B (3:48)

I mean, probably a bad analogy and I don't want to do like any copyright infringement. But you know what it reminds me of? It reminds me of Wall E. Like, if you've ever seen the movie, it just this cute, lovable. It's kind of like Wall E. So Rishi Mini obviously made a huge splash at ces. So tell me a little bit about. Well, one, the insanity of CES and kind of the Rishi announcement. But then kind of go into. I know that there was also another big splash with Nvidia and Jensen kind of coming up and engaging with the demo. And tell me a little bit about that as well.

A (4:21)

So over the past month, we actually been building up on the momentum before CES and trying to get everybody in Nvidia as well to have this sort of really nice presentation on why is having agentic AI capabilities on a robot important? Or why is it not only important, but fun and useful? And for that, that's where actually Nvidia or other projects like Dell come in. Because they provide the technology, the hardware, the computing that we lack within the mini. The MINI will be like the face, the interaction that you get by exploring this. Sometimes abstract AI models that you see on the web, like speech to text scanners, eyelid scanners as well. Something that sometimes when you hear them like that is quite strange, but you put them all together in a robot and oh, okay, now it makes sense because I can interact with this robot that behind it has an LLM. So now I get. Can get information, I can get everything that you do actually with your own LLMs in the phone, but in a much more fun way. Also more interactive, more customer civil as well. You can have kids around you, you can have him take pictures or you can kind of be really quite creative. But for having that, you really need also to have the computing and the AI integration. And that's where Nvidia comes makes sense.

B (5:46)

Okay, let me come to this. I want to show because there's two basically different options, right? We have kind of the Rishi Mini and the Rishi Mini Lite and we'll come back to kind of the compute piece. But you know, someone that is coming to the website that's interested, I mean there's some differences, right? A little bit difference in cost which we don't have to get into. They can see that on the website. But really it comes down to, you know, kind of have a little bit of, I won't say GPU compute, but a little bit of compute on the Arishi Mini, not the light version with the Raspberry PI. And then also the ability to have Wi fi versus Direct connect, like highlight. What are the differences? And then which one would you recommend to someone in like a use case? Right. Do you just go with Rishi Mini straight up? Is there certain people that should buy that over the light version? Kind of explain the difference.

A (6:33)

So actually they are quite similar. The. Well, the thing that takes me the most for importance why will I buy a different version is because the rich mini, I call it wireless is actually because it's like that. It's wireless. It has a battery, so you can kind of take him whenever you want to go and he has one to two hours of battery life. So you don't have to struggle with having it plug all the time or having it. The power went on and then Richie kind of dies because he has no electric power and. But the light version doesn't have it. So you have to always have it plugged. And yes, there is a little bit of computing on the wireless one. It's not that strong if you want to do like super fancy AI integration, but it's enough for really doing simple apps applications locally. So I will say that the wireless version is mostly for people that would like to robot to move around. For example, you take it to your living room to then I know, to the garden or whatever you want. And also for the. Specifically for the AI builders or hobbies that you want to use that little computing power to have a more streamlined. Streamlined applications or streamlined processing.

B (7:47)

Makes sense. Okay, so let's Talk a little bit about that compute piece, right? You kind of mentioned kind of Dell and Nvidia in there. And I know that we're working on some stuff, I mean obviously can't talk about it now because we're not finalized. You know, we'll do another episode whenever we kind of get the details worked out. But for example, this is an ideal and just high level for people that maybe don't know is that anything in robotics, right? You are going to have some sort of compute for that AI inference piece and that can be like an Nvidia Jetson, that can be a 4 or it can be something like a GB 10, right. So you know where we kind of see this is, you know you mentioned AI developers is having kind of a quote unquote kit, right? Or you have like something like a GB10 or some sort of RTX powered device that allows you to run kind of the local models, compute inference, you know, wirelessly off it, where you're able to kind of create and then ultimately deploy onto the Rishi Mini. So tell us a little bit about the, the, the demo where Jensen kind of showed up at gtc. Tell us kind of, you know, I know it's running the Nvidia Spark device but like tell us a little bit about the demo and what Rishi was doing.

A (8:55)

So it showcased really nicely what was one of the main design puzzle, why we create the Rich Mini. I mean our kind of dream was to have this cute fun robot that it was low cost and open source that could kind of help you every day with your tasks either, I don't know, helping you reimagine your new environment. And that's for example in the demo you will see that he takes a picture or about actually an actual drawing that somebody does on the draw room and then he kind of uses AI to reimagine that space alongside the user. Or you can use it for example, take care of your pets. So richmini can kind of check and see, ah, there's your cat, maybe he's doing that. I will kind of alert you and also giving you the liberty to build all those processes. Like for example, if you really have a really specific type of, I know, idea or project and you always thought, hey, I think that this could be useful for other people or just for you, then you can kind of build it because it's also open source. And then that's where the demo of that Nvidia and Jensen presented is quite cool because you can see all the layers of the integrations that you can have from the rich Mini software to the LMS to the different products that are on the market right now within ferrence with the computing and also with the AI or LLMs technology.

B (10:20)

So we're kind of taking a look at my my screen right now and if you know you're following on the podcast the obviously the video version of this you'll be able to see. But if you go to the Rishi Mini website and you get there just go to kind of the apps and it'll take you to like applications right? Which all these can be downloaded and ran in hugging spaces or you know, hugging face spaces or locally on you know if you have compute like a GB10 which I'm planning to do when I get mine kind of this weekend. So maybe pick like let's. We don't have to dive into the you know, the nuts and bolts but let's kind of talk about a few of these at a high level. Like obviously the, the conversational app, obviously Richie has the ability to conversate with but can you kind of give expand upon that a little bit? Is it kind of rag based? Is it more just, you know, designed to just answer basic questions kind of like in Alexa, like what is kind of the point of the conversation app.

A (11:12)

So that was one of the like the really one of the first apps that we created and then we kind of evolve it through time for us for reaching Mini to be successful and to be fun, you had to have fun while talking with him because if not the interaction was going to be weird. I mean you can already code or program something in your PC and it's cool. But if the fun thing about the robotics that you can interact more naturally with it and that for that conversation is central. So at the beginning it was quite, I mean like just putting an LLM inside a machine that talks to you. There were no movement, no kind of feeling to it. But now what we did, there are different kind of stacks that play at the same time. One of them is what we call emotions. So we pre recorded a set of emotions that have a targeted a specific moment for each one. So for example, if Richie is happy he will, he will wiggle his head and his antennas to kind of show to you that he's happy or he's sad or he's scared and whatsoever. So that the idea is that when you are having an voice interaction with it, the model will automatically make those emotions play. And that is decided by the LLM in itself. So that helper helps the robot kind of feel much more like a human does. For example, if you are talking with someone and you just say for example, okay, the robot will do try to do the exact same thing to keep you in the loop, to actually acknowledge that you are talking and to saying, hey, I'm actually here. The next steps for us is for that interaction to be much smoother, to reach much more complete and also much more cultural nuance. So depending on the person that he has in front of him or how you personalize him, then it will adapt. And also the voice levels. Does it talk like a human, does it take like a robot or does it, it talk like somebody else? That's also an open question that we are trying to figure out but at the same time leaving people play with it. Because for some people, I mean you will expect a robot to sound robotic, but some other people, they just prefer him to sound more like a human, like an Alexa or a Siri. So we trying to build the model as open as possible so people can go and kind of modify what they want to modify and at the same time evolve those interaction and communication stuff that I talked just before.

B (13:36)

That's very cool. So I mean if you check it out, I mean there is a ton, right? And I'm assuming, I mean you mentioned kind of open source, right? I know, I see some that are published by you and others like is, is it, is it an open source platform where, you know, this is beyond my skill set. Right. But say I do X to create for Rishi Mini, can I upload it and are people adding, you know, adding to the code base and adding to kind of the add ons or is it kind of controlled by Hugging face and Pollen Robotics?

A (14:06)

The idea is that for example, you had a crazy idea, but you are not so much of a programmer. So you take, I don't know, cursor or copilot and then you create your own draft of the app kind of works and then you publish it over the community and somebody on the community, I don't know, through discord or through your own networking contact, says, hey, that's actually a quite a good app. Let me kind of modified or upgraded. And so he will kind of copy paste the app, create a version two and there you go, you have the version two by yourself and then maybe somebody else will come and they copy paste the version two and you will have the version three. And actually that's what has been happening. There has been a lot of community work and networking about the ideas that some people had but weren't Quite strong in the application of it, but the idea was really nice. So then somebody else came, did a first draft and then revolve from that.

B (15:00)

Okay, I mean that makes total sense, right? Like some of the best ideas come from, you know, outside a company at the end of the day. So the fact that, you know, a good idea can be proposed and then, you know, obviously someone with better technical coding skills, like can actually finalize it and publish it, I think kind of amazing. So let's like jump into some use cases, right? What are like, let's say the. The coolest use case that you've seen that someone's put into practice.

A (15:27)

So there is a lot of it in the educational or kind of how it interacts with kids, higher ed and also kids in general, or for example storytelling. It's really attaching or how you kind of kind of connect to the robot and with the story especially because it's moving, it has the emotions so you can kind of feel the story moving alongside you and also the responsiveness. And then on education as well, but more on the higher education universities because you can kind of follow what you are doing, for example, and ask Richie, hey, I don't understand this equation. And as he has a camera, you can kind of show what you were trying to do or show your text. So he will then use the Internet and the LLMs to kind of browse the answer in a much more kind of useful way as he will serve as if it were like a. A pal that you have next to you, a colleague from, from work or university studying together with you. Another special case that I really like was by playing chess actually, or any board game. So Richie was just sitting there watching. We were playing live chess with the other opponent. So each time I made a move which you will going was going to make like a sort of emotion to tell me how good the move was or not in an objective way.

B (16:40)

That's cool. On chess or checkers, you can do

A (16:43)

it on both, but we did on chess because it's a bit more complex and even you can kind of play with him. I mean, he will tell you he doesn't have arms for the moment. So he will tell you, yeah, move the pawn to, I don't know, E4. And then you will react to that and he will go, ah, I see that you have played the Sicilian defense. That's nice. But what happens if I play this? And you can actually use the already made LLs that are quite good actually with chess, but as you integrate it with the robot you can kind of feel like somebody's there playing with you. And you can kind of apply the same thing, but for everything that you want to explore. If you want to have a more philosophical question, you can also go to it. But for me, just in that, just in having like this educational, fun approach is a really super useful. And then you have like more, I would say type of corporate or social applications. That is really nice. You can think of people that. Oh no, without your glasses you can see much. So you want some help on navigating your surroundings. Then you can kind of have Richie or I know for people, for elderly people that want to set an alarm or want to have fun at a certain time if they are bored or they want to get the news or they want to navigate some information that without having to go to their phone because maybe they don't feel quite connected to their phones besides having a conversation. So I mean, it's really. The possibilities are quite open. It depends a lot on the. How much people are motivated to construct them.

B (18:23)

Okay. I mean, that makes sense. I mean, I think the, the use case, I think the chess is great. Like, I, I love that it's kind of open. I think that's what's cool about kind of robotics, right? Like is. It's really whatever your use case kind of can be and need and it can be modified to that. Which I think is so cool. I would have never thought about, you know, playing chess. Like the education thing I think is awesome because, you know, I try to keep my daughter up on, you know, AI things because I mean, it'll affect her way more than ever did me. Right. But, but I don't like necessarily like with the fact where she can go in and like ask an LLM on her computer to like answer a question and like it'll give her the answer. I mean, most of the time, I mean it's not always right, but like we'll give her the answer versus saying hey, what do you think? Or like, hey, here's two possibilities, right, Which I kind of love the. And gamified is not the word but like the assistant type where that could be coded in where it's like, hey, I know the answer, but hey, give me a guess first. Like, what do you think? And let's talk about it. I love that I can see this, like, this is a crazy use case and I'm going to try because I'm getting mine, but our dog's getting a little elderly and he's having some stomach issues and stuff like that. And it, it's becoming a thing where he's going around the house and just not to be vile but you know, throwing up a little bit. And like we're finding it in the oddest places because he's very embarrassed. Right where I could see Rishi navigating the house, potentially looking for this and then alerting us or staying on Truman to follow him where it is and then basically using like the vision to like alert us when he's doing something. Do you know what I mean? I don't know. I think it's really cool.

A (20:01)

Yeah, actually that's one of the like as I told you before, like the specific cases that you thought that it was super specific to you but actually makes sense a lot for other people. So that's where you're kind of create that specific app that will help you in your daily life but then it will help you like other people that have the same issue or a different one. And it actually is quite simple. Just need like the good user case as you said, as you mentioned. I mean having just reached me and telling me, hey, watch out, there is a little thing on the floor when you enter or even connecting it if you have a kind of cell phone so you can receive an email or a phone just to alert you, just to let you know, hey, your doggy is not really doing good today. Watch out when you enter home. That's super nice.

B (20:45)

That is super nice. So it is cool because you could apply that use case. It could be following around a baby, it could be following around someone who's a, you know, a stay at home, who's a little bit elderly. So there's a lot of different use cases which I love. So let's talk about. We're not going to get super technical, right? But like in terms of like operating system of the Rishi Mini, like what kind of, you know, code base is it running that we develop on? Is it, is it Python? Is it something else? Like maybe just high level, kind of walk through some prerequisite skills you would need to be able to go develop your own app.

A (21:19)

So I'm not much of a programmer, but for me I mean it's quite. If you have some really basic knowledge, for example, if you have done one of those online courses before or even if you have seen the code of a web, it makes sense already to kind of how do you use Ritchie? So basically you can use it with Mac, so with Apple, with Windows and Linux. So it's quite open already and then most of the code is Based on Python. The most complex part, and that's the robotics part, is a bit of a bit hidden in the actual Python code with libraries, and then those codes are different ones. But you. If you don't want to go there, you don't have to actually, to just stay to the high level of the programming on Python and then you can do kind of do most of the stuff for the movements for connecting other libraries. If you want to have, I don't know, Rich Mini, use a specific library to get some info, and then using an LLM to kind of sort it out for the voice commands and also the applications. The high level of the connection is also done in Python. So the idea here is that, I mean, you have. You need to have some. Some kind of background if you want to develop your own, but if you don't, actually, with the assembly guide that we provide with the robot, helps you a lot to actually, even if your first time using Python, you can kind of already start doing it. Because that's also a big chunk of why we were doing Rich Mini, because we wanted to give a real experience of how it feels to create your own robot. I mean, from. You have the screws, you have the connected cables, the microphones and everything. And then, yeah, there is a little bit of code, but you don't have to try to be afraid of it, actually. So you can kind of be more. At the end, when you have built Richie and you have launched successfully your first app, you are like, yeah, I actually done something for myself that I have never done before. And it works. So that kind of magic, that will feel you more connected to what you have just built.

B (23:26)

That's really cool. I mean, that's what I'm actually planning on doing, is that the robot's supposed to be here. Like, I think it's Friday and my daughter has a few sporting things, but this Sunday we're going to sit down and, like, do it together, because I think it'll be a good experience, right? Like, and I think it's like, it's one of those things. It's like, when I heard that it was coming and you have to build it, right when I was reading, I was like, oh, man. Like, is this like IKEA furniture?

A (23:47)

But.

B (23:48)

But it's like, it's the investment in it. Do you know what I'm saying? Because I'll understand. Like, all right, well, here's where the Raspberry PI. Here's where the computer vision camera is. This is how it all works. Like, I think it's super cool how it's done it. And it's not in a scary or overwhelming way, but it's done in a way that, like, teaches and pushes you a bit, but not to an overwhelming level. And like, just the chatter that I see, I mean, I don't. Chatter is not the right word. But like, when I look at, you know, on hugging face and I just see the level of support and I mean the amount of units, I mean, this thing's going to be big. I mean, I'm super excited, to be honest. We can't tell.

A (24:21)

That was also, like a strong debate between us. Like, are we going to send Richie already assembled? Are we going to send it as an assembly kit? And actually, we. We kind of tried to push people, as you said, to build them by themselves. Even if you. When you hit it like that, you get some sort of negative emotion. Because I have to build it. It takes one or two hours. I don't want to. Can I buy. Just. Just already built. And so I unpack it and play with it. But actually, this is a really important point because as you go through the whole experience, not only you understand how the robot works mechanically on the hardware, but only it gives you a sense that you have created something for your own that is your own robot that you have built, not just another one. And also to avoid kind of for the robot to. Okay, yeah, I abacked it. I just launch it. I play with it 10 minutes and then it's gone like this. As you created and you have spent 1, 2 hours as you have spent those hours, now you want him to work and now you want him to do something that is useful for you. So actually we have seen that it helps a lot on the whole experience that people have to actually invest more on the time that they will have with Richmini and say, okay, now I've built it. Why not learning some programming? Why not applying some apps that other people in the community have done? Because I spent already some time on it. Instead of just purchasing it, open it, doing it already from the scratch, and then that's it. Like a Tamagotchi, you can kind of forget that it ever existed.

B (25:50)

I love it. I mean, I'm so excited. Like, you know, this is. This has been great. One of the better episodes we have in a while. I do want to ask. We're getting kind of close to the end. I do want to ask, like, a couple of quick questions before we wrap up. You know, we talk a lot about robotics and kind of agentic AI. So you've kind of mentioned both. Like, does Rishi fall more into the robotics bucket? Is it more into the agentic bucket? Like, why would you say one versus the other?

A (26:16)

I mean, both at the same time. For example, from a more technical or roboticist point of view, Ritchie has everything that you will expect from a rally. It has degrees of freedom of movement. It has microphones, sensors, so cameras, and also has some way of moving or some. How do you call it physically kinematics to it. So the way it moves that those are really robotics concepts. So you can kind of learn also with Richie about robotics in a general term, but also it has agentic AI capabilities, meaning that you can automate or you can create your own capacities using different tools or different pieces of AI tools to create a whole process. And that's where agentic AI comes. For example, you use the tip to speech, tip to speech, speech to text, the camera, and then an LLM and maybe and also like a whole library. And you connect everything and at the end you get the whole application for, I don't know, Rich Mini takes a picture of the drawing board and then it creates an image, as you prefer, for what you. What he was seeing. For example, if you are, I don't know, not a good artist. So you create a sketch, he takes the pictures, and then he converts it into a nice picture on your PC using AI. So there you get the agent capabilities. It goes from one, let's call it, place one medium that it was the camera to another virtual medium that is your PC.

B (27:51)

I love it. So, I mean, I think I could keep talking, but, you know, we're. We're gonna, we're gonna stop because we're gonna have a more technical version of. Of this conversation releasing very quickly before GTC as well. But before we go said, yeah, it was great. What kind of takeaway should. What. I mean, if somebody was listening to just came into this episode and you were giving the 30 seconds on Rishi, kind of give them that 30 seconds they need to take away on Rishi Mini.

A (28:19)

So actually, what I will say is that if you thought always heard about AI and you were kind of scared or, I don't know, had negative emotions about it, but at the same time you were curious. And especially with robotics, I think that regimen is a really nice experiment to have and to test by your own where you can kind of a robot today. What can a robot do? What AI models can do and what they can do to not only help you your daily life in smooth way. For example, with taking Pictures, setting an alarm or just by having a fun moment, but also kind of learning and discovering what is behind the veil with these models, with these robots and how you can also contribute. So not only taking a passive kind of pose where you let the technology evolve and we are next to it, but also kind of be more participating, giving also your opinion. Because as, as we are open source, you can always share your thoughts with the community.

B (29:19)

I love that. So you know, where can, if someone's interested in Rishi, you know, Hugging Face, Poly Roblox, where can they, where can they find you on the Internet and then where should they go to purchase one?

A (29:32)

If you want to hear more about me, you can just type Santiago from Polyden Robotics or Hugging Face and you will find me either in Discord or LinkedIn or on web. You are for getting your own rich mini. You can just type rich Mini on Google and you will see the page from Poly Robotics and also blocks from Hugging Face. You can use both to get your rich mini in just around five clicks already.

B (29:55)

That's amazing. So with that great episode, really excited more to come between Dell kind of Pollen Robotics, Hugging Face, Nvidia around the Rishi Mini. I will definitely following me on LinkedIn, you're going to see a lot of updates because I will be setting it up, trying to get it to do some fun stuff. So with that robotics, another example of robots are not here to take over. This is a great example of, you know, from, you know, higher ed to corporate use case where you can kind of kind of meld both worlds like robotics on kind of a smaller level, but then the agentic capabilities of AI, ML, et cetera in something that can kind of make your life easier. So with that being said, another great episode. Santiago, appreciate you having on you coming on. I'm sure we'll do it again. And with that until next time, we'll see you on the next episode.

A (31:03)

Do what you want. Do what you want.

B (31:05)

This podcast was produced in partnership with amaze media labs.