00:18:07 · JC Lozano, Ryan Deng
[00:18:07] S1: Yeah. So... Gonna do with [00:18:08] S0: those those materials that you're getting out the ground, like the ore? Yeah. What is the future that you envision for kind of what those materials can be used for? Because obviously like, you know, before I did our interview, I read that China does control majority of the the mining and the refineries that's occurring in the world, so. They control a lot of the mineral refining. [00:18:27] S1: Mhmm. That's for sure. Yeah. Mining is pretty distributed Mhmm. Because the resources are tend to be distributed. So we mine the material in The US right now, and then we have our own processing plant. So you've like mined this stuff, this material from underground, bring it to the surface and then you need to process it. And like mineral processing or mineral refining is basically this sequence of chemical operations. Does the money [00:18:51] S0: that you've raised go by fast in an industry like yours? Because to me, seems like it's, like, relative. Right? Like, whatever industry you're in, we have to raise a certain amount to be able to perform and operate [00:19:00] S1: Yeah. The company. Yeah. So, like, part of the thing we're focused on is, like, we wanna use VC dollars to fuel kind of the research Uh-huh. And, like, innovation. Yeah. And then we actually raise the majority of the funding for the actual mine, like, the actual CapEx funding. Actually That comes from like a different pool of investors. So that it invests in the project correctly. [00:19:21] S0: Okay. [00:19:21] S1: Yeah.
00:19:21 · JC Lozano, Ryan Deng
[00:19:21] S1: Yeah. [00:19:22] S0: Well, VC is different than like the actual expenditures of like operations. [00:19:26] S1: Well, yeah. So you're in accounting. So... [00:19:30] S0: Hope not anymore. Let's let's not go back to accounting, Juan. If this garbage truck doesn't stop. I'm kidding. Kidding. Kidding. I do wanna talk more about robots because I find this subject fascinating, especially as we're seeing hardware evolve into a space where, hey, there's general purpose robots for homes. Sure. Right? There's specialized robots in like manufacturing plants like pick and pick, you know, pick and place Yes. Whatever. Like [00:19:53] S1: Those have been there for a while though. Been there for [00:19:55] S0: a while. Okay. [00:19:55] S1: Yeah. [00:19:57] S0: Mining robots.
00:19:59 · JC Lozano
[00:19:59] S1: Well, so like the... Like think about a Waymo. Right? Like Waymo didn't actually like... They don't build the vehicle. Right? Like they took an existing vehicle, they tapped into it and then they like essentially retrofitted it. Right? So they added a bunch of sensors, they added a bunch of cameras, sensors, they had some actuation points so they could actually control the vehicle. And most of it's actually digital. Right? Like you don't see like an actuator on the steering wheel that's like Right. You know, it to turn. Uh-huh. You don't see like a linear encoder on the gas pedal that's telling it how far the gas pedal is pressed down. Right? That's because most of the, like, modern cars, right, are powered by a computer Mhmm. That translates, like, the analog signal from, like, you pressing the gas pedal into a digital signal and back into an analog signal to power parts of the engine or to take control. So it's kind of the same thing. Right? Like, you have all this large scale mining equipment. It's actually, you know, like, there's a ton of engineering that goes into it, and most of it is is controlled by an onboard computer, and it flows through something called the CAN bus. And so, like, this is a digital computer that's sitting on a large piece of equipment. So you can just, you know, hack in... Well, you don't need to hack into it. You just tap into that thing Uh-huh. And then you can put an onboard computer there. So a GPU, a CPU, some memory, a way to communicate, and then you can, you know, do some electrical engineering, some controls engineering. Mhmm. It's like wired directly into the the canvas. And then you can actually read the signals coming off the canvas that are telling you, here's how far the gas pedal is pressed down, here's the angle of the steering wheel....
00:21:20 · Ryan Deng, JC Lozano
[00:21:20] S0: And everything inside the vehicle. [00:21:22] S1: Well, they come with the vehicle. [00:21:23] S0: Come with the vehicle. Okay. So it's already there. [00:21:25] S1: It's already there. And the same thing's [00:21:26] S0: in your car, [00:21:28] S1: by the way. Like, is how... And so you just, like, you read all the signals off Mhmm. And then you can also send signals back into that computer, which will take action on the robot, right, on your piece of equipment. Mhmm. So it's actually like, you you could think about like, you know, equipment as just like a bunch of like robots sitting there without a brain. Oh. And so you can just like put some computers on there, you can put some models on there and and start controlling them. [00:21:52] S0: Are you just using, like, general models that are available, like, just traditional LMs?
00:21:52 · Ryan Deng, JC Lozano
[00:21:52] S0: Are you just using, like, general models that are available, like, just traditional LMs? [00:21:57] S1: Yeah. It's a great question. So we are doing kind of a a variety of approaches. We're... Okay. So we're vertically integrated. Right. And so... But that doesn't actually mean you do everything from scratch. Like, we don't actually build necessarily all of our own equipment. That doesn't necessarily mean we do all of our own autonomy. So right now, we're partnering with a couple of autonomy companies Mhmm. To automate some of the parts of equipment, and we're doing some of it in house Mhmm. To to understand how it works. And so the approaches that folks takes are are pretty varied. There's, like, classical robotics approaches where you sort of do, like, your traditional sensor fusion and route planning and mapping, and then there's more data driven approaches like VLAs, like from physical intelligence. [00:22:34] S0: And and remember, our audience is nontechnical, so Okay. We're probably hearing a little bit of a Okay. A buzz buzzwords k. Buzzword alert.
00:22:34 · Ryan Deng, JC Lozano
[00:22:34] S0: And and remember, our audience is nontechnical, so Okay. We're probably hearing a little bit of a Okay. A buzz buzzwords k. Buzzword alert. [00:22:41] S1: Okay. Got it. Yeah. Data driven approaches just mean you've, like, trained a model Mhmm. On it, and, like, think of a large language model or a foundation model, right, like Chatuchipiti or Claude. It understands images and then it gives you text back. Mhmm. Vision language action models are kind of a similar sort of architecture. There is what's called a vision language model backbone and then there's like an action expert that basically takes his input like a prompt, go do this thing, some pictures of the world, some sensor feeds. Right? And then it spits out like, here's the the way to control the vehicle. Oh. So that's one approach. There's like, you know, a bunch of traditional approaches, you know, sort of like auto... Autonomy in this in this space isn't super new. Mhmm. What's new for us is that we're moving a lot faster. [00:23:22] S0: What percent like... Do you have any metrics like compared to just like traditional mines in The US compared to when you guys onboard and take over point in the mine where everyone's just like, this mine is cooked, but you guys are coming in and buying it up? [00:23:34] S1: Yeah. That's a good question. So I would say, like, the lithium facility that we're building in East Texas, we'll get that done and ramped next year. Mhmm. So from project inception to, like, full production ramp was less than three years. [00:23:51] S0: Mhmm.