Rhoda AI and the Future of Robotics: Andrew Wooten with DVC's Nick & Marina Davidovs

Background is incredible. Six companies.

I genuinely think this is the next industrial revolution.

Why did you choose stealth strategy?

Rhoda's approach to robotic intelligence is very different.

Few-shot learning.

So, we can officially call all other robots unimaginative.

We're solving the problem of bringing the intelligent robots into the real world. It doesn't rely on robot data at all.

What do you think motivated him to keep silent?

So, effectively you're giving robots imagination.

That's that's a beautiful way to put it, yeah. What's your kitty's name?

Welcome to another Deconstructed VC podcast. Today, we have an amazing guest, Andrew is a co-founder and chief product officer of Roda AI.

And we are very excited because we wanted to talk about that for so many months.

It took so much effort to keep my mouth shut about this company. I was just like every time people would like "Oh, I'm investing in Figure. It's such a cool technology." It's like I shouldn't say stuff.

We appreciate the sacrifice, Nick.

Yeah.

But now we can.

Finally.

Finally. Welcome to the podcast. Like literally like I'm so glad we got you on the day of the announcement.

We're excited excited to be here.

edit and make it like polish it and post it as soon as possible because I am dying to ask you all of the questions.

And share all of that answers with all our audience because yeah, we are really excited and today is a special day as Nick said that Roda AI came out of stealth and we have couple of questions like to to cover like blitz questions about the company. So, when it was founded?

Yeah, so we started in October 2024. So, we think of October 1st, 2024 is our first day.

October 1st, okay.

So, we'll make you present.

So, you've been in stealth for 1 year and a half.

yeah, yeah. And you got to have a birthday, so we, you know, we got to got to put a pin in it.

How many people in team?

we're about 50 right now, just over 50.

very quickly, what the company is doing?

Yeah, so what we're building at Roda, actually, I'll start with the problem. The problem we're solving is that of generalist intelligent robots. So, I I like to think that there's two categories of robots in the world. there's factory robots. These are very reliable, very strong, very robust robots, but they just do the exact same motion over and over again, every day, all day for years. And that's very helpful in factories, but most tasks can't be automated that way, because most tasks have a little bit of variability. Anything from, you know, the way a box or a package comes in, it's a little bit different every time. tasks change all the time. So, while this group of robotics is very powerful, it's just insufficient to, you know, automate the vast majority of tasks. The second category is what we call intelligent robots. These are robots that leverage neural nets or or deep learning.

They collect data, and using that data, they can do some variability in their tasks. However, this second category is stuck in labs. So, there's a lot of demos on YouTube, Twitter. There's a sec There's effectively zero deployments, and and even the demos we see, they're very carefully choreographed. So, what we're trying to do is we're we're solving the problem of bringing the intelligent robots into the real world. And that's that's the mission of Roda. We want to create generalist intelligent robots that can accomplish tasks that never been automated before.

I like that you used the word choreographed demos, because like all of the dancing robots, all of the kung fu robots. What do you What do you think about those?

a product person, I didn't want to dismiss anyone, so I started by studying the TAM for dancing robots. And after after a bit of time, I realized that it's not a market we should be going after.

Mhm.

It's probably quite small.

There should be some market for that though. You know, like belly dance robot.

Could be. I I didn't spend enough time in that market to comment.

Where does the name come from?

Yeah, so the name comes from robot data.

Okay.

That's that's how we that's how that's kind of robot got stitched together. It is a bit of a Greenland Iceland name because we actually use less robot data than any other robotics company out there. But that that's where the name came from.

That is really cool. Let's talk a little bit about your personal journey. So, you were at Crimson Education scaling at that. Chief Revenue Officer, right?

Yeah, yeah, that's right.

How did you get into robotics?

Yeah, so I've always been very you know, excited about technology. I I I studied computer science back in undergrad. I've always wanted to work on, you know, building new products and creating something would have a real impact on the world of technology. When I was in my sophomore year of college, I met Crimson Education founder his name is Jamie and he had a very exciting team and growth trajectory and I really wanted to get practice, you know, building a company. I joined first to you know, start one of their country offices, but really over time scaled the company internationally. And that was a really exciting experience, but throughout all my years at Crimson, I always you know, wanted to be in tech. I knew that the experiences I was building was going to be valuable for you know, scaling and building companies and I met some amazing people along the way.

But I've always had an itch for developing technology. So, when I you know, finally made it out here to Silicon Valley, I I started diving deeper into AI and spending more time, you know, with with my now co-founders. Once the opportunity came up to get into robotics, I was really excited to dive into it.

And how did you meet your co-founders?

Yeah, so Eric Chan, is our chief scientist, and I've known him since undergrad. so,

From Stanford?

I went to Yale in US, in Singapore. He went to Yale, in the US, and, he was visiting, our campus. We got to know each other. and he's just a very He's a very brilliant guy, also a lot of fun, and really humble, just a great guy to be around. So, we've stayed friends for coming on 10 years now. and we've wanted to build a startup together, but I've always wanted to do things that were closer to, you know, commercial attraction.

Mhm.

He wanted to do things that were deeper, in research. And so, I think the moment which, robotics became at the intersection of research, and you know, commercial attraction around around the corner was was the right moment. And he shared he shared his idea, his approach, with me, shared a couple papers, and I got really excited about the space. And then he introduced me to, Jagdeep, the CEO. Jagdeep is, you know, has an incredible track record of entrepreneurship, but he is, you know, also just a very kind, very humble, very thoughtful person, and I've really enjoyed working with the two of them. And Chengan Chengan is also, our co-founder.

Jagdeep's background is incredible. Six companies.

Yeah. Yeah, this is the seventh.

Seventh company. so, two are public, one got acquired for 1.3 billion by Google, I think.

That's right.

Infinera. He's done like incredibly successful deep tech startups. And he still keeps building.

Mhm.

What do you think motivates him to keep building?

You know, his answer is he he would be doing this even if it was, you know, for free. Like he just he just loves, going from whiteboard to bring something into the real world. and I think that, you know, he's he's better at that than anyone I've ever met. and I think he's going to keep doing it. And, you know, it's I'm I both feel fortunate to be working with him, but I also think it's great for the world that there's, you know, someone with a lot of experience attacking these really big deep technical problems and building real commercial solutions. So, I'm sure he has decades more of generational company building ahead.

So, the technology that you presented today and some of the videos on your YouTube are wild. a lot of folks are showing, you know, robot baristas brewing coffee, but they're we know they're teleoperated.

Mhm.

like the one in Tesla Diner.

Mhm. Mhm.

some can be semi-autonomous, but very imprecise. But, one demo particularly struck me, the one where it's returns processing.

Mhm.

So, I'm I I usually see like how Marina is one of the drivers of like online fashion e-commerce. She orders a lot of stuff, and then she returns all of it.

So, it keeps bugging my mind like how much actual manual labor there is in processing. I think you you you you you can you can break even a robot from just like Marina's returns.

But, that is actually a very hard I'm I'm sorry.

but but I'm keeping like this US economy actually.

Yeah, she she's keeping the US economy alive.

us in business, Marina. So, thank you, Marina.

But, so so I saw the robot, it unfolds it from the parcel, whichever way it was packed by the actual end user customer, which might be completely messed up. Then it has to look like unfold it, check it from both ways.

Mhm.

And then fold it back, put in a transparent envelope, seal it.

Mhm.

That is just so much unpredictability. Two questions. First, how does the robot handle that? And maybe you could also do a little dive into like the general the general technology. What makes it possible?

Yeah. Roboto's approach to robotic intelligence is very different from how everyone's been solving robots. So, I would say the the conventional approach right now, almost maybe the consensus approach, is something called the vision language action model. And how the vision language action model works is it's a model that's pre-trained with a vision language model, so it's like a chat GPT or a Claude, a language that's you know a a model that's trained on web-scale language. And then this model is post-trained with some robot data.

So, this is this is how a VLA works, this is how most of robotics is being solved. You have a language model that can maybe observe a scene and say this is a copper, this is a phone, and then you have you know people collecting robot data through teleoperation or some sometimes through simulation and and the model's being post-trained by that. the problem with this approach is that the entire understanding of physics comes from that robot data set. So, understanding of like what an object is, like this is a cup, this is a table, that comes from the pre-training cuz it knows language. Understanding of how to grasp something and pick it up comes from the narrow robotic data set. And in generative AI, a collected data set is way too small. An example of this is, you know, if if you're training a language model, you need to train it on trillions of tokens.

You need to download the entire internet to train a language model. but in the world of robotics, there is no pre-training. So, everybody's been, you know, collecting their own data for post-training, but they didn't have a solution to pre-training. So, you know, they just they just plugged a language model in there. what Roboto's doing different is we're really focusing on the pre-training. We realized that to solve a problem with any kind of general generality and reliability, you need to have a strong pre-training prior on motion. Eric's belief is that you can't collect your data to generative AI. You have to go to where the data already is. So, what we did is we pre-trained the model on internet-scale video data. And once we had that internet-scale pre-training, it started to understand physics, understand how things move. And you know, to to really just make this practical, the way a robot works is let's say the job is to pick up this glass.

First, it's going to imagine itself in video picking up this glass. So if if the arm is here, it'll generate a video of the arm, you know, approaching the glass and closing it. And then once it's generated that video, it then converts that video into actions. And that sounds like a pretty, you know, intuitive approach, but there's a really powerful distinction here, which is the reasoning part, which is thinking about how do I do this task? It doesn't rely on robot data at all. It's relying on the video data that you collected because you're you're solving the task in video. And what we found is that if you've solved the task in video with very high fidelity, there's only one way you can convert that to action. So there's no reasoning left on the action model. It's purely a translation problem. So essentially, once you've deconstructed this the problem this way, robotics becomes a generative video problem instead of a robot problem.

And we have a lot of data to solve generative video problems. So that's that's a little bit about the technique. The big question is, you know, why isn't anyone done this before? And there's really two main reasons. And they both have to do with video models. One is that video models are generally, you know, most video models you find online, they hallucinate. So even though they can generate a very beautiful image, if you ask it to pick up, you know, pick up this glass, it might do something like this. Like the the arm won't quite close, but the glass will magically lift up. So the physics are just wrong. And the second bit is it's slow. It might take 10 seconds to generate this video. And what we've spent the last 18 months doing at Rota, you know, our incredible research team has has built technology that both enables a video model that's much faster, so it runs in real time, but not, you know, every couple seconds real time, like literally multiple times per second generating a video real time.

And then secondly, the physical accuracy is to a higher level than, you know, any video model out there. So, we built a model that's both very physically accurate and much, much faster. And once you've built this model, you can actually use it to control the robot to do tasks. So, that's a little bit of a, you know, technical deep dive. Going back to the returns processing, you know, a task that you mentioned. If you were going to try to solve that problem with robot data, I I don't know how long it would take. It would take an incredibly long time because you need to train every configuration of the package.

Hello.

what's your what's your kitty's name?

Emmanuel.

Emmanuel. I

Emmanuel cat.

I I love that Emmanuel has an interest in in deep tech. but so you have to

essentially, you know, for this task, you have to Hi, Emmanuel.

he requires attention. He's like, "Okay, can a robot pet a cat?"

Yes.

This is this is maybe the

He's scratchy.

Are you a human test?

Okay.

You're stuck with me.

But but for

I I I like him, too. so, for this specific task, you know, if you were going to try to train every configuration of where the package is or what clothing item there is, you would have to collect, I don't know, tens of thousands, maybe hundreds of thousands of hours of robot data. So, it's just completely not feasible. But a video model can absolutely generate, especially, you know, an accurate video model can generate a video of opening a package, turning it around, taking out clothing, folding them, inspecting them. So, we were able to do that task with, I don't know for sure in the exact quantity, but, you know, call it no more than a couple tens of hours of robot data. So, essentially, very, very data efficient.

That's few-shot learning.

Exactly. Yeah, absolutely.

That is wild. So, effectively, you're giving robots imagination.

That's that's a beautiful way to put it, yeah.

so, we can officially call all other robots unimaginative.

I love it. we should have We should have looped you in on on the marketing copy.

thank you. The other question I had So, the robot like during this task, the robot goes through a lot of states that look exactly the same.

Mhm.

So, like I'm inspecting two sides of the t-shirt.

Yeah.

Does it mean like I've already inspected the other side? Does it mean that I've already inspected at all and I'm supposed to fold it? If it's folded, like did I inspect it? Do I need to unfold it, fold it back? Do I need to seal it? How do you solve the like logic? Is it some somewhat like of an agentic approach?

I think one thing that's very exciting about our model is compared to, you know, almost any other robot policy out there is that we have a very long context. So, what that means, you know, to to kind of break it down, in the same way language models, you know, they have contexts like how much do they remember of what you previously said. And when ChatGPT first came out, it was very little. And now it like remembers your entire chat history. So, that that context really allows you to, you know, reason and and and execute long horizon tasks. So, in the context of robot models, this problem that you're talking about is very real. It's If you have a short context, then if the t-shirt is laying this way, you don't know if it was just put there or if you already inspected it or if you, you know, where you are in the process, especially if it looks the same.

And most VLAs have a context length of maybe one to three frames, which is a tenth of a second, which of course is, you know, not enough to be able to tell the history. because we're you know, have a very different model architecture, we actually have a much longer context. It's in the hundreds of frames. So, we our robot actually has, you know, a memory of the task as it's doing it. And it's using this task to It's using this memory to, you know, reason about the task. and this is a really key distinction because when you don't have context, you usually have to fall back on things like language. So, you have to deconstruct a task. Like if you're doing returns processing, you have to build it into maybe 20 modules. Like first you do this, then this, then this, and you have either some deterministic or language model like walking you through those modules as you execute.

Whereas we saw the task through long context, which means when we collect, you know, that limited robot data, we collect data on the whole task, and that's enough for it to remember the whole task. We don't need like a backbone that's telling it step-by-step how to do the task.

Similar to our house routine. Like now I'm deconstructing a lot of tasks like little ones for you.

Oh, yeah.

Maybe should like

I have a very short context window. Every time we try to increase it, like needle in haystack task is not being passed. I

Well, Agentyc AI's on the rise, so you're you're you're in good shape.

I'm yes, I will I will use it a lot.

I can we ask maybe it's appropriate to ask here. So, why did you choose stealth strategy?

Yeah, so this is really Jagdeep's kind of you know, wisdom coming into the equation. I think I would have defaulted to sharing things earlier on, but I've really liked, you know, how we how we chose to to build Roda. the way Jagdeep puts it is there's three reasons to have a public profile. It's to you know, find investors, find customers, or find team members. building a business isn't about, you know, having your having your name on the website or or or seeing your name in the press. It's about, you know, creating value. And those are the three parties that you need to create value. Our thinking was from the start if we need to come out of stealth to you know, build more relationships with one of those three key audiences, then we will. And if we don't need to come out of stealth, then then we won't.

So, we almost flipped it around and said it Roda was always about why come out of stealth. It wasn't about, you know, why why stay in stealth. And for the first year and a half, you know, through our network and through our our researchers' reputation in the science community and Jagdeep's reputation in the business community, we were able to, you know, both recruit incredible teammates, get really high-value add investors on board. And yeah, and and actually both, you know, our team's network and our investor network has been really helpful in in starting those initial customer conversations. But now we feel like we're at a point where the team is getting to a size and technology is getting to a maturity where we want to start, you know, accelerating the flywheel and we want to start building our team faster and we want to start, you know, building relationships with the customer and investor community faster.

So because of that, we chose to come out of stealth.

We actually started like discussing that how it was to answer questions about your job.

You know, so we had we even when we were in stealth, we we we all got in a room and said, "Okay, what are we going to what are we going to tell people?" Cuz people want to know what you're working on. You know, they're worried what's going on if you don't if you don't talk about your job for too long. So we agreed that we would we were okay saying we're at the intersection of robotics and AI. So I've just been telling people that I'm working on building robots and honestly it's a pretty it's pretty it's a pretty fun conversation starter.

Do you know how we call project inside our company?

I think I've I think I've heard it's Asimov, right?

Yeah, Asimov.

That's a great name. I love it.

Well, like we we had an incident with another portfolio company where like like a screenshot of our internal system got got to the founders and they're like, "How many people can see that?" And we're like, "We should probably for like stealth mode companies we should probably come up with project names." And that's why like when we invested the first time we're like, "Okay, this is going to be project Asimov because of the, you know, our means robot and laws of robotics, which I think having imagination can solve at least couple of laws Asimov's laws of robotics.

Mhm.

You can imagine that if you like hit a person in the head, the head goes boom, and that is a bad outcome. Should not be translated into action.

Well, we're at we have we have lots of systems in place to make sure that whenever happens. So, that's that's even pre-imagination, but but for sure. Yeah, that's

Do you have that in like any heads exploding in the training set?

We tried We tried to avoid Yeah, any any kind of violence in in in the in the even in the internet scale video pre-training. So, we have some some good filtering mechanisms in there.

How did you collect so much data off manipulating objects high accuracy? How did How did you like Did you have an external force of people annotating that? Did you Did you do all internally?

Yeah. Oh, this is a This is a great question because Alex Bergman, he's our Chief Data Officer. He's He's been at the team pretty much from the very beginning. and he he single-handedly runs our data collection pipeline. So, probably one of the most, you know, talented and efficient people I've ever met. I think yeah, he's he's now downloaded hundreds of millions of videos from, you know, hundreds of resources. and he has built the entire pipeline of downloading, processing, evaluating, scoring, you know, preparing, chunking for the AI video model essentially single-handedly. So, just just an incredible you know, both researcher and engineer. and I think what what from what we've heard at other companies can take, you know, 20 or 30 people, Alex, with the help of some AI tools, is just is just a machine.

It's a thousand X data scientist.

Absolutely.

So, thinking about the competitive landscape, there's there some big guys like Figure raised a lot of money. Tesla, there's Nvidia with their recruit. and like physical intelligence and a bunch of other guys. None of them use your approach.

Mhm.

Do you think once you know, like and none of their stuff actually works. Like it does some good things, but nothing's near human performance. Now since you've announced this quite widely and like all of my X feed is filled with Rodan.

Mhm.

Do you think they're going to start pivoting towards what you're building? Are they going to be like new entrants? Would Google go after the same field since they have like YouTube?

Yeah, I think that's that's a great question. this is definitely a very hot space and I think the reason is that a lot of folks recognize that this might be the biggest, you know, market ever. when you think of how much how many things are being produced and moved or, you know, how much how much manual labor there is, the the opportunity is just really enormous. and I I personally think that the key factors that go into how quickly GDP can grow is how much, you know, labor availability you have and how much capital availability you have. And that formula has been the same for the entirety of history. It's about, you know, 2/3 labor, 1/3 capital. And once you can, you know, start to create labor, which is what I think this generation of robotics is going to do, you know, we can see a exponential growth in global GDP and the economy and consequently in, you know, human prosperity.

So I think I I genuinely think this is the next industrial revolution. and in the same way the first one you know, bumped long-term GDP growth from 0.2% to 2. I think this one's going to bump it from maybe 2 to 10. The world's going to be an incredibly interesting and bright place as this technology matures.

It's actually super interesting cuz it might keep compounding. Like because the it's only the first layer of robots for human needs. Right? Then there's going to be robots for robot needs. Then there's going to be extraterrestrial robots because like why send humans if we can send, you know, like to build a Dyson sphere or to to send starships to another star system. We could just send robots with microbes.

Yeah, I think I think this is absolutely right. I think people you know, we focus we're we're starting the company by focusing on manufacturing and logistics. And you know, some folks will ask why manufacturing logistics, why not home robots or, you know, hotel robots or something. But I think the reality is if you can make a step change impact in how things are made and moved, that's going to permeate into every other part of human society. And you can you can look at that, you know, once again going back to the Industrial Revolution. The Industrial Revolution was about the steam engine, which has nothing to do with you know, maybe necessarily things like healthcare or public safety or life expectancy or you know, breakthroughs in other, you know, unrelated fields of science. But how it played out was when the Industrial Revolution happened, you know, more more money was invested in these fields, things were produced faster, moved faster.

and that that GDP growth tailwind just had an enormous impact on human civilization. I think the reason that humans have the life expectancy they do and enjoy, you know, the comforts that we do and the quality of healthcare and the quality of, you know, culture and art, a lot of that comes from the resources and efficiency that was unlocked by the Industrial Revolution. So, I think that, you know, manufacturing and logistics might not be where the average person spends a lot of their a lot of their time thinking about, but in- inventing a big technological change in those fields is actually going to permeate through every every part of of humanity, and I think create a really exciting future. And you're right. I I think that the timeline to s- you know, space have have, you know, making real gains in in productivity in space is going to be vastly accelerated by robotics.

You know what? That was your dream to travel to the moon. Maybe it would be not you, but the robots.

Yeah, the moon's close. Like I don't want to go to Mars. I like I I want robots to go to Mars and to build a Dyson sphere and like to create infrastructure on the moon so I can go there for a vacation.

Yeah, data data centers on Mars. It's a it's a it's a good business.

We were just joking that since there's no atmosphere on the moon, you can you can like you you can move energy without any loss and then you can maybe use the fun part of low gravity. You can build like a large tent with oxygen in it like with atmosphere and you can give people wings and let them flap them and fly around.

Sounds like a good time.

Yeah, I mean like that would be my I hope my

And we can move money Monday.

Yeah, or that that's going to be my assisted assisted living facility like for a retirement house. Like I'm going to just fly around like an angel.

You know, at least make sure that you will be on the right side of the moon so you I can look at you and wave you.

Okay.

Sounds like you're not joining for the trip.

So also thank you one of the videos you you posted is is going to help me fall asleep because it's a 90-minute uninterrupted autonomous box decanting.

Mhm.

And then another one is a 2 hours 40 minutes of container breakdown.

Yep. I was I was really hoping you would you would watch the whole thing end to end.

It is impressive. That's like I just scrolled through it for now, but I will watch it. Like and Anyone else in the market can do that.

well, so there's I would I would say, you know, it's a matrix of task complexity and duration and I think that I haven't seen a task of that duration of that complexity shared by anyone. There have definitely been videos that were long duration where the task is very repeatable. Like maybe picking something up here.

slow.

Yeah, like maybe like like folding like the the same towel over and over and over again or like moving a parcel from here to here. I've seen videos like that for, you know, 20 minutes, an hour, maybe even longer, but yeah, something like opening and unpacking a box with paper, with plastic, with you know, heavy materials and running that for an hour. I I haven't seen anything like that yet. And these videos they're really important for the robotics community because if you're if you're doing a demo of a task one time, you can have a very bad policy and do a very impressive task one time. So it just at this point it it's just not helpful when when when folks post videos like that. I think long duration videos need to be the norm. I understand that they're not the best Twitter material, but that's okay.

People can speed them up, you know, not everyone has to watch them. A couple experts can watch them and say you know, I've I've watched this, it's real. But I feel like every demo should have a video like that.

People are weird. You'll be surprised what they watch for like hours and hours and hours and like, you know, you put this to a good music and maybe somebody wants to smoke some pot and just like watch the robot like do do its thing for some hours.

Well, at least we have to do something when we are relaxing on the moon.

Oh, yeah. Like well, we'll just do that. But if So right now you're selling like a robot as a service if I understand it correctly. So if I'm like a potential customer, let's say I like I watch one of the demos and I was like that's exactly what I need or something similar. How do I how do I buy a robot? Like do you have a pricing strategy?

Yeah, so so our website has a you know, request consultation button. We're One of the reasons we launched is we really wanted to you know, build more partnerships with more customers. And the way we think about it, if you solve you know, 90% of everyone's problem, you don't have a business. You have to you have to solve 100% of a problem for some for some people. And we're obsessed at Roda about learning 100% of the problem so we can then solve 100% of the problem. And the only way to do that is to talk to customers. And that's why we really, you know, have valued even before the robot made its first move, talking to customers, learning about their use cases, and understanding, you know, what what's a use case that doesn't just make up for a cool project in the corner, but really moves the needle for their business.

So, for folks that are running, you know, warehouses or factories or just involved in those processes, whether it's, you know, commercial, industrial, warehousing, or even other, you know, industries we might not be targeting right now, learning about the use cases is really valuable for us. And, you know, we'd absolutely love to partner with them. So, I'd I'd highly encourage them to to reach out. To reach out, yeah.

So, I understand that right right now about just 1% of industrial operations have been automated by robots that, you know, use fixtures and and fixed programming. So, you're addressing the 99% but that's also has a power law like a Pareto distribution. There's a long tail of stuff that you won't probably be interested in doing. Would you be open to selling like software as a solution for like longer tail integrators? Maybe maybe companies that were like, "Oh, I'm I'm going to do a robot for specifically doing milkshakes."

Yeah, yeah. Essentially, how we think about it is right now, I guess, you know, you can always think about building a company vertically or horizontally. You know, the benefits of vertically is you solve the problem end to end. You guarantee that user experience. The benefits of horizontally is, you know, you focus on one component and you can have access through other players. I think at this stage, a lot of the, you know, robotic stack isn't quite there to where, you know, if we if if if we posted our model online tomorrow and said, "Hey everyone, use this." I I don't think a lot of the other components are are there in enough volume and reliability to produce that

That's where we're at right now.

so, for now, we're really taking, you know, an end-to-end approach, where we want to make sure that the first adopters of this technology have really good experiences, that they get technology in their factories and warehouses that solves their problems, is reliable, is safe, is certified. so, we're focusing on on the full stack right now. That said, you know, as as the space matures, which is happening very quickly, we very much intend to go into more of a plat- platform model, where our AI model can be used both by, you know, other hardware manufacturers and providers, and also, you know, folks that have a specific industry that want to they want to go after and build, you know, a vertical integrator or deployment type of company. So,

Mhm.

we we definitely want to enable, you know, the future of robotics and automation to come sooner. And to do that, we have to keep you know, take care of two things. One is make sure that we build real solutions that add credibility to the space instead of just hype. And the other is that, when the time is right, we, you know, give entrepreneurs and builders the tools to go and create lots of value. And you know, we we firmly believe that, while we try to get a lot of the world's best brains into Roda, the majority of the world's smart people will always, you know, not be working at Roda. So, we'd love to, you know, build a system that enables them to to create value and and help, you know, move this technology forward.

I think that's a really good way to look at it and also promotes pretty good culture, which is a good segue to my next question. So, last couple of weeks we've been hearing a lot of drama around Anthropic, D O W, and Open AI undercutting them and just striking the deal for, you know, couple of execs quitting from Open AI over this use of AI agents for mass surveillance. That is scary, but like as somebody who was born in Soviet Union, I have no idea, no delusion of privacy. Robots, however, are scary. What do you guys think about like defense tech? Would you explore that? Is is it strategic cuz everybody speaks about, you know, China, they're a huge bet on robotics. How do you like as a company see this?

Yeah, we we we really see ourselves focusing on kind of manufacturing and logistics for now. We definitely want to make sure that our technology is used to make lives of people better and and not cause harm. So, internally we have, you know, pretty strong beliefs on on on not causing harm with our technology. And this is something that we got a lot of alignment on within the team very early on. So, we don't see ourselves, you know, we we want to avoid implementations, deployments, technologies that can directly cause harm to people.

That is really cool. That that goes very well with our belief system, our our ethics code. You mentioned you have like guardrails against actually causing harm to humans. Can you can you dive a little bit into how that actually works? Is it like the Is it in the language model? Is it in the like how do how does the robot know not to harm?

Yeah, so so there's lots of there's lots of layers to it and you have to have lots of layers. but essentially, you know, without going into all of the details, the where you start is around the data. You know, we have to ingest a lot of data to train our model, both, you know, a lot of video data and a little bit of robot data. we want to make sure that there's no data that demonstrates, you know, harmful scenarios or that, you know, could could introduce harmful kind of scenarios into the robot's training. So, that's that's stage one. You want to make sure the data coming in is is clean from potential things that may generate harm. The second thing is you want to make sure that the video that's being generated is being screened for any kind of harmful, you know, behaviors or controls.

So, you have to have this data input data output for things like language instruction, you also have to screen the language instructions. And then, even once you've built, you know, a lot of safety into the AI model with, you know, multiple layers of kind of monitoring both during training and also during inference, you also have to make the hardware safe. The only way to make the system safe is to have redundancy for safety in the hardware.

Mhm.

And we're we're keeping a lot of our hardware efforts under wraps, but some things that we are sharing is that our hardware design is is built from the ground up for safety. And there's a couple architectural choices we made that are very different from, you know, all of the all of the humanoids out there. One is that we we don't want to use legs. We either have, you know, mobile bases or stationary systems. And this is because every customer I've spoken to, they want an emergency stop on a system. Everything that moves in a factory has a red button on it that you can push on to stop the machine. And you know, I I I I I just don't believe that humanoids on legs, which are a much higher variability system, are going to be excluded from that, you know, very reasonable requirement. So, our system is on a wheelbase, which means there's a lot of buttons on it that you can push and and it'll freeze.

and it'll freeze in place. If you're on legs, you might tip over. If your battery dies, you might tip over and maybe harm a person. But, if you're on if you're on a

Have everything.

Yeah, if you're on a wheelbase, you can freeze in place, which is a lot safer. The other thing is we built torque sensing into all of our actuators and brakes. And this is really important because we can detect when a human's around the system. And if a human is around the system, then you can limit the upper amount of torque that the robot can exert. So, physically, the robot can't harm a person. So, you kind of we built it from is the AI model training safe on inputs? Is the AI model safe on outputs? Are the instructions it can receive and training it can receive limited on harmful inputs? And even after you make the AI model safe, you then want to have redundancy on the hardware side to ensure that even if the AI model gives, you know, harmful instructions, the hardware is on low-level controls is is just blocked from executing them.

That is a really cool approach. that was I I think a lot of work.

Well, and you know, I I I want to be transparent where a lot of these systems, you know, we're actively designing, continuing to iterate on. for now, you know, we're we're very cautious about how the robots are placed and how they're set up. but as we continue to develop these systems, you know, we're we're very focused on safety and ensuring that these robots can work next to humans without those humans feeling a danger.

And that will probably put some limitations on how fast you can deploy, how many systems you want to have live like in by end of this year.

I don't know if that's a number where we're sharing publicly yet, but I I would say our goal for this year is to have a small number of robots within a, you know, slightly larger number of companies and and larger number of use cases. So, our hope is to deploy, you know, with x amount of customers and have, you know, n amount of use cases in each of those customers. Maybe just a couple robots in those use cases. And really, the reason is we don't want to start by trying to, you know, scale to very large numbers. We want to start by getting into use cases, learning about these environments, hardening our system so that it works in a variety of of use cases and everything, you know, from the user experience to the hardware to the AI model performing in all these use cases.

And then once we feel that, you know, we we deliver lots of customer value, then then roll out into scale.

And then yeah, every customer can can grow with other use cases within their facilities and just like more of the same use case.

Yeah, maybe we can just discuss right now the team, how many people people were initially, how many people you have now, maybe deconstruct a little bit like what what they're doing, the functions.

Yeah, so when when we started, it was Jagdeep, Eric, myself and Changan and Gordon who's a co-founder and scientific advisor. He's he's professor at Stanford. So we started with the five of us and we were all, you know, kind of focused on on different things, but we knew that the key hypotheses we need to prove out is, you know, A, can you generate videos that are fast enough and physically accurate enough to control a robot? B, can you you know, use those videos to control a robot in real time? And C, does it scale with compute and video data as opposed to scale with robot data, which we think, you know, just doesn't doesn't practically work in the market. So we really focused on proving out those hypotheses and building, you know, a pre-trained video model takes time and resources. So we we spent a lot of time building and developing and iterating on that.

We spent a lot of time talking to customers to understand what are the use cases that move the needle and we kind of did these two efforts at the same time. Now the team is much bigger. We're at over, you know, as I mentioned over 50 people and the I guess the rough breakdown is we have a AI research team. This is the team that does, you know, fundamental science in the video model, inverse dynamics model, essentially everything that goes into the the AI and and robotic and

technical stuff members.

Exactly, yeah. But but even, you know, beyond that there's there's a research team, there's the robot software team, which is everything from robot real-time controls, low-level controls to, you know, UX and kind of how you how you think about controlling the robot, giving it instructions.

Is it by the way, is this job title specifically for like cuz everybody gives calling AI researchers and engineers just technical stuff. Elon calls them all engineers. Is it anti-poaching?

No, actually I don't think that's what it is. I I This is something that I I really respect about, you know, our our team members and researchers. They don't necessarily care about, you know, they don't like titles and hierarchy. They like to work collaboratively. They like an environment where everyone's in the team and everyone's contributions are valued and you make sure that you listen to the best ideas in the room as opposed to default to, you know, who's more senior or junior in the room. So, I think I actually think that the naming conventions of of, you know, referring to folks as as members of the technical staff is is more because of, you know, the culture that researchers like and how they like to, you know, build and collaborate together as opposed to, you know, trying to trying to keep them from from leaving.

From Zuckerberg

approaching everyone.

So, that part is technical staff. How is your GTM built right now? Your product management? Do you do you have like a department or

Yeah. So, there's on the technical side we haven't covered everyone yet. So, you know, we have there's the AI research team, there's the software team and there's the hardware team. So, those are those are the three components of the technical team. At Rodo, we try to be, you know, very engineering scientist first. So, we've we've been very intentional about scaling the engineering teams before, you know, scaling the business side of things. But, on the business side, you know, we're starting to build the team and and have, you know, infrastructure in place to both do things like go to market, you know, efficiently manage our finances and and and keep our our burn low and steady. And then also on the product side, ensure that we have the capacity to talk to a lot of customers, understand their needs and and communicate these needs into how our product is being built and designed.

So, yeah.

Do you have any open roles, like vacancies right now? So, I

Yeah, absolutely. There There's actually there's there's many open roles. I I want to make sure this video is always up-to-date. So, I'll just say that, you know, there's a careers page on our website that lists all the open roles, but we're we're we're always looking for smart, enthusiastic, mission-driven people to join that care about, you know, solving this problem. And we think it's we think this is probably one of the most exciting problems to work on right now. the way we think about it is there's only going to be, you know, the the the moment in time where robots can do everything physically that humans can do is going to happen, you know, not once in a decade, not once in our lifetimes, but like once in human history. And I think that the world is going to be before and after, you know, that moment in time.

so, it's an incredibly impactful and incredibly technically interesting problem to work on. So, if that if that resonates with folks, I I would highly encourage them to apply. We try to list our roles, but we're also hiring opportunistically. So, if there's a role that doesn't quite fit on the website, I think sending an email anyway and and sharing about your background is something that you know, we really encourage. We And we have hired people that way.

Are you all in-person in Palo Alto?

Yeah, we have a very in-person culture. So,

even for robots, robots should be

All the robots are in-person, too. I think if you're if you're building hardware, you have to you have to be there. you you can't you can't do it over Zoom. So, we have a very in-person culture, you know, 5 days a week, and I I really

is working from home.

well, actually,

not not not really. Yeah. So, we we all we all work from the office every day. And I think that, you know, A, it helps us get more done and communicate better, but B, it just creates an awesome culture. Like a lot of my really good friends, you know, I've I've I've made at Roda, and it's it's it's awesome to come into the office every day and work with like-minded people. I think it's something that you know, during during Zoom everyone kind of went away from working for the office, and there are benefits from remote work, but there's also a lot of great benefits of especially when you're building something mission-driven and exciting just being in the room with the people you're doing it with. so, if folks are are tired of working on Zoom and want to want to get back into the arena, you know, I think we have a great culture for that kind of thing.

working with robots in person as well. We did discuss it like they could like two trends, either you're working everyone in person from the office or either like everyone is remote and there's nothing in between because something in between is just not working.

Hybridization is weird. so, let's talk a little bit about fundraising. Okay. so, you announced 450 million. Bloomberg said it's 1.7 billion valuation.

You're free to comment that.

Yeah, and

and then I assume that is like two of the rounds even three of the rounds combined. like everything you raised before like before you came out of stealth including this like newer bigger round. How did it like how's the fundraising environment for robotics right now? Was it hard? Was it like always oversubscribed?

yeah, so there's there's definitely been a lot of a lot of demand for for the fundraising and we've been kind of fortunate to be in a spot where we we've had more demand than than room in the round, which has enabled us to, you know, choose really value-add investors like yourselves.

Thank you.

So, yeah, there's a lot of demand. I think I think investors and I don't want to speak for them, but I'd actually love to hear your perspective on this, too, but I think there's two kind of conflicting things happening at the same time. On one hand, I think there's now a consensus that this technology is going to create a huge market with a lot of opportunity and that this opportunity the timing is right, that it's not, you know, now is the time to be investing in robotics. So, there's that, which is a tailwind, and then the other side of it is there's a lot of showmanship and hype in robotics and it can be difficult.

Yeah, it can be difficult to tell, you know, the credible real stories from from the hype that's not going to stick around that's not going to, you know, create value. So I think that when speaking to investors this this is a trend that everyone, you know, we've seen battle with and to their credit I think a lot of a lot of the best investors have identified that this this situation that on one hand it's a huge opportunity on the other hand there's a lot of you know, bubbly companies that are probably going to go to zero. And I think that they've used that as a reason to get really technical and I've seen, you know, I've interacted with a lot of investors who have read a lot of the, you know, scientific papers in the space and who have spoken with researchers in the space and they know that you can't, you know, you can't just rely on reference checks.

You have to you have to dig deeper into the technology understand it for yourself and and choose a play that's really credible. And I think that to their credit a lot of the investors that have joined us have taken the time to understand the technology have, you know, taken the time to understand how it's differentiated and and why it's different from the typical VLA approach. And then once once they understood that, you know, they already have exceptional understanding of, you know, markets and and growth and scaling. So to them the part that we need much less robot data is the trigger point to where they they know that this is going to be a different kind of business. It's not going to be stuck in the lab because if you only need dozens of hours to deploy to a use case, that's a very different kind of, you know, go to market opportunity than if you need to spend a year preparing for every use case.

When we spent those for I think first two hours in your lab before we invested first time, it it was actually painful for me because we worked on very similar approaches back in 2017 for our AI cameras.

Mhm.

And we had so like so many ideas we had that but we we were too early. We didn't have your like researchers that are way smarter than us.

And also the resources that you apply to it were like so I I was just like looking at all that and it's like okay. I knew it's going to work in 2017. I just didn't know when and who's going to make it. Now I do and so for us it was a no-brainer and I think for that exactly reason I only know two types of investors that either don't like had hadn't known about you or the ones who invested.

Mhm.

I really never met an investor who looked at Roda and said like yeah that's great but we would pass.

Well I I I appreciate your support and yeah you you your your experience in particular you know enabled you to really understand the vision systems and AI and and how these things would play together to solve our product. So it was great to be speaking the same language with you even for our first meetings.

Okay so another question that I had is if you were like an a founder who's deeply interested in robotics

Mhm.

without ability to raise 450 billion from scratch in stealth mode which is I'd say rather rare. Let's say you can you can raise 3 mil seed. What would you build?

Mhm. I think at 3 million there's there's two strategies you can take. I think one strategy

Take for a million and run.

That's maybe with that one it's three but I I would advise against that one. I I think there's there's two directions you can go in. One is you can focus on one of the specific sub problems and solve it really well. So you know some examples of this is you know you you can look at the full robot stack and say, what are the parts that are missing? And you you can really zoom in. Like, some examples are I think as of today, no one has a good hand. We've seen We've seen a lot of, you know, cool hands that can do cool motions, but there's there's a chasm there because the hands that are very dextrous are extremely unreliable. Like, I I would be surprised if they last a week in a factory. on the flip side, the hands that are reliable are really mostly grippers and and have pretty limited dexterity.

You know, working on hands is an interesting one. Or you can even go deeper and say, well, what about what about the touch sensing on the fingers? You know, what what are the sensors on the hands? Or what are the actuators you need in the fingers? So, I think if someone's very deeply technical and they want to solve an engineering problem and they have, you know, 3 million to go make an impact, I would I would choose a component somewhere on the robot stack, somewhere from cameras to actuators to materials to sensors to you know, maybe network processing solutions to cloud inference infrastructure solutions to I think I think data collection is is riskier. If if folks want to build a data collection business, they should reach out to us and we'll share what kind of data, you know, we think is needed for robotics, but with with data at this stage of an AI, you know, all the folks that have spent time building data pipelines for VLAs, I I think, you know, took a took a risk because there's both the risk of will VLA companies buy this?

There's the additional risk of will VLAs even work in the first place?

Cloud-based systems, they've been like virtual environments. People have been trying lots of stuff that ultimately will probably go in a trash bin.

I agree. I think the the simulation engines and the data, you know, collection, those are riskier cuz there's like two stages of risk to it of, you know, will people buy this and does the approach I'm You have to bet on an approach that you're building towards. But things like, you know, hardware components or networking software components, those those could work, you know, that's a good trend to back.

that would probably not become core competence for people who are raising a lot of money and solving the the full vertical stack.

Yeah, and there's, you know, ultimately the the play might be to get acquired by one of those companies if they if they're going to verticalize or to build a business where you sell to, you know, if if there are multiple companies that can build robots that work then then you can sell to all of them. So that's one direction and I would I would recommend that path for folks that really want to get their hands deep with a, you know, engineering problem. and then on the other hand, I think now is actually a good time to start building kind of vertical deployment businesses for robotics. So choose a specific industry, choose a set of use cases, understand these use cases really well, build, you know, understand what are their certification requirements, what are their safety requirements, what are their use case requirements, KPIs, and start experimenting, you know, with the models and the hardware you can find to start to solve those use cases.

Candidly, today the models and the hardware that are open source will probably not enable you to do that successfully to production standards, but I think that is around the corner. You know, you working with Rota when when we open our platform or, you know, buying hardware off the shelf. I think that's that's going to reach an inflection point maybe 6 to 12 months from now. So I think it already makes sense to start building companies that are very industry specific, industry deep, vertically integrated. So I and that might be a good business to build for someone who's excited about go-to-market and about deploying robots into the real world and wants to have those customer relationships and see their robots, you know, move and and build things. So I think with 3 million in the bank, those are those are the two categories I would choose from in robotics.

Either, you know, vertical deployment if you want to focus on the go-to-market side, there's still going to be a lot of technical challenges to solve there. So plenty of, you know, engineering meat to dig into, but more go-to-marketing heavy. And then the other is building, you know, components, parts. I think I think with 3 million trying to build a full stack AI model to robot is is probably

What's the entry ticket you think's like?

Oof.

Given that they would have to catch up now.

Yeah, I think Well, I guess it depends how you define full stack. Like in some ways you you can raise a lot of money like like, you know, like some of the software companies out there like Skilled Physical Intelligence. They they raised a lot of capital and yet they're still only focusing on part of the problem, which is AI model. so in some ways, you know, if you're attacking the foundation model side of it, you can't raise too much money. Like it's it's it's going to be an expensive problem to really prosecute. On the hardware side, there's there's also this really big gap between making a robot that looks good in a lab for a demo and a robot that's hardened enough to work in a factory for 3 years. And if you want it to work hard in a factory for 3 years, you need to, you know, you're going to have to go through multiple iterations.

So, it kind of depends on your like funding strategy. If you're trying to fund end to end from initial robot design through to large volume manufacturing, you need hundreds of millions of dollars. But if you're going to do it by first building, you know, a design of a robot or a differentiated actuator and then raising another round based on that to build the whole robot then maybe to go into new product introduction manufacturing. Like you can you can do it kind of stage by stage. With the foundation models, it's harder to do it stage by stage.

Are you already open to like acquisition opportunities that, you know, like looking at smaller startups maybe pick up some team, pick up a piece of technology or a value chain that's going to be part of the road map?

We have a We have kind of, you know, two general philosophies at Rodo, which is we even though we've raised, you know, capital from a great group of investors, we still have, you know, limited resources, right? So, we want to spend our our limited resources on I'd say we don't want to spend our limited resources on problems that have already been solved. So, if someone's already solved a problem and they've solved it well, we'd rather just partner with them or buy it. On the flip side, if if the problem isn't solved, we're never afraid to go after it. You know, if there's a very specific technology in the cameras or the sensors that we just don't see a good solution for, we're not afraid to build it. and and and build out the team to build it. So, I'd say we're always opportunistically, you know, looking to solve problems, build differentiation, and do that both through, you know, hiring the best people and and sometimes the most efficient way to hire is to kind of bring on a team.

so, these are things we're open to. but yeah, I I would say reach out and and we'd love to hear more about

if if you are working on one of these ideas, reach out to us or to Andrew. either we would invest or they would acquire you. if we're crypto founders

Yeah, yeah, yeah. First, we need to invest and then that that would be our course cuz we need to get the bigger share of Rota. and I also have a few ideas for crypto founders out there. So, like first of all, you can as usual work on a distributed robotic system where basically different parts and screws are distributed across the room like and they somehow work together.

there's there's a like a robot payment solution, of course. So, like robots need to be able to afford modding parts like a marketplace of NFT robot parts that you can like modify and look awesome, right? They they would need that. I think there's going to be a large market for that.

Yeah, I I I'll I'll

Not an investment advice, but

I'll keep that off my list for now.

I think I have the last question, but I I'm very I I I I couldn't not ask it since your experience in Crimson. And we've got lots of questions about kids and education and what to do with colleges, especially right now in this constantly changing environment. What would you suggest to high schoolers who are applying right now, who are interested for example in computer science?

And also, does what you do at Roda eliminate the need for trade schools?

I think for a while we're going to be, you know, we're going to be creating more demand for, you know, interesting manufacturing technician jobs. I don't think we're going to be ruling out trade schools for a while. I think there's, you know, we're we're targeting kind of the dull, dangerous, and dirty jobs that people don't want to do, that companies struggle to find talent for. And even within that category, you know, the the the triple D as they call it, there's there's millions of robots to deploy. So, I think for folks that are, you know, high skill and if anything, you know, if if if you're in the US, I think robotics is going to create more manufacturing jobs. And what And what I mean And I mean And I mean that sincerely because most things in the US you can't manufacture. You can't make shoes in the US.

You can't make t-shirts in the US. It's just from a global competitive framing, it's nearly impossible. It's It's, you know, it's frankly difficult to make cars in the US. We're able to because of a lot of protectionism from the government, but in in a open market, it's it's very difficult to manufacture things. But with robotics, I think we can actually manufacture anything in the US. And I think that's going to create more jobs in manufacturing logistics, not less over time. As as more and more things are are getting built and made closer to where they're consumed, which is also, you know, really good for for the climate. I wouldn't rule out trade schools. We have We have a long ways to go to, like, you know, a robot plumber that can walk into a home and fix a pipe. So, I I I I wouldn't be too too worried.

But I think yeah, the world is changing so fast right now that it's difficult to see It's interesting. It felt 10 years ago like predicting the next 20, 30 years was easy. Like there were a lot of experts that had strong opinions about, you know, here's the trajectory of world peace that we're on, the trajectory of technology, and it feels like every every 20-30 year predict prediction ends up being incredibly wrong. So, I don't I don't want to try to predict what the future is going to look like.

Maybe pipes will just regrow themselves.

Who knows, yeah. there's there's some interesting things going on in material science. but I would just say you know, I think I think finding the intersection of what you're passionate about and what brings value to other people is should really be the focus. I think that

What's the What's the Japanese name for the three circles?

Yeah, it's what you're good at, what you enjoy doing, and what other people need. I think I think those are the three. The other advice, you know, I heard and someone shared this at business school was that if you ask successful people do they like what they do, they'll say yeah, they're very passionate about it. But often they weren't passionate when they started it. They became passionate about it because it's exciting to do something that you're good at. so, I think that when when discovering, you know, what you're passionate about, you shouldn't just do it from the perspective of, you know, kind of sitting on a couch and thinking, what do I want to do right now? How can I make that a career? You should do it by applying yourself, trying different things, trying to learn, even things that might not appeal to you initially.

Like, you know, try building a financial model, try designing, you know, a simple circuit, try coding up an application. Like, apply yourself first, you know, take the hours of effort that it takes to achieve, you know, a basic level of competence, and then if you see that you're good at it, then you'll probably grow into a passion for it. So, I the general, you know, cliché advice is do something you're passionate about, but I would caveat that with not just a passion of unexplored interest, but a passion that comes from working hard on something and and and building some skills in it.

I love that advice so much because I have so many people around who are just trying to find out what they passionate about. Like what they are actually in this loop like trying to find it, trying to think about like spending a lot of time like with coaches and like therapists and other people. But actually they have to spend more time trying to do it. And

I have never heard a single example, a single anecdote of somebody who has actually found their passion via talking to a coach or a therapist.

You know what? I know the person who who knows what he he he wanted to do since high school. Actually, and when Nick has been asked at school what he wants to do, he said like I want to do

I want to write business plans and strategies and sell them to companies.

I I think I think a coach or therapist can help get someone, you know, in the mental framing for where they're they're ready to go and apply themselves, but finding finding the passion won't come from that conversation. It'll come from applying yourself. So I think, you know, if if if people struggle to, you know, find the motivation to go to go and get started, a coach or therapist can be a great level of support, but the passion won't come from that conversation. It'll come from putting the work in.

Still, the last last question, if high schooler wants to work in robotics, where he or she should apply?

I think it it there's a lot of incredible schools out there. You know, it it it depends on kind of the exact area of robotics, but going to a lot of a lot a lot of universities actually are building robotics programs. You used to have to choose between, you know, computer science or mechanical engineering. Now there's actually more and more robotics majors that'll combine things from, you know, computer vision to computer science to actuators, you know, mechanical engineering, electrical engineering. So you can actually build that full stack in a college program now, which is really exciting. I've heard that the programs at, you know, Stanford, UC Berkeley, Georgia Tech, CMU, University of Washington are really exceptional. I'm sure there's, you know, many others I'm not doing justice by not listing, but yeah, I would I would talk to alumni who went to programs, get their specific advice, and and and see see what programs people have had great experiences with.

I think this was amazing. Very interesting. I would actually enjoy rewatching it.

Yeah.

I think we did we did a great job of

Deconstructing

Deconstructing the the whole thing, which is which is a huge topic, and I I think we we kept a pretty good pace, so people won't have to listen at 2x speed.

And I think we we covered a lot of insight, because like I'm I'm pretty sure that it would be the topic for the next like 5 years for sure.

Oh, yeah. People going to going to keep returning to this.

Awesome. I really enjoyed this conversation, and thank you for your support in helping us build our way.

Thank you so much.

Thank you for letting us be a small part of this journey, because well, this this is very exciting.

Awesome. And I'm so glad that I can now like

Talk about it.

Talk to everyone.

Well, you're you're already coining great great marketing terms for us, so you know,

I I had I had a lot of those in my head. They're just sitting there.

Amazing. Well, thank you, Nick. Thank you, Irina.

Thank you.