Rajat Bhageria

Episode Summary

In the first episode of "My Space," host Manav sits down with Rajat Bagheri, the founder and CEO of Chef Robotics. The conversation unfolds Rajat's journey from building his first entrepreneurial projects as a teenager in Cincinnati, Ohio, to pioneering AI-driven robotics in the food industry. Rajat shares his early experiences creating a media platform for young writers, his deep dive into computer vision through assistive technology for the visually impaired, and ultimately, how those experiences shaped his approach to building Chef Robotics.

Listeners get an inside look at Chef Robotics’ mission: automating the labor-intensive, repetitive tasks involved in food assembly and manufacturing, rather than front-of-house food service. Rajat discusses the massive market opportunity created by the labor shortage in food production, the technical complexities of handling diverse food items, and the thoughtful balance of hardware, AI software, and data strategy driving Chef's competitive edge.

The episode also dives deep into Rajat’s perspectives on entrepreneurship, the realities behind the “passion hypothesis,” and the importance of solving real pain points in big markets. The conversation wraps up with advice for aspiring founders, reflections on the role of software in the physical world, and Rajat’s reading recommendations for those interested in innovation and impact.

Transcript

Manav [00:00:00]:
Third eye being an investor and then.

Rajat Bagheri [00:00:02]:
Chef Robotics, I was, like, 16, 17 years old. And there's people in Brazil and Argentina and China who are using this product. A few angel investors were like, why don't you turn this into a company?

Manav [00:00:10]:
What exactly are these robots doing?

Rajat Bagheri [00:00:12]:
So where each robot's kind of doing the work of two people? Robots don't fail. They don't slow down. They just, like, go. Those are the two things that kind of got me excited. It's a big market. It feels like there's a big pain point. And I guess one more thing which.

Manav [00:00:21]:
Is, like, advice would you give to people starting something.

Rajat Bagheri [00:00:24]:
The passion hypothesis says that passion leads to success. You look inwards and you're like, what am I passionate about? But the truth is that, like, foreign.

Manav [00:00:37]:
Hello, everyone. Today we have Rajat Bagheri on my show. He's the founder and CEO of Chef Robotics. Rajat, welcome. I'm so excited to have you on the show. How are you doing today?

Rajat Bagheri [00:00:47]:
I'm good. And thank you, Mana, for having me. I'm excited.

Manav [00:00:49]:
Thank you. I want to ask you many questions about Chef Robotics, but I really want to ask about you first. Can you give us a little intro about yourself, starting thirdeye, then moving on to being an investor, and then starting Chef Robotics? How has been the last decade of your life been?

Rajat Bagheri [00:01:04]:
Yeah, that's a. It's a broad question. That was a big question. Yeah, I guess I went to, like, high school, and a lot of my kind of, like, schooling was in Cincinnati, Ohio, basically during school. I mean, in high school, I was very focused on kind of like, my number one goal is I wanted to go to a good school. So I was like, I worked really hard to get into a good school, and then kind of like, when I got accepted, I was like, okay, fine. Like, I basically last, like, five years. I was super focused on that goal. And I was like, okay, now I'm free. I can, like, do whatever I want. And that's really when, like, entrepreneurship really started. So kind of my senior year of high school, I worked on a few different projects. One was called Cafe Mocha, which is basically this. Like, at the time, medium really didn't exist very much. So there's like, Tumblr, there's these different blogs, like WordPress blogs that people are doing. But I was like, okay, well, I like to write. And there wasn't really a good platform for young writers to publish their writing or their research or their poetry or what have you with the world. So this is really where I Learned how to do software engineering and then actually shipped this thing. And at the time I was like 16, 17 years old in Cincinnati, Ohio, like suburban Ohio. I was like, wow, there's people in Brazil and Argentina and China who are using this product and it's pretty cool. And so this is relatively formative experience for me as a young person.

Manav [00:02:03]:
And this was blog articles, like, not like Twitter, right?

Rajat Bagheri [00:02:06]:
Yeah, no, it was more. It's essentially medium, essentially the way you can think about it, essentially. That was the idea, right? So you could have like profiles and you can kind of publish different stuff and there'd be like a feed based on like who you follow and stuff like that. But really for written content, long form written content, I would say.

Manav [00:02:19]:
Got it.

Rajat Bagheri [00:02:19]:
That was kind of my first kind of real entrepreneurial endeavor. You know, right before college, I really got excited about like a different project which is like education reform. Again, just like, I was like curious and just playing around with stuff. So I wrote like a small book about education reform. This sprung out of like, honestly, like senior year, one of my English teachers, we had this assignment which was give a critique about the school system. So I really got into that project. I was really excited about it and it felt like the school system I went through was very focused on extrinsic motivation. I did the process, the game, and then I was like, okay, well could it have been different and more intrinsically driven and motivated? Anyways, I was just like, I have nothing better to do this summer. I was like, okay, I ended up writing a small book about education reform from a perspective of a 17 year old, basically. Which is of course at the time I didn't know much, but it's like, okay, I just went through this, here's my thoughts, right? And basically how do we get biology and chemistry and physics and math and English and all these different subjects to be a little bit more intrinsically motivated then my goal during college, basically freshman year of college, my goal was to find a co founder for Cafe Mocha, the social networking website for young writers. That was my goal. And so I was doing computer science at the time and I was like, okay, who's the best kind of engineer here? I found some really great people that became very close friends. And interestingly what happened is that we entered this hackathon pen apps and at the time Alexnet had just happened and computer vision, it felt like it was having its stride again with deep learning CNNs. So we were, okay, well why don't we take that idea, we combine it with smart glasses The Google Glass was having its heyday at the time as well. And what if we build like a product for the visually impaired for this hackathon? Just for the hackathon.

Manav [00:03:40]:
How are you helping the visually impaired though, with the glasses?

Rajat Bagheri [00:03:43]:
Yeah, so the idea basically was that, so it's funny, like our, like there's three of us and one of my co founder at the time's grandfathers was visually impaired. So that was kind of like where we knew about the problem, I guess. But basically the idea was that, you know, the visually impaired kind of had this idea of like learned helplessness, right? Like you kind of go through life and you kind of constantly need help. And I think our idea was like, if we would provide them a product that they could put onto their face, like smart glasses, and they could use a verbal signal like, okay, Glass, recognize this or something like this. And then we could take a picture or video stream and then do real time object recognition and detection. Then we could tell the person verbally, hey, you're looking at Ibuprofen versus Advil, or it's a $1 bill or $5 bill, or here's what the menu says at the restaurant. It's Main street versus Market Street.

Manav [00:04:19]:
That's such a good idea.

Rajat Bagheri [00:04:20]:
Yeah, we honestly, it was like a very simple idea. You know, honestly, we didn't know much about computer vision or anything at the time. We were like two weeks into college. Right. But we ended up doing really well at this hackathon and we just hacked our way through this and at the end of it, a few angel investors were like, oh, why don't you turn this into a company? You know, again, like, I was pretty excited about companies at this point. I had done kind of moca. I had done like the book. Like, I'd done a bunch of like, entrepreneurial things. And then I had, you know, basic technical jobs. My co founders were extremely good engineers. I was also an engineer, but like, they're much better than me. I was like, okay, like, why not? What do we have to lose? So we decided to kind of do this. This is what became third eye. You know, a lot of my kind of college experience was third eye, right? And basically learning how to hustle, right? So it's like, how do I convince people to like work here? And like, hiring is really hard. How do I convince customers to buy this thing? What's the business model? What's the pricing model? How do I get press? How do I do marketing? Just all skills I didn't have even basic stuff like Incorporation and stuff like that. I have to figure this out. Anyways. Worked on this for around three and a half years and ultimately became a decision of do we drop out and do this full time or do we kind of sell the company? For various reasons that I just sell the company. And so third day was this really exciting experience for me because again, we were all very young, and yet we had done something that people were using, and it was a pretty cool product. It wasn't like hardware per se, mostly a software product, but it's cool to see people using it. It does have a visceral nature to it.

Manav [00:05:32]:
That's so smart of you to not launched that company after you graduated and during the school you doing that. That was actually like, really smart. And so you must have learned a lot about like image recognition machine learning during that process as well, right?

Rajat Bagheri [00:05:46]:
I. I think exactly. And. And I think that that's exactly right. Which is like, that was my first experience really with AI, right? And I became really excited about this prospect of AI. You know, this was like 2018, 2019 type of timeframe. And. And of course, like, AI had been. Just to be clear, AI has been a thing for like decades, but. But there's always these, like, winters, basically that happened. And, you know, it felt like 2013, 2014 with ImageNet and AlexNet was this kind of like, spark again. And we kind of took advantage of that. So anyways, but for me personally, I was like, okay, like, this thing seems exciting. And it. And it basically was like, okay, like, I had a few things that I was thinking about next. I was like, okay, like, first of all, I want to, like, really learn how to build companies. Like, I don't actually know how to build companies. So that was like one of my goals from basically like a mentor, right? Somebody more senior. And then the second thing was like, okay, what's the next thing I want to do? So I tried to do parallel process both these things. So from. In terms of, like, learning, I convinced. So there's this guy, Slava Rubin, who's the founder of Indiegogo. He was giving this talk at Orton. So I was like, hey, like, you know, this is who I am. I did this thing with Third Eye. He was launching this new, like, equity crowdfunding product for Indigo. He runs indiegogo. So he's launching this new product for equity crowdfunding. And I was like, well, like, I'm. I'm guessing you need help with deal flow. What if I help you build, like, some new vehicle to get deal flow? Using a bunch of scouts all over the country as founders, basically. So I was like, basically create my own, try to convince him to create my own job with them. And like, okay, well in return can I just follow you around and attend the meetings you attend and things like that. So he did say yes. So basically that was a really awesome experience because for a few months I just followed Slala around. I went to all his meetings and he was thinking about acquiring companies and I was helping with the model and attending, helping make slides for even board presentations. A bunch of pretty intense stuff that as a young person I probably shouldn't have access to. But he became a really great mentor. So that was during the day and then the evenings, of course. I was thinking a lot about, okay, what do I do next? And I became really excited about essentially two ideas. One was AI, of course, just continuing AI, and the second was energy. I think it's just like it feels like those two things in various breadths and scopes are the two big things that are going to affect our lives. Of course, just given the third eye experience, I was more excited about AI. I was like, okay, what's the right product and company to build? So then I really took a deep dive into the market, right? Like, who are the customers for this? And I became excited about this idea of like AI in the physical world. And what I mean by this is like at the time like the AI companies that existed were like, people were doing like machine learning for like spam detection, spam filtering and stuff for like Netflix recommending you content. It was mostly in a cloud, all software based. And I was like, okay, well I think there's something exciting about AI in the physical world because like 90% plus of GDP is in the physical world.

Manav [00:08:09]:
Like Tesla's figure, robot, humanoid robots. What else is an example of that?

Rajat Bagheri [00:08:13]:
Well, I think, I think like my thinking was, okay, like what are the foundational industries that represent most of gdp? It's, it's the labor market, which is half of gdp.

Manav [00:08:21]:
Like retail jobs, nursing, hospitals.

Rajat Bagheri [00:08:24]:
Yeah. Transportation, construction, even things like mining. I mean these industries was just so gigantic. I was like, okay, well what's the biggest part of the physical world? Well, it's the labor market. So then I was like, okay, well I want to do something in the labor market which obviously takes the form of robots. Right. AI enabled robots. Right. But again, didn't exactly know what the right industry was. And so this is when I kind of the impetus for the fund happened. So another very close friend of mine, Nandit, while I was doing third eye in startups in college. He was doing a bunch of. He convinced a bunch of really great LA funds, LA based funds, to essentially do the same thing I did with Slava, essentially take him as essentially like, I'll do whatever you need me to do. Which of course was a lot of like, you know, helping with getting deals over the line, but also like fund operations. He did like, whatever it took, basically.

Manav [00:09:05]:
I can tell you enjoyed that experience a lot.

Rajat Bagheri [00:09:08]:
Yes.

Manav [00:09:08]:
Like the investing and looking at deals.

Rajat Bagheri [00:09:10]:
Yes. So we decided to kind of say, okay, well, why don't we do something together? We became close friends, we launched prototype capital and the idea was like, it seems to be the case that there's going to be this big influx of companies who are using AI in the physical world or even IoT or ML. So let's go after them. But to go after them, they don't just exist in the Bay Area or LA or New York. Those companies are like everywhere else. They're in like since Cincinnati and Idaho and Atlanta and all these other cities in the US and around the world. So Silicon Valley VCs are not looking for those companies. So how do we get access to them? And we said, okay, well the way we can get access to them is by finding out who the founders hang out with for fun. And the truth is most founders hang out with other founders for fun. So if I want to get after a really great company in Atlanta for insurance tech, let's say hypothetically, then if I can find that founder's friends and convince them to be a scout with us and I give them carried interest, if they refer us great deals for deals that we invest in, then hopefully we can get some really great under the radar deal flow.

Manav [00:10:05]:
And these are pre. Seed. Seed checks.

Rajat Bagheri [00:10:07]:
Precedeseed checks, yeah.

Manav [00:10:08]:
Which is the highest form of risky.

Rajat Bagheri [00:10:10]:
Yes, it's the highest form of risk for sure. This is kind of what we did. So we had around 70 different kind of founders all over the country and some of these were actually later stage founders, like with CSC founders and stuff too, because they also know early stage founders which are like their friends. And yeah, we basically invested in a bunch of different companies.

Manav [00:10:24]:
You raised a fund and then.

Rajat Bagheri [00:10:26]:
Yeah, we raised a fund. Yeah. Which is not hard, which was just where. Hard in it of itself.

Manav [00:10:30]:
Yeah.

Rajat Bagheri [00:10:30]:
But yeah, so the prototype experience was quite, quite good actually.

Manav [00:10:33]:
And it's still ongoing?

Rajat Bagheri [00:10:34]:
It's still ongoing. And we made, you know, we've returned the fund multiple times over and that's been cool to See, knock on wood, it continues going up. Yes. But from the chef perspective, it also kind of got me in front of some of these potential customers. It got me in front of potential customers in construction and agriculture and food and all these different industries. And of course I ended up focusing on food. I always knew that I wanted to like the whole idea for prototype was like founders investing in other founders. Like I didn't want. Me personally, I really like building. I never wanted to become full time vc, at least in the short term, a medium term, I guess I really want to.

Manav [00:11:05]:
You're too young for that.

Rajat Bagheri [00:11:06]:
Yeah, I really wanted to build stuff and I was like, okay, prototype is going to be a thing, I always do. But like I want the full time thing mostly to be building. So anyways, like I. The food industry was very exciting and the reason it was very exciting is because because of a few kind of macro trends. One macro trend was just the size of the industry. I learned just looking at data I was spending a lot of time, like once I selected there seems to be something in food, I looked at a lot of data about what's the size of the industry. What I learned is that the biggest industry on planet, so in the US is kind of nursing and personal aid. The second one is kind of retail salespeople and the third one is food preparation, food service, food production. My perception was that the first two are not tractable by AI anytime soon. So it felt like number three is actually arguably the biggest market tractable for AI, just given this idea that the proxy for market size of AI is the number of humans who do that job. So I said, okay, well it seems like there's a really big industry. And then I was like, okay, is there a big problem? Because of course in startups you need to be solving a problem. So with that, again, I took an anecdotal approach as well as let's look at the data approach. Anecdotally, I talked to anyone I could talk to, from food truck operators to fast casual operators, to airline catering, to ghost kitchens, manufacturing, anyone and everybody. They all said, basically my number one problem is a big labor shortage. It's like on a given day, I don't know which 80% of my people are going to come to work because of that, I'm leaving revenue on the table. And some of them even went as far as to say, like, you know, this is an existential risk for our business, which is to say, like, if we do not fill these jobs, we don't know if our business in 10 years can survive. And so they were really kind of thinking about like, you know, do we offshore parts of their supply chain? You can't offshore everything, but can you offshore parts of the supply chain? Especially with the manufacturers, the big guys. So it felt like there's a big life from anecdotally, it felt like there's a big pain point. And then I confirmed with Data that the BLS in 23, 2023 reported that the food industry is actually the number one labor shortage in the US more than like retail, more than manufacturing.

Manav [00:12:48]:
And this is combining assembly, food preparation.

Rajat Bagheri [00:12:52]:
It's just like a food industry basically. Okay, so basically those are the two things that kind of got me excited. Which is one more thing which is like it's a big market, feels like there's a big pain point. And then going to all these fast casual chains. At the time we were said about fast casuals, it felt like the status quo AI robots could kind of scoop food and make a chipotle bowl or a sweet green bowl. Like it felt like that's like something I can imagine. So technically of course we hadn't done a ton of homework at the time, but it felt like the puzzle pieces came together and that's why we decided to kind of focus on broad scopes, that industry.

Manav [00:13:18]:
Yeah, I've seen a bunch of companies like I made a video on Hanson Robotics. They did the coffee machine. Then there was Bloom Pizza. They did the pizza. One thing I like about you is like you're targeting the robot that's behind the scenes, like the invisible robot, which is in my opinion a much bigger market comp to the ones that are just like in the front line serving food. So can you talk a little bit about what exactly are these robots doing? Because it's hard to manipulate food. Right. Because it's so complex.

Rajat Bagheri [00:13:46]:
Yeah. So today our go to market is really food manufacturing, which is different than a lot of people think. A lot of people are like ah, robots for restaurants. Right. But, but like, like you alluded to, we are actually focused on manufacturing. Essentially the way you can think about it is anytime you have kind of a meal that you might have in an airplane or like if you get frozen meals from the grocery store or you go to the grocery store in the deli section, the fresh food section, all those prepared salads, you might anyways, meat packing, all these kinds of meals are actually made by people and they're made by people in these big facilities, basically food factories. And the way it kind of works is you have these long assembly Lines. And on the assembly line there's like 12 people and each person has a big tub and they're kind of scooping trays, scooping food into trays or burritos or wraps or sandwiches.

Manav [00:14:27]:
The most mundane tasks humans should not be doing that.

Rajat Bagheri [00:14:30]:
There's no future where humans are going to do this. It's just not going to be a thing. Right? So today that's what we focus on. Right. So we focus on the food assembly, which means that kind of like, you know, like you have to scoop food from a big hotel tub and you have, you know, not crush it. You have to work with any portion size. The customer owns 53 grams of shredded chicken. You do 53 grams, you know, you got to be consistent. Then you have to detect and track and place the carrots into whatever compartment the customer wants, but also spread it the way the customer wants. And of course, you have to do this in a way that's scalable so that you're not making custom software, custom hardware per ingredient or per tray or per customer. But rather it's really an AI driven, more flexible solution. And we do assembly, by the way, because the assembly side is actually a lot more labor intensive than cooking. So most people imagine that the cooking part is the labor intensive part, but what you find is actually the cooking kind of scales sublinearly. In other words, it takes one person to cook at your house. You cook for a few people, like your family. But then if you cook for a thousand people, you actually don't need thousands of humans. You need a few humans who can do big quantities, big batches, but the assembly actually scales linearly. If you need more output, you need more humans.

Manav [00:15:31]:
You see this at In n out. Yeah, there's so many people, like just assembling. Yeah. And person can make 50 flip burgers like 50 times, but not the assembly, which is really smart in my opinion. And I feel also like you would rely more on a human for the cooking aspect.

Rajat Bagheri [00:15:47]:
That's right.

Manav [00:15:47]:
Because they're like, visually there's much more data points that they're tracking. So it's a good relationship between the robot and the human and you're actually taking the mundane task out of it.

Rajat Bagheri [00:15:58]:
That's exactly right. There's even one more thing which is like, because, for example, you could talk about burger flipping, right? There's one person doing 50 burgers at the same time. It's really hard for a robot to generate an ROI in that. Why is that like the pitch to like, let's say burger joint Is hey, we're going to put a robot that's going to flip the burgers or do the frying. That sounds great in theory, but the issue is that robots aren't perfect and sometimes they fail. So you essentially still need that human. You didn't actually like say reallocate that human to do different tasks. You still need that human. It's like, okay, we're going to pay me all this money. I as a restaurant owner have to pay you all this money for a robot and you still didn't even get rid of a person or like really meaningfully generate roi. That's why cooking is hard, because there's so few people doing it that you really need to be super autonomous to really add roi. On the other hand, assembly side of the house, there's so many people doing it. If you can even help a little bit, then you can actually generate a big roi.

Manav [00:16:45]:
One thing I like that you were talking about how there are tough conditions like really low temperatures or really high temperatures and people like really get exhausted like when they're doing these tasks. One thing I was really thinking about is like how is this infinitely scalable? So for example, self driving cars, they're infinitely scalable because more data they're collecting. So it's like scalable. But in your case case it's going to. Is this more of a custom solution like for, let's say for Chipotle, for Cava, like you're going to have to program each customer separately.

Rajat Bagheri [00:17:13]:
It's a good question. We think a lot about this, right? So even, you know, even the self driving car case, I would say it's not actually, it's not just that instantly scalable. It's instantly scalable so long as you have the right training data. In other words, right? Like I can make a self driving car that goes around San Francisco, great. But then if I go to Phoenix, maybe Phoenix is a bad example. If I go to like New York City, a little bit more complex environment, it's a very different layout and you need more training data of New York City city, right. So it, it really comes back to train media, we kind of think about chef a a similar way from a chef perspective, like what are the things that change from customer to customer? We'll just talk about it. Well, they have different ingredients and of course ingredients is a big deal, right? There's different ways of cutting the food, there's different ways of cooking the food, there are different way of growing the food, different suppliers of the food. Anyways, there's A big kind of whole thing about ingredients. There's different portion sizes, there's different trays, they usually have different conveyors and then maybe have different placements within the tray. And there's other dimensions, but those are kind of the five dimensions. So the way we think about it is, well, how does a human deal with this? Humans have to deal with all those five dimensions too. And of course, what a human does is they just walk over to the line and they use their eyes and they have a hotel pan of food, a big tub of food, and they say, okay, well where in this food do I pick from? To be consistent? And then of course, they have their arm and they generate emotion, they have emotion, they use with their arms, right? And then same thing with their plate. With their eyes, they figure out where the tray is, they need to place, and then they figure out where the tray is and they place. So in other words, just to simplify a little bit, it's like, okay, well, the human has eyes, they have, let's say, cameras, they of course have brains, computer vision, ML, an arm which you can kind of replicate. Not exactly, but of course good enough. With a six degree freedom robot arm. And they have different utensils. So we designed our system based on this idea, which is that we're going to have kind of a human equivalent. It's going to be the same footprint as a human. So in other words, every customer, no matter if it's a Thai place or Indian place or we don't care, Chinese place, we don't care. So long as you have humans there before, our robot's going to be able to fit. And so long as the robot can fit physically, we have the same components at every single customer, which is a camera 6 off arm interface for different utensils. And then you can kind of divide the problem into picking and placing. The picking is where you have to deal with things is like the ingredients. And on the picking side, the way we think about it is like, okay, well, so long as we can abstract away these material properties of food into an AI model and the right utensil, then we can essentially pick anything. And one way to think about this is like the utensil you're going to use to pick up a long green bean is going to be very different than the utensil for a sauce, right? And that's going to be different than the utensil to put oregano on a, sprinkle oregano on the top of it. Of course. So you need like a class of utensils but it's not like you need thousands of utensils. You need, like six or seven different classes of utensils. And then, of course, you have to deal with all these material properties. So, like, okay, well, you have to work with any kind of tomato, no matter how the prep cook cut it today. Because of course, there's different temp workers who are cutting it slightly differently every single day. That's really where this AI policy can be useful. It's like, okay, given this topography, where do I pick from to be consistent?

Manav [00:19:51]:
Isn't that like infinite permutation combination of.

Rajat Bagheri [00:19:54]:
No. So then basically what you can do is. Then what you can do basically is like, okay, you have a hardware system. System, and you mass manufacture the hardware system. You have a library utensils, and you mass manufacture the library utensils, right? And then you have one code base, and every customer gets the same code base. And then there's basically different AI models for different classes of ingredients. And then you train these consistently. Kind of like online learning, right? So, like, the more of these robots you have in production, the more training data you get, in other words, the more diverse, the more. Like, one day you saw this kind of tomato cut, and this, the other day it was cut a little bit differently. Okay, great. That's training data. And you take all this production data and you make your models better so that next time when you see a new customer and they cut their tomatoes slightly differently, you can adapt to it really quickly, right? So it's not like there's no custom software. It's all kind of different AI models. And that same thing is true for placement, right? So, like, you know, different customers have different trays. Like, sometimes they're, like, round, sometimes they're black, sometimes they're white, sometimes they're like. There's different combinations, compartments. But we have a model that detects and tracks these containers or these wraps or these breeders or whatever. But then once you see enough of these, when you see a new customer container, you don't have to do any work. It's just your model works. So and same thing for conveyors, right? Like, you know, there's white conveyors and blue conveyors, high conveyors and low conveyors and slope conveyors. But we have a lot of perception and computer vision. We built so that if we go to a new customer, we don't have to. We don't have to do anything. So it's basically like a flywheel at.

Manav [00:21:12]:
The end of the arm. Do you have, like, what kind of sensors and cameras Are you using at the end of the arm?

Rajat Bagheri [00:21:18]:
Yeah, that's a good question. At the end of the arm there's no cameras actually. So we, the robot has fixed cameras. And so you don't have to do any transformations or anything else, but just like there's like a camera that's kind of pointing at the ingredients and there's a camera pointing at the conveyor. It's kind of fixed cameras. The end of the arm actually does have a force torque sensor. So we can basically detect how much force you're kind of like the robot's feeling. And that's useful for things like, you know, you want to pick up like certain, for example, certain ingredients, like leafy greens. So leafy greens, like it's often like weight's obviously important, but also volume's important. You want to make, you want to feel like your salad was like full. So we can use like a force torque sensor to figure out how deep to go into the bed of leafy greens because there's a lot of air inside of it. It's not just weight based. You can't just use perception. You actually have to have some force sensing.

Manav [00:21:55]:
And you still need humans to fill the tray back up. Right?

Rajat Bagheri [00:21:58]:
Yeah, yeah, so exactly. So there's actually a separate job. There's a separate person whose job it is to refill. And that's because of course, even for.

Manav [00:22:05]:
Humans you have to refill and that cannot be automated.

Rajat Bagheri [00:22:07]:
Well, I'm sure it can. I'm sure it can one day. And there's, I mean, you can use, imagine using amrs or different. Yeah, there's different ideas you can take. But I think it's not as. Again, it's kind of the cooking problem I mentioned there's only one person doing per line. Right. Whereas assembly has like 12 to 15 people per line, times like potentially 30 lines. So just like in terms of market size, yes, there's a lot of things we can, could do, but I think we want to like really focus on the biggest part problem.

Manav [00:22:32]:
Got it. I think I want to talk about data. So OpenAI, like other LLM companies, they've been using like online, socially available data on Twitter, Facebook. So I know you guys have had this chicken and egg problem of how you're going to collect the data you need to deploy the robots. So how did you solve that problem and how long did it take to solve that problem?

Rajat Bagheri [00:22:54]:
I think also. So it's a really hard problem and I think the biggest thing we did is, well, so maybe I guess one piece of Context, Right. So about data for robots. The reason OpenAI and LLMs were able to kind of succeed is because they had data they can download from the Internet. The issue with robots is there's not a lot of data you can download from the Internet. Like, I can't just go to, like some website and here's like, how you manipulate this bottle or this can. Sorry. Or like food without. Like, here's how you pick up blueberry without crushing it. There's no data about that off the shelf. Essentially, you have to generate. And there's a few different ideas on how to generate it. So one idea to generate it is like, you can do it in a lab and data in a lab is a good idea, but it's a good idea in theory. But the issue is the physical world is very complex. It's very random. It's not very robust. It fails.

Manav [00:23:35]:
But you're still in an enclosed environment.

Rajat Bagheri [00:23:37]:
We're in an enclosed environment, but the issue is that. So we did this, right? You get the customers, let's just say a tray of onions. You literally get it from the customer site and you bring it to your office and you bootstrap a data set for chopping chopped onions. But then the next day that person was let, or they left because it's a temp worker. They brought a new worker and that person cut them slightly differently and it doesn't work anymore. And there's just so much variety in food. Like some days like the onions a lot. Yeah, it's just. It's very highly dimensional. So anyways, you can bootstrap a data set in a lab. It's just not robust to dealing with all these edge cases you see in the physical world. Okay, so idea number two is use simulation. Simulation is actually potentially a good idea. But the issue with food in particular is that there's no deformable physics. Physics simulations out there. So in other words, what I mean by this is, if I'm Waymo or Cruise, then a lot of the machine learning is actually in simulation. Or even if I'm doing bin picking or palletizing, these kind of physical. If I want to pick a box or a case for manufacturing, a lot of that can be done in simulation. But food is deformable. You can squish the blueberry. That's hard in simulation. So simulation doesn't really work. Okay, well, there's another idea which is, okay, like imitation learning, which is another new field, especially with transformers, have been pretty powerful. Powerful. We're actually doing a lot of imitation learning. As well. The issue with imitation learning and the idea there by the way, is a human mimics emotion, the human does emotion and then the robot copies it. So this is actually like if you've seen Optimus and they're doing a lot of imitation learning and Tesla FSD12 is doing a lot of imitation learning. So imitation learning is actually quite useful. But again, it's good to get a starting policy of how to do something, but it's not robust. Again, I would say to the physical world there's so many edge cases you need and so that's number four. And then the final idea is of course production data. So we said, okay, we can take and choose bits of these, but we really think the best way to get production robots is production data. That's when we face this kind of chicken neck problem. To solve this, really the big thing we had to do is work on our go to market. So what I mean by this is when we first started the company, we were focused on fast casuals, specifically assembly for fast casuals and even more specifically delivery orders for fast casual, if that makes sense. Basically.

Manav [00:25:37]:
Like, are you talking about Amy's and all those?

Rajat Bagheri [00:25:40]:
No, no. At the time we were focused more on brands like Le Chipotle, Sweet and, and all these guys. At the beginning, the issue we came up against is that. Let's just talk about sweetgreen. The person who's at the sweetgreen counter making beer salad, she isn't just, she's not just on the assembly counter. She's there for like four hours on the assembly counter for lunch rush and dinner rush. It was like two hours, three hours apiece. So what is she doing for the rest of her shift? Well, she's like cutting things, she's cooking, she's cleaning the floors, she's cleaning the bathrooms. So if you want to have a one to one equivalency, you need to essentially have AGI. Does that kind of make sense? Because she's to have a one to one equivalent to Sally who works at Sweet, you need to do everything. So that can't happen. Let's just say we limit the problem to just the plating, the make line. Even then she's doing 90 ingredients. It's not like Sally's doing the leafy greens and Bob's doing the meats and Joe's doing the sauces.

Manav [00:26:25]:
Because they ask the customer, you want brown rice, you want white rice and based on that they're serving that.

Rajat Bagheri [00:26:30]:
Yes. But let's just say even if it's a delivery Order. So you order on doordash and they give you a little. It's an API call. Even then, one human is doing every ingredient. If you're a robot company, your one robot needs to do every ingredient. And if you can only do 30 of those 90 ingredients. Ingredients, then you didn't really give them anything. It's the same problem as what we talked about with cooking. Slightly different. But we said, okay, well, this seems like an exciting market, but it feels like not the right go to market. So what we said is that to solve this data problem, we need to fix our go to market. We need to change our go to market. So then we discovered the manufacturers. And the idea of the manufacturers is that because it's higher volume, instead of 1,000 trays a day, you're making 50,000 a day, 20,000 a day of meals. Well, you have a lot more volume, so you use an assembly line format. So now Sally, Sally's on Station 1 of the assembly line. Sally's doing rice for the next four hours. Sally's doing rice and Bob is doing the chicken. And Bob's just going to do the chicken. And of course, it's not that simple. Like they're going to change over the line and they're going to do different meals throughout the day. But my point is that Sally's now doing five ingredients throughout the day. Sally's not doing 90 ingredients, as in the case of Sweetgreen. So five ingredients is a small number. Five ingredients is something we can bootstrap in a lab. So that's what we did. We said, okay, first of all, let's change our go to market. So we selected food manufacturing, which is kind of a high volume, relatively low mix product, mix environment. Then we said, okay, we're going to use, we're going to ship one robot. And to get one robot into production, we need to onboard five ingredients. We bootstrapped five ingredients in the lab that were relevant for the customer. We use that to ship one robot. Now, as soon as we ship one robot, we get production data, which is the key. Now, once we have production data, we use that to train our models and the AI gets better. And so now we do really well with those five ingredients. So now when we try to do the six and three ingredients, then the model performs better. We keep on getting more and more data, and now we can do 15, 20 ingredients. So now it's like, okay, let's ship robot number two. And then you get the idea. Now the more of these robots we ship into production, the more ingredients we can do. Now there's a better claim that we can ship and another robot. Another robot. So that's how we scaled within our current customers. And then we took this to the next customer and we're like, well, customer. Well, customer number two, we've already done a lot of your ingredients with customer number one. So from day one we're going to have high utilization. And they're like, okay, well great, let's put your robots there. So we ship them to robots and now we get more diverse training data. It's slightly different. They cut their ingredients slightly differently, they cook them slightly differently, which is great. Now the AI gets even better and we do the same approach. Okay, well let's scale within that customer. Then we go to the third customer and you get the approach. So basically the more of these robots we ship, the more training data we get, the more food manipulation we can do and the, the more manipulation we can do. Not only can we scale in manufacturing, but one day when we want to go back to the lower volume customer segments like fast casuals, prisons, hotels, universities, K12 stadiums, cruises, all these other food service establishments, we've already learned how to do their ingredients. We know how to spread cheese and not get it clumped up. We know how to spread sauce so that it doesn't get into the edges. We know how to not spill low viscosity stuff all over the place. We know how to not crush the blueberry. We know how to place shredded carrots in one little compartment versus not having them spread all these little things. Things. We know how to be consistent. We know how to do food inflation well. So by the time we get to lower volume players, we're the best game in town. We have more data than anyone else.

Manav [00:29:25]:
When I saw on your website you guys are doing Indian food, I was like, Indian food is the hardest to. Because Indian food has so many variables, like so many little ingredients that goes on to it. But okay, so back to data. Some people would argue the data that you're collecting in itself is equally or if not more valuable than the rest of the hardware component that you, that you're building. Would you say so?

Rajat Bagheri [00:29:47]:
Yeah, I mean, yeah, here's the thing, right? Like, I think Silicon Valley especially has like this idea of, okay, like software can solve every problem in the world. Like, but the thing is like software needs a body, right? Like food. Like things like food. You can't have pure software make food for you. You need a body, right? So I think it's kind of like, I mean, one example I like to give is like, I think a lot of the innovation, like, certainly the iPhone had a lot of hardware innovation, like the battery technology and multitouch and there's a bunch of hardware innovation too, but really iOS and the software is a big innovation. Or the Kindle, like, certainly there's a lot of hardware innovation, but there's a lot of software innovation. In other words, I think the hardware is kind of a, it's kind of a vessel or a vehicle to deliver the software. Right. I don't think the data is possible without the hardware. In other words, it's kind of like, yeah, sure, the data is a valuable thing. That's the moat, that's the edge. That's the thing that separates us. But could we have gotten it without hardware? No, absolutely not. It's very, it's like intertwined.

Manav [00:30:36]:
But you could possibly license it to other robotics companies.

Rajat Bagheri [00:30:39]:
Yeah, I mean, we wouldn't do that, but yes.

Manav [00:30:40]:
Okay. It's a possibility.

Rajat Bagheri [00:30:42]:
It's a possibility. Well, and I guess one kind of corollary to that I would say, say, is like one thing we could do is like, or we were already doing is like our software and data is agnostic to hardware. So like, let's just say our current ARM manufacturer is giving us troubles or whatever, then we can very trivially go to another ARM manufacturer, or we want to go to a different camera, or we want to go to whatever, whatever new thing we want to do. We built our software stack and our data interfaces to be kind of agnostic to hardware.

Manav [00:31:07]:
Got it. All right, let's talk business. So average assembly line worker would, I don't know, would make for 50, 60k, like something around that a year. So how do you price for your services? Is it like a yearly subscription?

Rajat Bagheri [00:31:18]:
Yeah, it's a yearly subscription. So most of our customers kind of run two shifts a day.

Manav [00:31:22]:
How many hours is that?

Rajat Bagheri [00:31:23]:
So each person's working eight hours. If there's overtime, more like 10 hours. Right. And then there's two shifts a day. And the reason there's two shifts is there's shift in the food industry is always. Not always, but often cleaning shift, kind of do sanitation and stuff. So anyways, most lines are running 16 hours a day. So our robots are usually running 16 hours a day.

Manav [00:31:38]:
They are. Wow.

Rajat Bagheri [00:31:39]:
And they usually run, I mean, it depends on the customer, but five to seven days a week and then. Yeah, so we're charging a yearly recurring fee. And that's of course less than humans. So each robot's kind of doing the work of two people. And of course again, nobody's being fired. Right. Those people are going to do a different task. It's probably better to do that other task which is less redundant. And so, yeah, like our business models, we're going to charge them a small kind of implementation fee we call like an NRE non recurring expense. And that's mostly just for like the initial configuration installation. Like we're going to have a couple of applications engineers who fly out and deploy the thing and train the team, things like that. But there's no big capex upfront. Right. So they basically say, okay, look, you're going to pay a small non recurring.

Manav [00:32:16]:
Expense and that every month or annually.

Rajat Bagheri [00:32:17]:
The NRA is just one time. Just literally as soon as you sign the contract. That's just for us to say, okay, we're going to ship you this thing. It's like official. Yeah. And it's like there's some labor cost actually doing the installation. Right. And then once that's done, there's no capex. You're going to pay us a yearly recurring fee. That yearly recurring fee is going to be less than the cost of your people. Those two people. And then the ROI for them is really like, yes, Chef is cheaper, but that's honestly like number five on the list. The biggest ROIs are really like, you have to remember there's a big labor shortage shortage. So there's like.

Manav [00:32:44]:
And high turnover.

Rajat Bagheri [00:32:44]:
And high turnover. Right.

Manav [00:32:45]:
So which means you need HR people to find people.

Rajat Bagheri [00:32:48]:
Yeah, so there's a lot of things there. And not only that, that is definitely true. And the other thing is that like they often can't run all their lines. So if they have 10 lines, they can only run seven because they just don't have people. They don't have enough supply of labor to meet the demand from customers. So they're under producing. So if we can say, okay, well line eight over there, why don't you put these eight robots and line eight can run now that's a ton of money for you. Right? So increasing revenue is something we think a lot about. Oftentimes we can help incre increase throughput. Average throughput robots don't fail. They don't slow down. They just like go. Whereas like six hours into the shift, humans, they get tired. So Chef can usually increase average throughput, which is again a lot of revenue. We usually think, I think the best businesses are businesses that increase revenue more than save cost. So we help with that and then we also help with yield. We waste less food. Well, so it's interesting, right? So when you go to the grocery store and you get a meal, you're promised some amount of calories, you're promised some amount of weight. Yes. It's not always accurate, but these food companies really strive hard to make it as accurate as possible. But to do this, and their thinking is like, look, if we underpick the customer, me really annoyed. We really don't want to under undersell the customer. We'd rather put them, we'd rather be a little bit higher. So what they do is they buy us up on average. So if the target weights 8 ounces, they're like, okay, let's do 8.2 just to be safe. In the industry. This is called giveaway. Humans on average are over depositing. And this is just the way the industry works. There's no other solutions for this. This is the way the industry has accepted to be. But of course you have robots, then you can be very exact. So we can help say, okay, well if the target's 8 instead of putting 8.2, we're going to make sure it's 8.01 smudge over what you need it to be.

Manav [00:34:19]:
This reminds me of the Chipotle CEO came out recently and he's like, they're giving excess food so they throw less food. Yeah, so that was quite the interesting thing. Okay, so this is going to be infinitely scalable in the coming years. You're going to get, keep getting better. The AI is going to keep getting better and better. And this will be shipped worldwide, I'm guessing.

Rajat Bagheri [00:34:38]:
Yes, we're already in Canada now. Canada's obviously not too much farther from the U.S. but that was a kind of first international deployment. Now we're, we're kind of actively thinking about the UK and Europe, but that's kind of the next ones. And we're like thinking about Australia as well. More than likely UK is going to be the next one that we go to.

Manav [00:34:51]:
All these countries have aging population, aging population.

Rajat Bagheri [00:34:54]:
And not only that, like one thing that's nice about food. That's why honestly, food is so exciting, is like every human on the planet Earth needs food and assembly. Universally, the plating aspect is 70% of labor for everything except fine dining. Assembly is the labor sucker. That's the one that takes the most labor. So I think we're well positioned. There's no reason why we can't go to more or less every country.

Manav [00:35:13]:
I can see infinite applications, frozen food ready to go, meals, meals at Trader Joe's. There's so many yeah.

Rajat Bagheri [00:35:21]:
And then down the line, I mean, again, the idea is not just manufacturing down the line. We want to take this to actual fast casuals. Again, but to be clear, not just front of house, the fast casuals also have back of house lines for delivery. So, like, Chipotle has a back of house line just for delivery, like doordashing, Uber eats and stuff. So we want to go to fast casuals. We want to go to ghost kitchens, universities, K12s, corporations, schools, like, sorry, K through 12s, stadiums, cruises. I mean, there's so many applications. That's why food is so cool. It's just gigantic.

Manav [00:35:46]:
Yeah. Would you be going to India one day?

Rajat Bagheri [00:35:48]:
Yeah, I mean, certainly. I mean, I think we definitely want to go to India as well. The only trouble with countries like India is that the wage is already quite low. Right. So, I mean, the way that a lot of food companies think is, like, if they're gonna get a robot, they want an roi. They really care about an robot roi. That's important to them. The labor environment is a little bit different. Right. The reason that the pitch is so compelling in the US And Canada and things like that is, like, there's a big labor shortage, but India just, of course, has so much population that there's less of a labor shortage and the wages are lower. So. Yes, But I think the product's going to have to shift. It's going to have to get even cheaper, it's going to have to get even more autonomous to be able to generate an ROI for those guys.

Manav [00:36:21]:
Got it. So what's next for Chef Robotics? Like, are you guys going to be raising another round? Are you guys going to be, like, what are you going to be focusing on, Focusing on for the next, like, let's say a year or two?

Rajat Bagheri [00:36:31]:
Yeah, let's see. So there's a few things that are top of mind. So one thing that's really exciting about Chef is we have a really good set of, like, existing customers that are quite big customers. I mean, these guys have, like, lots of different plants all over the world. And so really landing and expanding, which is a lot of customer success. Right. And really, like, just essentially living with them, like, really making the product extremely good for them. Obviously, there's a lot of product and engineering work to do that, but we really spend a lot of time on that because we think that if a customer buys two robots, that's not that impressive, but if that Same customer buys 50 robots, that's extremely, extremely impressive, because they're not going to buy 50 unless the thing really works. But of course, we have to really make it work then. Right?

Manav [00:37:02]:
And that's the sweet, sweet recurring revenue per robot per year. So it's good for your company, good.

Rajat Bagheri [00:37:08]:
For the business as well. And by the way, the data, flywheel, of course, helps everybody else too. So it's kind of this big flywheel. So, yeah, kind of scaling within current customers. Now we're at a point, we obviously kind of announced what we do recently, and now we're at a kind of a point where we do want to scale, go to market and sales and marketing. Honestly, we've been very quiet. I mean, as you've probably seen, we've been very quiet. And the reason we've been very quiet is like, we felt like we had something, we're onto something, and we're so focused on current customers that getting more sales folks or marketing folks wouldn't really do much. We can't even handle the demand. But now we can. Now we feel like the product's ready to scale and we have the team to execute against it. So it's like, okay, really scale and go to market. So that's getting net new customers. And then I would say number three is really kind of continue to really invest heavily in AI. So we have this dedicated AI team now who are kind of using imitation learning and learning from demonstration and also decision transformers to kind of say, okay, well, let's use production data. Let's combine it with, like, imitation learning to learn new SKUs, new products, new ingredients. And let's try to build, like, more of a generalized food manipulation model. And, like, make that flywheel even faster. Right. How do we. How do we assist that flywheel? How do we make the flywheel even more effective?

Manav [00:38:12]:
Wow, that's a lot on your plate.

Rajat Bagheri [00:38:14]:
Yeah. I mean, and I guess maybe to that point, we continue to grow out our leadership. So we hired a coo. We had hired a great head of software.

Manav [00:38:20]:
How many people are you total right now?

Rajat Bagheri [00:38:22]:
I think it's like 33. 30. 33, 34, something like that.

Manav [00:38:25]:
Everyone bases him Francisco.

Rajat Bagheri [00:38:26]:
Everyone's in San Francisco. We have a policy of like, everyone's in the office every day.

Manav [00:38:30]:
That's amazing. I love that. Where the entire world is moving to remote. Like, I like that. If you were not doing Chef Robotics, what other, like, opportunities, like, you see, or like, what's. What's an area that's being underserved right now, or you would want to see more engineers working on that problem?

Rajat Bagheri [00:38:45]:
Yeah, that's a good question. So I think my general Thinking is that the most important problems are in the physical world, right? It's like these, like, yes, software's awesome, right? I don't like marginal cost is here. There's a lot of great characteristics of software, but the real important problems like food safety and hunger and water crisis and terrorism and war and all these are physical world problems. So I'm really excited about using all the beautiful properties about software. Right? Marginal cost of 0, easy to scale, infinite scalability, all these, not infinite, but high scalability, all these things and kind of apply them to the physical world. I think robots is a very natural instantiation, but it doesn't have to be like Uber, for example, I think is a physical world software instantiation. It's like Uber is a purely, not necessarily purely because they're servers and stuff. Like mostly essentially a software product, but it has a physical impact. So I think those kinds of companies are really exciting. Or like Anduril, for example, is using software and combining with hardware to have military applications. I think just more of these kinds of companies, software plus hardware. I think if you look at the most powerful companies in the world, every single one of them from Apple, Nvidia, Microsoft, Google, they're all hardware companies at some point. Part of it, like, you know, if you look at even like, you know, like, obviously Amazon is right, they're doing a lot of like physical delivery and stuff. But even like Google and Microsoft, you don't colloquially think of these companies as hardware companies, but a really big part of their business is the server business. Like they're leasing out servers.

Manav [00:40:03]:
The only company that's not too much hardware is Facebook.

Rajat Bagheri [00:40:05]:
Yeah, but even Facebook has huge data centers. I mean, Facebook is like spending a ton of money on compute and GPUs and data centers. Like, data centers are in some regards, like, very similar to robots. Which is to say, like you think about the data center center business model, it's like they buy, you know, Nvidia GPUs, they buy like Server Rex from Dell and all these people and then they get these facilities and then they kind of rent out the servers and they'll install a lot of software to make these servers useful. Like platform as a service or infrastructure. As a service, they'll install software, make it useful. They're essentially leasing software plus hardware as compute or gpu right on the cloud. That's essentially what robots do, right? Like almost all of our hardware's off the shelf and then we install a bunch of software and we're essentially leasing hardware and software so you can Essentially think of chef as a server rack on a cart in some regard. I know that's not totally fair, but it's a similar analogy, actually, which is off the shelf hardware plus software and rent it to provide value for customers.

Manav [00:40:54]:
That's a beautiful analogy. What is a book that you recommend other people read? You can name, like, three books. And what is a contrarian, unpopular opinion that you have that a lot of people will disagree and argue with you on Twitter about?

Rajat Bagheri [00:41:08]:
It's a good question. Yeah, I mean, I guess. Okay. Like, I mean, there's a few that come to mind. I mean, I think, like, it's a trite answer, especially in Silicon Valley. But the book that probably had the biggest impact for me is the Steve Jobs biography by Isaacson. So, like, again, this was like, you know, 2011, right? Or something. So, like, so for me, this was kind of like high school, right? And before Cafe Mocha and all these things I told you about initially. And so, like, I remember, like, him passing away and I was like, I didn't even. Of course, I didn't know much about Apple or I had gotten the iPhone, but I didn't really know about this guy individually that much. I hadn't really done a lot of research. But then he died, and I read this biography and I was, like, really, really inspired. And like, arguably this book had a really big impact on me. So anyways, I think it's just a classic and it's one of those books that are really good. And then generally, I really like biographies. I would say I really like Disney's biography by Neil Gabler. I really like the Rockefeller biography.

Manav [00:41:53]:
Did you read the Elon Musk one?

Rajat Bagheri [00:41:54]:
Elon Musk one's quite good. I mean, there's like a lot of, like, yeah, again, like, classics.

Manav [00:41:58]:
Richard Branson also has a good one.

Rajat Bagheri [00:41:59]:
Yeah. So I think these biographies are really powerful. And then this second book, I guess this another book. I'm going to say this is going to be more contrarian and people have their opinions about it. But, like, I really like the Fountainhead.

Manav [00:42:09]:
Never heard of it.

Rajat Bagheri [00:42:10]:
Ayn Rand. A lot of people have opinions about Ayn Rand. She's very extreme. But basically, like, the fountain has this idea of, like, it's a philosophy of what's called. She calls it objectivism.

Manav [00:42:17]:
Like, capitalism's a good.

Rajat Bagheri [00:42:19]:
Capitalism is a good thing. Which is like, basically, it's like the individual. Like, basically she like, lauds and says it's okay to be individualistic and just focus on what you want to do. And, like, Build stuff and like be capitalistic. That's an okay thing to do. And by the way, if you do that, society's better off because of it. Like, in other words, like, like by having Elon Musk be an individualistic and make a lot of money, the world is like way better off.

Manav [00:42:39]:
Which is, which is true.

Rajat Bagheri [00:42:40]:
Which is true. So. And I think at the time I was like, you know, it's like as a young person, it's like, it's like, oh, do I like, I think there's a lot of like, it's very easy to be like, ah, let me do all these different things. But I think I read this in early college and I was like, wow. She's basically just saying focus on making a thing really successful. Basically what she's saying. Instead of trying to be impactful or publicly cool or societally known, don't try to worry about that. Just build something really useful and make a lot of money. And if you do that, then society will be better off.

Manav [00:43:05]:
So I think people will argue that that actually makes everyone much more productive. But on the downside, it causes loneliness, isolation. Do you agree with that or No?

Rajat Bagheri [00:43:16]:
I don't think that has to be so. Individual is. Individualism is not. That doesn't mean like, I don't mean individualism as like lack of societal connection. Individualism is as like go off and do your own thing from like, let's start a business or, you know, build a thing or something. But I don't think it has to be like, I think they're like orthogonal in my head. Like can be separate dimension. Which is like, you can have a great social life, I think, and actually like, if you like for example, another biography really like is like Picasso one. I forgot the author about it. But like he's extremely social, like, and outside of his, like I read his.

Manav [00:43:43]:
Routine, he had an interesting routine. Like he would hang out during the day, meet all his friends, and then he would work from late afternoon till late night.

Rajat Bagheri [00:43:52]:
So I actually think you can be extremely prolific and productive, but also have a great social life. All I'm saying is just focus on yourself and just make a successful thing. And that's going to make society better off. So that idea I thought was cool.

Manav [00:44:05]:
So what I've observed from researching you, talking to you, you are really good at picking the right problems. And I feel like a lot of people, myself included, have been unintentional with a few vent. So what advice would you give to people, like, when before they're starting something, what should we decide to focus our time on.

Rajat Bagheri [00:44:24]:
Yeah, this is. Obviously, there's a big can of worms here. Right. I think broadly there's like, two different schools of thought. I've heard that I kind of, like, think about. It's like one school of thought is this passion hypothesis, which is kind of you look inwards and you're like, what am I passionate about? And you do you do that. I think that can work. And if you're passionate about something done easy, especially if the passion matches up with a big market and, like, a big opportunity. But the truth is, is that, like, I'm not like, before Chef or, like, before prototyping, Like, I wasn't inherently passionate about, like, the visually impaired or food. Like, I love robots and AI. Yes. But honestly, food is not. Like, I'm like, I don't. I'm not like, I never loved cooking or, like, went to culinary school or anything else. I. So I kind of. Yeah, So I wasn't deeply passionate about it. And, like, before Chef, like, I wasn't, like, passionate about really this thing. And I think a lot of my founder friends are like this. It's not like Aaron Levy's super passionate about cloud storage. He's like, you know, you. You don't wake up and be like. Or like, HR software. It's like, let's go now. Like, most people are not super passionate about these things, Right. So anyways, there's this other hypothesis which is like. Which is like. So the passion hypothesis says that, like, passion leads to success. And this is by Cal Newport, by the way, which I really like his ideas here. And there's another hypothesis which is like, that. It's the equation's kind of. You know, it's inverse, right?

Manav [00:45:28]:
Is it impact?

Rajat Bagheri [00:45:29]:
It's. Well, it's. Success if you're successful that leads to passion. I tend to. I tried the. The passion thing, which is like, okay, like, what am I? What am I? What do I care about? And, like, yeah, I guess I figured out that I like AI robots, but then I did the opposite, which is like, what's the right company I can start that'll be successful? How do I increase probability of success? Because as soon as things start to go well, it's very easy to become passionate. Like, if you're really good at something, if you're just winning, then you'll become passionate. So for me, this took the form of, okay, let me find a really big market. Like, okay, let's look at the data. What is the biggest market? It's like labor, industry. It's like, okay, okay, food is the biggest market tractable. Then it was like, okay, there's a big problem. Then it was like, okay, assembly is the right go to market, not cooking and prepping. Then it was like, okay, manufacturing. Like, today, if you look at chef and you're like, why did he end up or we end up in food manufacturing, it's kind of like, not sexy. It's kind of boring right on the surface. But then if you look at this history, it's like, ah. Like, you follow the steps, like, aha. Like, today they're building this, but tomorrow they want to build that. Then it's like that. That's really. It makes sense. So I tend to think, like, figure out what's going to make you successful and then passion would follow. But that's, of course, only true if you don't have something inherently passionate about from get. From the get go.

Manav [00:46:30]:
Well, with that said, I think you, like, thank you so much for coming on the show. You answered, like, amazingly. Thank you so much.

Rajat Bagheri [00:46:36]:
Yeah, thank you, Manav. I appreciate it.

‍