Welcome everyone to the OG Pod. Today I'm very pleased to be joined by my friend Eli Finkelshteyn, who is the CEO and founder of Constructor IO in AI First Startup that's now over 100 employees and worldwide. Eli and I know each other from back in the day. We've known each other for about 10 years now. We met in Silicon Valley working at the same startup backplane. How's it going, men?
Good. Thank you so much for having me. It's wonderful to see you. Oh, it's great to see you. It's been too long. I'm really interested in Constructor and what you guys are building because I remember when you started out, I kind of had this initial thought of like it's an auto-complete. I mean, I guess that's a company. You know, no offense or whatever. But obviously you built, like you built so much value.
I remember like when you got listed with MPM and like you had these other kind of breakthrough moments throughout the story of your company. But yeah, tell us about Constructor. How you had the vision that something as seemingly simple as search could be so impactful and how you've scaled your business to be over 100 employees in worldwide. Such a good story. Yeah, I'll give you this short version because I don't want to bore your listeners. And if
there's anything you think is interesting, it can give you a longer version. But before you and I met, I was working on Search at another e-commerce company. And I was after backplane, which was where you and I hung out. I was doing consulting on it a little bit afterwards with other e-commerce companies. Mostly data science stuff work, right? Yeah. Yeah, data science and specifically, an interest area for me was always product discovery within data science. That's like Search,
browse, recommendations, basically. You see a user, how do you decide which products to show to that user given the inputs that they've given you. And before I started, before we started doing constructor, when most people think about Search, they'll usually think about this word relevance, like how relevant the results look. So you're looking for a pair of pants or something like that, like are you actually getting returned pants? Are you getting returned shoes or something
irrelevant like that? And the thing to me was when you're looking at that in the context of e-commerce, like relevance isn't really the main thing. The main thing is, are you showing people something that they would be interested in buying? Like if I go on to like a clothing website and I search for pants and like you show me a bunch of like women's pants or like kids pants, like they're technically relevant, I search for pants, you're giving me pants. But it's not something that I would
be interested in buying. A much better thing for me would be like you show me things that I might actually purchase. And even if they're not called pants, if they're called jeans or something like that instead, we can have our argument about which one of those is more relevant. But at the end of the day, like I as a shopper care about finding things that I want to buy. And if I was a retailer, I'd care about giving things that a shopper might be willing to buy to that shopper.
So it was solving a problem that you knew existed within many companies was just like, you know, even within backplane, we wanted to surface all sorts of different types of content to customers around a fan based like Lady Gaga. So you were already thinking about these type of problems and differentiating between is it like relevance and intention or relevance and like, I guess like what the company is trying to sell. Like what is that?
The mention that we that we use for it is attractiveness, which is this term that was coined by engineers at Google and Cornell. And like basically what it means is you're taking into account everything that you know about a customer's click stream. So like what are the things are they clicking on? What are they adding to cart? What are they scrolling right past? Like what are they telling you about themselves through their activity on the site? And then based off of that,
you're adapting the site more and more to them. So like the way that we want to think about it is your experience on a site powered by constructors should be very similar to your experience on like a Netflix or a Spotify. Where you are personalized. Exactly. Very personalized, but you actively wanted to be personalizing to you because it's a better experience for you. It's not like this word of personalization that you might have on like Google or something like that. We're like,
oh that's creepy. Remember that thing that I looked at on some other website way back when it's like, no, I'm coming to like American. It's helping users fill their intention with the results that they would actually be looking for based off of all of these different signals. So what are the signals that you might get besides just like looking at this page or like what are kind of some of the more abstract signals that go into how you rank attractiveness?
Well, one of the big things that I think is not done often enough online right now, but at least my take on it is I think it's going to be done more and more is what we call zero party data. So it's kind of a weird term, but basically the way that a lot of folks within the space will think about it is first party data will be like your click stream. So like what are you clicking on? What do you got any card that stuff we were just talking about? Zero party data is
why don't I just ask you a question? Like as you're coming to a website, like right like, you know, are you looking for like gluten free stuff? Like you can just answer that yes or no and then based on your answer you expect the website to adapt to it. If you're coming to a clothing website, like what's your pants size? Do you like flashy colors or do you like stuff that's a little bit more seduced? Like are you looking just just by the best pair of pants regardless
of the cost or are you really trying to save money? Like you know what I mean? So that's going to be a really clear strong signal from signaling the user's intention and then of course the the prompts, the questions that are asked, those are determined by the company kind of like whatever they're marketing, whatever they're product items. Yeah, I mean I think it kind of like hits this underserved use case right now where like we're used to searching for things or browsing
for things online but sometimes that's not really like going to scratch your itch. So like as an example, if I'm a person that's kind of hitting analysis paralysis, like I've got two kids, maybe I want to buy them a birthday part for present. I'm not sure what the right present is for to buy for them. It's not difficult for me to find toys on a website like I can search for it, I can browse to it. My problem isn't finding toys, my problem is deciding which one is the right toy.
If I was going into a store, I could go and find a store associate and hopefully know what they're doing and I would like ask them questions but online you can have an even better experience than that. Like you can have the very best merchandisers within that store create a series of questions to ask the person that very basically are like, okay, you know, does your kid like Batman or Superman? Do they like building things? Maybe then it's Legos or something like that. Each one of those
signals gets fed into some kind of a machine learning algorithm. Like your company is very AI first, right? Yeah, exactly. So everything that we do is it's all probabilistic. It's all based on AI and very much for us it's like about using statistics, using probability to give people the thing that they're most likely to want to buy. Yeah, yeah. So talk to us a little bit about the story of constructor. Like how did you get fundraising? How did you hire your first employee? Those kind of
things. Well, for fundraising, I told them that I knew you and then they just offered me millions and millions of dollars. Yes, yes. That's how it works. Yeah. As a thank you, I wanted to come on the podcast. Well, I appreciate that. My pleasure. We're even now. No, the fundraising was a slog. Like you were mentioning in the early days, the very first thing that we built because we knew that building a search engine from scratch would take many years,
we built auto-suggest and that's all that we were selling in the early days. And I think a lot of the investors that we spoke to, they had the same reaction that you did at the beginning of this podcast, which was like that. Yeah, exactly. That's like a nice feature. It's not a company. It just shows that founders have to have this vision and they have to approach it from like the MVP is we all know whatever the minimal viable product for people that don't know the acronym.
But you have this vision of how you can start with something and then elaborate it over time. And like only you see that and nobody else really gets that, you know. Well, you're trying your best to convince these people that like you will be able to build it and they kind of have to make up their mind about whether this like random person that's talking to them that they've never met before in their entire lives is going to be able to build this thing
that like they promise they're going to build. Yeah, I think something like 95% of the time that's not true. Right. They've got a hard job. VCs totally. Yeah. Yeah, that is a hard job. So how did you convince like your initial seed round or like how did your initial funding get happen? So we got lucky that our biggest investor in our seed round was a person that my co-founder and I had worked with previously at a company called Shutterstock.
So he was the former president of Shutterstock and he had had experience working with especially my co-founder more than me and he was willing to take a shot on us which is very kind of him. And how did you spend your initial money? Were you hiring people? Were you like buying servers? Were you what did you do with like the yeah to grow? We took the money and we built a hot tub on the roof and we thought that that would be really cool to know I'm just going to. Hey, my work you got
to track town somehow. Exactly. Yeah. Now there's like it's like a meme or something like that. Like a lot of people they'll raise money and they'll go and spend it on something stupid immediately to like celebrate it. It's like, Nuna, you got to use that money to build a business. Like you got to learn that lesson hard way. Some people. Yeah. But no, we used it to go and hire people. We used some of it for servers as well but we were lucky to get into some of the start of programs
that the cloud companies have. So we got some credits that way so it was primarily for hiring. So you have your initial idea, your initial product. How has your product line evolved over the years from auto complete into it sounds like it's a full blown search. And yeah, like what products do you offer to people? Yeah. So it's everything that we think of in the category of product discovery. So anywhere that you can show a product on your retail website, our job is making sure the product
that the shopper sees is personalized to them. It's attractive to them. And it is aligned with the business metric that the retailer most cares about. And you're tracking every unit of data. Just knowing your background and data science and and using quantifiable results to be able to show like, hey, here's the impact of using constructor on your business. That's exactly right. The auto
suggests thing like we were talking about before. We laugh about it a little bit now but the reason that it worked and why we were, you know, we didn't die at that time is that a lot of people don't realize but every little part of your product discovery experience can make a pretty big impact. And so for the auto suggests, like we were lucky that we were able to convince a few companies to
test it. And when they tested it, like the primary difference in it doesn't sound very sexy but it was the type of tolerance was much better than anything that you would get anywhere else. And so when they tested it, you would see anywhere from on the low end, it was like a 3% increase in revenue for searching users and on the high end, one company even saw a 15% increase. Wow, that's a huge jump. All from type of tolerance, right? Which like, it's terrible for sales.
Like nobody hears type of tolerance in their life. You would not really think of that type of tolerance. I've never even really thought of that. But of course, like, is it GEEN or GEEN? Like, you know, those kind of things probably make a huge difference accumulated over, over like the full volume of search. People, it turns out, are terrible at spell. Yeah. Well, we've all been punished by auto complete. I feel I don't even try to spell any more. I'm like, ah, the computer will fix it.
Like, computer will fix it. But like, if you're going on to a retail website where a lot of the time, what they're selling isn't actually an English word. Like, you think about like a wine website, right? Like, you try to spell it. Yeah, like, you just heard it at a bar. You maybe didn't even hear it perfectly. And then you're trying to like, type that into a word. You definitely need some kind of typing
type of fixing. Yeah, I mean, like, even right now, like, you probably ask like a good chunk of the people that are listening to this podcast, like spell Cabernet 7 on it. It doesn't spell the way that like it sounds, right? Like, you kind of just have to know the way that it's spelled. And if you think you know the little one, yeah, there's like a you and the staffer on or what I can't even say it. Yeah, but there's a lot of examples like I could do Cabernet. There you go. All right.
But yeah, the idea is that it actually turns out that exactly to your point over a large amount of data, it actually does make a difference. And when you send half the users to like your old algorithm and half the users to the new one that we invented, we were able to show these revenue lifts. And so these companies were nice enough to keep us around. And then eventually. I have to imagine that's a big part of your company. So constructors eight years in?
About a year. Yeah. Okay. So that is like proven building thing. But yeah. Yeah. Yeah. But that is like a proven durable business. And you have worldwide presence. And I'm sure an amazing client list. And it's like eight years ago, MPM was on the client list. And that's like pretty cool for me. Anyway, um, but I think I think it's got to be so like, uh, it's kind of like when you have a really good product, you don't have to work that hard to sell it. And what's a really good product,
well, one that you can show with data works. And I think that probably makes some durable clients. I'm probably the worst salesperson in the world. Like, yeah, I mean, you're my background's data science. Like you don't want a data scientist as a salesperson. So like the pitch for the longest time, but honestly, it's, it's not that far away from it now is like, please just try it. You'll like it. Yeah. Yeah. Totally. Maybe you could see that your users are having a better
experience. And like you can see that mathematically. Like you don't need to take my word for it. You don't need to ask them like, you can literally see that you're going to be getting more purchases on the side that's using constructors than the side that isn't. But that's like the same thing that makes Tesla such a strong business. They don't have to market themselves. Like the word naturally spreads from people that use it and they just enjoy it. Try it. You'll like it. Yeah,
exactly. Yeah. Like the whole experience comes in automatically. So you, you have this idea based off of your past work of a problem that you can solve. You get your initial funding. This was deep learning. The deep learning breakthrough happened around 2012, right? So you're right at the beginning of AI. How quickly, how, like, how does your company leverage AI and how quickly was that coming into play in terms of the product experience and personalization?
So there, there have been like a bunch of breakthroughs over the course of, I mean, from, from before the company was founded all the way up until like, right? I mean, you guys, the people are really listening to this properly have played around with G chat GPT. Totally. Like that in and of itself is a massive breakthrough. The technology that made that possible, which has a funny name, it's called the Transformers. Yeah. Generative. Transformer, protocol or whatever.
It's like a big truck with like a non-moscow. Yeah. It's the savers of the world. Yeah, exactly. It's one of the meets the eye. Sorry, I'm full of, I've got two kids now. So I have to make dad jokes now. It's like a obligatory. It's a requirement. Yeah. The second they're born, the mantle falls upon you. You don't make dad jokes. We will take your children. All right. All right. I'll start. But so you've been leveraging breakthroughs for years now? Well, we're trying to. I mean, like,
there's a lot of people we're trying to and sure. Right. I hope that we do a decent job for our customers. But one of the ones like that I'm most interested and excited about right now is actually Transformers, which is the same thing that the chat GPT is based on. Basically, it's a way of letting a computer understand much more context within what somebody is typing. And so like, one of the things that we've been convinced of as a company for a while is that the way that people
search right now, we kind of search like cave people, right? Like, we're just like typing in these words, like without any sort of sentence structure around it. Like if you walk into like a store, like you're not going to go up to somebody and be like, blue shirt Nike, like that would be like weird. Right. You're going to explain what you want and like what you're looking for and like, they'll get it based off of that. But when you go onto the website, you're like, blue shirt Nike
and that's for normal. With Transformers, the idea is that instead of just needing to match on those keywords, which is kind of what people are stuck with right now, or they have been stuck within the past, it will let us add much more nuance, much more explanation, much more expressivity to that search. And so you'll both be able to speak to it more like a human and it'll be able to understand you better instead of just like hoping that you get the right keywords.
Can you share anything about how this technology actually works like at any level, probably the highest level? Like what is a transformer? I mean, I've tried to understand a little bit myself. It seems like what they're trying to do is always predict the next word. And you get these huge models that like read the entire internet and everything that's ever been written. And from that, I guess you get like this matrix of connections. And this is the black box part that nobody can
exactly like, you know, cite the source for how it determined the answer. Is it non-deterministic? What do they call that? Like staccatica or something? Just the results of these Transformers. So it's probabilistic. Probably. Right. That would make sense. It's based on an older technology that actually might be
the thing that you're referring to back in 2012. Basically, one of the things that people realized that is both fascinating and painful to me is that you actually get much farther in human and computer understanding of human language using statistics than you do using linguistics. Which like, I love linguistics. So I was like, shit, that sucks. I wanted to do more stuff with linguistics.
Can you explain a little bit more of what that means? Like linguistics is like talking, statistics is math, but like, how does that come together? Yeah. So like, if you're looking at this stuff, maybe like, call it 20 years ago, maybe a little bit less than that, but call it like 20 years ago, plus, it wasn't clear to people whether the way, the best way to get a computer to understand human language was through basically like understanding
how human language works. It's like the rules behind it. And that's linguistics. Makes sense. So like, you can think of it as like similar to like the grammatical rules that you were learning as a kid, right? Like you have to say like whom if it's an object and it's who, if it's a subject and like, you have all of these different rules like that. And if you teach the computer all of those rules, then it'll both be able to generate and understand human language
much better. But what it turned out is that the rules mattered or were less effective than just the straight probabilities learning based off of massive data sets. Is that the case? Yeah, exactly. Yeah, exactly. And people started realizing that first in translation. So like machine translation, like Google did a much better job of that than anybody else had beforehand. And there's was not really based on linguistics. Like they weren't, it wasn't built by
linguists. It was built by statisticians based like machine learning engineers. That's so counterintuitive though. You would think that the way to teach a machine a game would be to teach it the rules of the game. But then isn't that not how like the Google bot passed go where like instead of just learning the rules, it just like learned from a million iterations of the game. Something like that. Yeah, exactly. But the interesting thing is like when you think about it,
actually makes sense because how do you learn a language? That's true. Right. A million iterations. Yeah, like as a kid, you just heard a bunch of people speaking English and so did I. And like now of a sudden you speak English and like you learn the rules after the fact, but the rules aren't the the reason you learn you know how to speak. Yeah, yeah, that's totally true. You can speak before you know the rules. That itself provides a clue for why the probability method works.
Yeah, and like I'm over simplifying all of this stuff and like anybody who's listening to this. Oh, it's a complex already. I think no, no, no, but I mean like anybody who's losing this would find like 30 different holes in everything and what I just said, which would be completely reasonable. But like I'm
I'm trying to keep it interesting at the same time. This stuff is so interesting and I feel like people will appreciate just like the highest, broadest level understanding of how this stuff works. But yeah, so we're working with these language models probably, statistically, statistically, and through that we so how are you incorporating or how do you imagine
that these transformers might impact your business down the road? I just think that similarly to how we when when you're working with chat GPT and it sounds like you've played around with it like I think it's about to be on earth. I talked to it like an hour a day. It's a problem. I talked to it until it tells me that I'm out of requests and then I'm like, okay, that's a good, you know. I mean, it's like playing with magic like it's it is. Dude, totally that's a good way to put it. Yeah.
But yeah, you can speak to it in in full sentences like you don't need to use keywords. You don't need to be like very, very careful about how you phrase something like you can phrase it all sorts
of different ways and for the most part it'll it'll understand you. And when you're comparing that to searching online right now like it's kind of the exact opposite right like you have to be very, very careful about exactly which keywords you use and like you don't want to accidentally hit like a synonym that can like mean some other thing potentially like it's almost like a skill that we've we've learned at this point we're just used to like how to search for something. How to search.
And kind of the big question in my head is is that going to be what our kids do as well? Or like will our kids have something much closer to chat GPT and it's easier to speak to it as a human? I think it has to be the latter. I think it has to be the latter because the advantage of GPT is that you don't necessarily so two things come to mind. One, I do think that talking to these AIs and interfacing with AI will be a skill similar to like how searching is a skill. And there are
people that get better and worse results at Google. They just that's how it is. And so there'll be people that get better and worse results with AI. But it's already so natural that it seems like Google is always a couple of clicks. You're providing the keywords and that's what your business is all about, right? It's about reducing friction, surfacing the thing the customer actually wants and making that tied to what the business is promoting. And so with Google you type in some
keywords or whatever and then it's a couple of clicks and maybe you get what you want. With GPT it's sure it's right with misinformation today, lots of flaws, whatever. The long term you type in your question, your query, not keywords, a query, and you get a solution, you get an answer, you get a result. And that's way different than going and reading a couple of pages of Stack Overflow to find a answer to a question. The big question in my head with this is I think
that folks like me, we get a little bit drunk on the algorithms. Like we're like, you've got this like really, really cool algorithm and like it does things that look like magic and like that's that's the golden holy grail. The thing is in reality, like I think the UI is just as if not more important than the algorithms themselves. And a lot of it is just about convenience.
Like there's this open question in my head at least of is looking for something via a chatbot, via something like chat GPT, is that actually more convenient than looking for it by a keyword? I don't know. I mean, sorry, I'd go on. No, I'm just going to say like if the answer to that is yes, like is it always yes or is it yes
for just some certain subset of queries? Right. I could say it being yes for some subset, but increasingly the more that we can interface with this AI like it's just in the room. Like imagine a passive mic, I know it's creepy, big brother, whatever, we're all getting spider on anyway, what doesn't even matter. Imagine a passive mic is just like you can talk to it like it's somebody there in the room with you. And you could just ask it, it's question in real time. And like
you said, so you're getting all this context somehow from these transformers. I don't quite understand how that works, but the crazy thing is going to be like what if it has context of your last conversation or your last 10 years and it's been personalized to your queries, your searching habits? Like that is going to produce amazing results. And you're going to be like what were we talking about last week? I wanted to go, oh yeah, it was that restaurant. I know those things.
Yeah, type of stuff. The personalization, like I've got no doubt, I think that is just continuously proved year after year that like as long as you're not creepy with it, like as long as you don't pull in external information. And what's creepy do you think? Well, I think that shoppers are giving you as a company their trust when they're going on
there and like you need to respect that trust. And so like if you're doing things that don't really help the shopper, but maybe make you a quick buck, like you should probably stop and think about like maybe short term that might be a good idea, but long term that's a terrible idea. Yeah. So at least to me, it's like always having the shoppers interest first.
Yeah. Yeah. Yeah. Of course, to business, we all recognize it's a business. You want to like you want to sell them something, but it's nice that like they're coming to your store because they want to buy something. They wouldn't be there like if they don't want to buy something. Well, and look at all the most, look at all the most successful businesses in the world. They all just focus on the product and they don't take shortcuts. And you know, Amazon, how many decades
was it decades before they made money? Like it was a long time and they started selling books, you know, and now there's the meme. I saw whatever the book I want. It's pretty good. I don't know if you've seen that. Rip Jeff Bezos and like a some kind of v-neck or some kind of a best. Anyway, it's funny. But yeah, like Tesla, Tesla is a product that sells itself Amazon. Same thing. Like the best thing I think that people can do business-wise is never take a shortcut,
front load everything. And do your best to just build the thing that works the best for everyone. Yeah. I mean, like respect the people that are your shoppers, respect the people that are your customers. Don't forsake their trust. Like you have a business. I think Jeff Bezos is like one of the first people that like really parped on that. And like it's kind of funny. Like even right now, like there are plenty of things that you could say Amazon does poorly. And like I would wonder
percent. I mean, I would agree with you on a lot of them. But they still respect the customer quite a bit. Like they're still very, very customer first. And it's absolutely one of the important things when you're building a business. You respect the people who put the trust in you. Well, everyone in tech knows now the term customer experience. And as far as I can tell,
we got that from Bezos. Is that right? I didn't know that. I think so. Yeah. I mean, I don't know where he got it, but where I traced it was from people that used to work for him. And that he would always like, you know, kind of like harp on the customer experience. We always have to think about the customer experience. And now I think about that with everything like with this podcast was the viewer experience with your, you're talking about the shopper experience.
But that's what people resonate with. The experience of something. That's what brings people back time and time again is if they actually enjoy the experience of it. Like that. Yeah. So let's hop back to a constructor. You got some funding. You've got some employees like where, what are some of the lessons? Like, what can you share with somebody who, let's say, they're a potential entrepreneur. They're in the same position you were 10 years ago, 2012. They have an idea. Maybe even they have
some cash or whatever. Like, what should they be thinking about? What are the pitfalls? What have you learned? Well, we actually were at a company together. And a lot of the pitfalls that I learned were from that company. Yeah, back plan exists. Let's just name it. Yeah. Yeah. No, I mean, it's fine that company doesn't, doesn't exist anymore. But I don't know if you've like mentioned it under your show before, but that was a pretty cool experience. I have recently actually. But yeah.
Oh, there you go. Well, we got to work with some cool people and we got to work on some cool stuff. But at the same time, like, there are things that we could have done much better. Like, I would say like getting people aligned on a single plan. Like that, we could have done that much better at back plan than we did years earlier. Years earlier. Having a very clear vision of how you get to success, I think we could have done that much better and like, align people behind
that success. Having an idea of what not to work on. Like, that was one of the biggest things that I learned there. Like, there are plenty of things that you can work on that are going to drive marginal value. But like, especially as a startup, like marginal value is not what you're looking for. Like, you're looking for like big heaps of value that you need to figure out real early before you run out of money. Right. There are things that could be great once you're, you know,
six, seven, eight years in. But like at the beginning, like, you go for something that gets you like a 1% increase in revenue or something like that. Like, if your revenue is like $5, like, that, that time we do in that, don't do that. Yeah. Yeah. Over optimizing before you have like product market fit, for example. Yeah, exactly. It's, it's exactly a good, a good example of it. So those. So was there ever time in the history of your company where you were just like,
this is hard. Like yesterday, I don't know. Okay. Yeah. It's hard to build a company. Every day and including today that I've been doing this company, that's a great answer. I love that answer. No, it's, it's, it's really, it's, it's, it's what makes sense. It makes sense that it would be that way because, you know, like these things are difficult to do. And if it was easy, everyone would do it. Yeah. Exactly. So what keeps you motivated, though, like in that landscape of
difficulty, like what keeps you being like, okay, why didn't I just sell this tomorrow? I'm sure you've had offers. I'm sure there's been like opportunities. What keeps you motivated? I don't know. What, what will you give me for Caleb? I, I gotta see some stats. 50 million, 100 million. How much you want? I'll give you whatever you want. You tell me. Love it. I want the original's podcast. No, I like the originals. I want to keep that for me. All right. All right.
But no deal. I'm sorry. Damn. No, what, what keeps me motivated is at least to me, like this is one of if not the most interesting problems that I could be working on. Like I'm, I'm a language
geek at heart. And I really, really like machine learning. And this is one of the few places where like if you're dealing with search with product discovery, like so much of it touches upon human language, so much of it touches upon the machine learning aspects of it, so much of it touches upon really understanding people like the psychology aspect of it, I think is really interesting. And so much of it touches upon just figuring out what does success actually look like?
So what does success actually look like for an individual person, like an individual shopper coming to the website? And I think that a lot of when they're looking at this problem, like they don't think about those things as much as they should. And I get a real kick out of thinking about that stuff. And I think that I'm always learning. And that's what keeps me motivated. Well, I think you have to be an empathic person to even have those perspectives and to be thinking
like from the perspective of the customer. When I am shopping for something like what is going to create the best experience for me? Now, were you bilingual, right? What's your, like, some kind of family history, if you don't mind me asking? I'm an Ukrainian. Ukrainian. Yep. Yeah. I'm a little. And has like, did any of that kind of like influence your path into this world of tech or data science or rather linguistics in AI? Or was it just other interests? No, I think so.
I got really into like a bunch of language classes when I was little and I kind of just stuck with it. Like it's a long story, but for different reasons. I grew up religious. I was like learning Hebrew and Aramaic when I was younger. And just like comparing like how those different languages work to like Russian, comparing that to English. Like that was just like a really, really interesting thing for me. And then, do you ever see the movie arrival? No. Oh, dude,
it's a great movie. I think you'd like it. Well, I think everyone would like it. But it's like this sci-fi movie where aliens come, they land on the earth, they got all this future tech. But the point of the movie is that when this translator learned to speak this alien language, it rewired her mind. And she was able to see basically through time, she got like this weird
power where she was able to like visualize through time back and forward. And I thought that was a really interesting idea that I think that other languages do change your mind, especially computer programming. That'll make you like more logical, maybe more robotic even. But it'll help you be more efficient and more, I don't know, rational. But I've heard many people say this expression that like there's no way there's no word to express in English what I want to say in
Russian. Or yeah, yeah, that kind of stuff. That happens all day. One of them, I think that's fascinating. I think that I think at least to me that's even more fascinating. Well, not more, but like also fascinating is that with some languages you can create meaning just through the like the structure or the more fallage of a language that doesn't exist anywhere else. So not
anywhere else, but like in other languages. Like as an example, if you ever read like Latin poetry, because Latin, the order of words doesn't matter nearly so much as English, what it lets you do is it lets you create extra like you can think of it as like poetic meaning through the order of the words. So like you could say something like there's this Latin poem, I think it's an op-it from remembering right, but he's describing Poseidon like the god of water, what I'm blanking on it.
He says where he sits between two large whales and like the word two is on one side of the word Poseidon and the word whales is on the other. So like he's literally like the word Poseidon is literally sitting between the words for two and whales. Right, which is like an extra layer of meaning that like or poetic meaning at least that you wouldn't be able to do in English at all.
Like it's not even like a thing. That is so cool that languages have this ability not only to have unique things within words that don't exist in other languages, but the actual structure
of the language can inspire different ideas or thoughts. Yeah, and I think it's almost like to me like you know we live like looking and thinking in three dimensions and like if somebody just like all of a sudden opened your eyes up to like a fourth of fifth, a sixth dimension and you're like holy shit like there's all this stuff here that like I didn't see before but like that's so cool. Well there kind of has to be like in science extra stuff that we can't see because we know that
dogs have such better smells than us and other animals can do like echolocation. There's a wide variety of senses and we're enhancing our sense of James Webb Space Telescope. Now we can see an infrared you know like we're enhancing the human senses but it does make you wonder like how people are going to evolve in the world where we have AI and infinite compute and like what are people going to do with all this stuff? I don't even know. I don't know probably cat pictures.
It's going to be memes. It's seriously going to be memes but like some people will be productive. So what do you see for the future of constructor? Like how what are you excited about like in a couple of years from now? Yeah, I just think that the way that we discover products online
right now is so much worse than than what it could be like every single part of it. And like one thing that we haven't even scratched the surface on as an example is when when you go onto a typical website the user interface looks exactly the same like you search for something you get rose and columns of products back right? Totally. And you don't you don't get those back because that's the best user experience. Like it's not that somebody sat down and they were like you know what?
Like grids like that's going to be fantastic. It's you get that stuff back because it's the easiest way to show something coming out of a database. Totally. Like it was very much designed by engineers not like people that had the customer in mind first. And can we show something that's that's better than that? Like when you look compared to what's available in brick and mortar?
No, I walk into a store like stuff is not there in like these like grids like they think very carefully about like what products they put on the mannequin and like which ones they put together. What products they have gripped off the shelves in a certain way? What's in the like cubbies in the back versus what's hanging on the hangers in the front? Like so is that a good way to understand
constructor? Like we're going to make your store front epic. Like whatever people are coming in integrating with your products, you know that like you're not walking into like you know some what's what's a dumb like electronic store? I don't know, fries. Let's pick on fries. You're walking into an Apple store like this is an Apple store. Is it like that's the kind of difference the product can make? As much as possible like we want to make the shopping experience enjoyable.
We want to. So what other UIs have you been thinking of possibly playing with? Because it does seem like yeah if I'm getting rose from a database that's hung in a surface it you know like I can't have been really think of how else I would display it. I mean I think it's still early days but I think that like showing certain products let's say with like I'm just coming up with like simple ideas but like maybe you show something with like a long long form description and like you show
like a much bigger more beautiful image of it and like you can't do that for everything because it takes up so much space but if you're very confident that one product is going to appeal to somebody like your top line yeah exactly like so kind of like how like Google search does these inline results almost where like if you search for a football game you're going to get the score right there with the logos and the time and they have like all of these yeah kind of like card UIs okay
that's a cool way to think of it. It's a good example of like this sort of innovation that you can do in the UI but like another example is like does it have to be an image like could it be a video like could it could it be something that like shows the thing being used it could is that a better UI or a worse UI I don't know but like we should try it out we should figure it out.
That is a really good approach to things is like I mean you have to have like the scale to be able to run those experiments so like when you're first building a company you don't want to start going down that optimization road now that you have all of the scale you can start to tackle those
kind of questions. I just think that the experience of shopping online can be so much better than what it is right now and like somebody's going to figure that out like somebody's going to figure out how to make it that much better and if we can have at least a small part in that like I think
that's bad-ass. Well you're so well poised so like you have like these transformers does that mean that in the future I'll just go in and be like hey I like the Nike brand and the bangles I'm a fan of their sport team whatever show me what you got and it's going to like take in all the inputs of everything else that I've ever looked at it blah blah blah and show me like
boom you want these shoes right here. I mean it's the idea is you want to marry the technology with like the user experience so there are open questions about like whether you want to use like everything the person has ever done but because you wanted to be scoped and contextual to exactly and you don't want to be creepy and all that stuff like we want to make sure that we
respect the users the shoppers trust. Yeah yeah I mean that's always going to be a problem with the AI company is like you're just gathering so much data and people I think have this weariness of how their data is used I don't know my philosophy is like cat is so out of the bag
in that regard there's no we should all of course respect everyone it's privacy as much as we're going to but like your data is out there I don't know what do you think about like data and privacy like it seems like the apple phone the iPhone can just be hacked by anyone like if you got like 25 grand you can get like a zero-day exploit and like take over someone's life you know have you
heard about that at all? I've thought about it I think it's terrifying. Yeah like why if I'm if I'm a hacker and I find an exploit why am I going to go give it to a bug bounty for five grand apple why don't I sell it for 50 grand or 500 grand on the black market and have it never
be discovered? Yeah so I think you're hitting upon like why companies like constructor need to be need to take security very very seriously like one of the things that for example we'll do is we never take any personate and viable information so even if somebody were to hack into constructor which
like we take pretty strong precautions to make sure it never happens and it never has happened in the past but even if they did like they wouldn't be able to tie the data we have back to let's say Caleb Ogden like they would have like a random identifier if they coupled that with like they also hacked like one of the places that you shopped maybe they could get a little bit closer but like you have to you'd have to do quite a lot and the data that we have at the end of the day isn't
all that interesting anyways. Well I think yeah no that is a good measure too and anonymize everything as much as you can but still like you need all of these data signals create a better end product because the more that you can capture of people's intention while shopping or preferences from even like their browsing behavior the better you're going to be able to
surface results. Sometimes I see an ad where I'm like yeah that's exactly what I want you know like that's exactly what I'm looking for and it's a satisfying experience and that's actually the best customer experience that you could create for somebody is like everyone's making money then because like the ad you want to buy the product you don't feel like you're just like getting marketed to by some whatever company it's personalized it's yeah that's a good experience.
Absolutely I unfortunately have to run. Okay we're at we're at about an hour in dude I freaking love this conversation I hope we can keep it going another time I think everyone needs to I'm like jumping around on to like random esoteric shit dude this is so fun just just give us a quick recap of constructor like what you're proud of with your business and where it's going how
people can connect with you that kind of stuff you're doing cool sharing. Thank you very much I really appreciate that constructor is a product discovery company geared towards e-commerce we work with businesses like Sephora, Benobos, American Eagle most of the large retailers that
you might shop at already and what we do is try to make sure that the shopping experience on those websites is as shopper friendly as possible for you it makes sure that the products that you see on their personalized to you they're attractive to you and there's things that you ideally want to buy.
That's so cool man well it's constructor IO everyone go and integrate with their product use their go apply for their company and I just really appreciate the discussion man I feel like we could keep talking for a long time but I want to respect your time I really appreciate you having me thank you very much thanks thank you my friend great to see you great to see you see you guys