece. Sustainability, reduced waste, ethical fashion, circular economy, you know, these things that it's been thrown around. But I don't think I've heard it in too many big news news channels. And one stat that was was on this on the website was that each year clothing generates five billion pounds of landfill waste and 15 million tonnes of carbon emissions annually, equivalent to the amount of trash produced by five million people in a year. Isn't that cool? Yeah. 

Nick Clayton [00:32:17] Yeah, the numbers are huge. It's there's five hundred and fifty billion dollars of waste annually globally that's just burned. Or actually France just recently passed a law saying that you can't burn clothes anymore. So they have to figure out something to do with it. Not. 

Speaker 5 [00:32:39] Trying to pick your knowledge of the fashion industry and your experience working with all these different retailers, I wanted to know what what is the typical lag for a fashion trend or a style to be caught by a retailer or company, go through the design process, go through the manufacturing and then mass producing that piece of clothing. What is the average duration that it takes and which ones? Which retailers are kind of. Doing the best at this and which ones are kind of lagging and having issues with trying to keep up with fast fashion and trends of never ending. 

Nick Clayton [00:33:26] It depends very much on on the type of fashion retailer. So you have the sort of larger, really high volume trying to produce sort of mass market clothes at a low cost and sell them for, you know, kind of small margins. Those tend to have much longer production cycles. So it'll be a year or two years for design to go from the inception stage to actually on the shelves or on the racks. There are a lot of fast fashion companies that take that cycle and reduce it to larger. Sort of fast fashion brands tend to be more on the scale of a few months, six months. There's some, you know, even smaller players working on. Really like custom manufacturing type applications, where the turnaround would be weeks, instead months, but that's still a developing field. 

Speaker 5 [00:34:32] What's what's your go to outfit like? If we were to see Nick walk in across the street. How would we see see you. 

Nick Clayton [00:34:41] Yeah. So I, I am typically wearing, like, mom denim jeans. I don't particularly care for the feel of denim though. That's just a personal preference thing. I think denim is a perfectly good fabric, but not non denim jeans and just a button down if it's warmer or like maybe a button down and a blazer or a coat or a hoodie. 

Jed Tabernero [00:35:11] What's the threshold for influencers just being called an influencer or a certain amount of followers that go into your algorithm? 

Nick Clayton [00:35:17] So we tend to look more at influencers with higher numbers of followers because the connectivity is greater there. In theory, there is no bottom limit to what would have some amount of predictive value. But typically the more followers, if Beyonce does something, matters a lot more attuned to the market trends as a whole than me to on on my Instagram with I think zero followers, probably. 

Jed Tabernero [00:35:50] But if you identify if I 

Nick Clayton [00:35:51] was be an indicator that something's happening for sure. 

Speaker 5 [00:35:58] I'm really curious to know how. How did you end up in an A.I. fashion company going from robotics and even before that, working for a consulting startup, S.W. Consulting previously? Can you walk us through kind of your your progression through your career and how you ended up Savitude? 

Nick Clayton [00:36:21] Yeah, sure. So I have always had an interest in in process automation to a certain extent. So way back in my when I was nine or 10, there was a online MMO called Soundscape. I don't know if any of you guys played that back in the day, but the way that I actually got into coding was creating scripts to mine in it was the first thing that I did. So they were you would click on rocks to to mine them in Ramsgate and you could write a script to look for the right color pixels and click on them. And that was like a baby's first programs. From there. I did a lot of sort of more casual development, was part of the first robotics program in high school, which really strongly impacted my direction in terms of interest. I think a really great program, really great place for kids to learn. And I was the programing captain on my team. We did, among other things, computer vision for robotics specifically. From there I went to college. There is a lot of great and computer vision courses at the University of Michigan, a lot of great AI and computer vision professors at the University of Michigan. I would say John Leonard and Sylvia Rassa are probably the two professors that most influenced me there. And I started tinkering with AI and computer vision back in the summer of summer. After my freshman year, I built a MLP natural language processing application that looked at Craigslist posts and determined whether or not they were a garage sale, found the address and plotted them all on a map so that you could find it. 

Speaker 5 [00:38:28] How did you do that? Was it was it like like an image of just like a driveway? I'm curious, 

Nick Clayton [00:38:35] how did that work? No, I was I was looking at the text data and I was looking for spans. So sections of the text that looked like addresses. And so I would find listings that contained an address. And there were a couple of other there's a type of classifier called a naive Bayesian classifier, which just looks at word frequency that I was using. And so it would say, you know, this is probably a garage sale and here's where it is. So throw it on the map. Wow. And you could look around. I got a job well done, mostly because I titled my app, like Craigslist Macros, thing like that. I think they objected to my use of the word Craigslist. So that was I did projects like that. 

Speaker 5 [00:39:30] That's pretty cool. So were you like flipping these things that you'd find at a garage sale? 

Nick Clayton [00:39:35] I so I got the idea because a friend and I were going garage sale, hunting, looking for parts for a stick welder, which is not a process that I would recommend to anyone. It is not safe. Do not build your own stick welder. But we were we were looking for four parts for that as well as later, some parts for a 3D printer. And it was just a lot of trouble to find garage sales. Yeah. So from there I started working at a startup called Deathcore Wiser, which later changed its name to all Fredi that was working in the computer vision space. A lot of my coworkers were were PhDs in computer science and they were working on a software application to do 3D captures of rooms in the real estate space. So you would spin around with your phone. We stitched all those images together and got a depth map for what the the 3D model of the room would look like from that perspective. And, you know, from there I started doing some a couple more independent projects in the computer vision and machine learning space. And eventually Camila's reached out to me about solitude, and I'm doing that for the last four and a half years now, I guess. All right. 

Shikher Bhandary [00:41:11] I want to focus a bit on the season dismissed because I think we didn't touch on that stress enough. We didn't touch on it enough. Yeah, I got about 

Nick Clayton [00:41:20] ten thousand downloads and then I got a C D, 

Shikher Bhandary [00:41:25] so 

Jed Tabernero [00:41:26] I said, you know, you made it. 

Nick Clayton [00:41:28] I didn't have the guts to not comply with the cease and desist at the time or fight it at all because I was a freshman in college. Yeah, but 

Shikher Bhandary [00:41:38] I'm glad you got it because the number of times so I'm big into sneakers so I'm always trying to get like, you know, the latest drop. And before I can even press the button, it's all sold out. So I'm glad I didn't. I bet all of that is just some programs, just some bots just buying. 

Nick Clayton [00:42:00] Yeah. 

Shikher Bhandary [00:42:00] So I'm glad Craigslist got to you and made sure society is not to gamed by your algorithm, but know regardless. That doesn't say. Does that count as close to being a convict on the things have changed. I guess this is a first is a first. Right. So your secret sauce is ultimately it comes down to having that model and training it. And how do you secure the model itself? 

Nick Clayton [00:42:34] Yeah, so there are two ways that we look at that. One is where where patent protection is available. We patent things. So we have four patent applications in the works for various aspects of our technology. That's. Traditionally, a lot of software is not patentable. So we are very, very careful about the way that we store our models and who we grant access to those models and how those models are run and that sort of keeping them as a trade secret. That is the best you can do when patents aren't available, which is very frequently the case in computer science. 

Shikher Bhandary [00:43:14] So is it possible to, like, platform ize the model, like give it a UI and UX to to make sure that now other people can access the front end and use the A.I. without getting into the trenches? 

Nick Clayton [00:43:34] Yeah, so we put an API in front of our model so that an application program interface so that people who want to use the model don't necessarily need to understand how the model works or even necessarily have the capabilities to to run the model. A lot of smaller. You know, we have a Shopify application, for example, retailers on Shopify frequently have no engineering expertize, so it has to be a fully automated solution for them. They don't need to understand the underlying machine learning to make it work. 

Shikher Bhandary [00:44:07] That's that's quite interesting because in essence, that's that's kind of like a source platform software as a service where people use it. But you guys do the nitty gritty stuff. That's that's incredible. 

Jed Tabernero [00:44:18] Before you started to write, you probably had a specific fashion sense or whatever, the lack thereof or whatever. What happened after did you start using 7C for yourself instead? 

Nick Clayton [00:44:32] I definitely learned a lot from looking at our databases and we've definitely improved since I was never completely devoid of fashion sense, but I was definitely more towards the the unfashionable side for the the earlier years of my life. And it's definitely looking at the the underlying factual data and working so closely with fashion designers, it's almost impossible not to pick up at least a few things. 

Shikher Bhandary [00:45:09] Nice. 

Jed Tabernero [00:45:10] Just imagine if that's your work every day, you know? I mean, yeah, yeah. Something comes out and it's like, oh my gosh. Like Scarf is in Sydney or whatever. You and your wardrobe had a scarf. So you just feel like, you know what, I'm going to wear the school 

Nick Clayton [00:45:21] like I never would have known what like drop or block shoulders are before working with Tavita. 

Shikher Bhandary [00:45:27] I still don't know what that is, but I guess it's important. I don't have to say I'm going to research and trust me tomorrow. I'll be stylish. 

Nick Clayton [00:45:38] Yeah. 

Shikher Bhandary [00:45:39] Do you have some final thoughts to kind of address with regards to how people could reach out to you, how people could know more about the company and the incredible tech that you guys are working on where they could follow you? 

Nick Clayton [00:45:55] Yeah, so the best way to learn more about Savitude would be to visit We have contact forms there or my email is outside of toothcomb and I think at