S2 E57: Inside Earth AI: Cutting the Cost of Critical Mineral Discovery

Dustin Olsen (00:40)
Hi everyone, welcome back to the Rare Earth Exchanges podcast. I'm Dustin joined by a guest, Roman, who is the CEO of Earth AI. Roman, how are you doing?

Roman Teslyuk (00:51)
Hey Dustin, I'm doing great. Thanks for inviting me.

Dustin Olsen (00:54)
Absolutely, Roman. So you are doing something pretty novel using modern technology. But first, before we dive into all of that, I would love to hear about your background that ultimately led you to Earth AI.

Roman Teslyuk (01:08)
Sure. I always was inspired by, you know, the frontier explorers going to unknown lands and I thought geology, exploring for minerals sounds pretty sick, going to the desert. So I thought it was a great job. I really passionate about it. I was born in Ukraine in a pretty normal sort of family.

picked geology, studied really hard, became the best in class and best in the country. We run these Olympiads, so I was best student in the country one time. And then I thought, I want to explore the frontier, worked in the Middle East, Far East, the Arctic, and then decided to do a PhD in Australia. Because Australia is the biggest mining market in the world, a lot of research here, a lot of funding. It was really fun doing the research, but I started sort of…

figuring out how can I help the industry and help the world. And it was kind of clear to me, A, there's a real big decrease in number of discoveries in the world. We used to be finding 150 new deposits per year. Then that number dropped down since I think it was like 150 into thousands. And then now it's like 40 to 50 a year globally. That's like discoveries means new ore bodies found.

I thought it's really, you know, massive decrease. And I thought, you know, the logistics is really slow and expensive. It's like cost 300 million to find new ore body. And also I just saw how much inefficiency there is in the industry in terms of like a lot of archives have been unused. And I saw this opportunity where I thought, okay, and Australia, by the way, keeps the archives for a very long time.

It keeps kept the geological archives since 1970s. So we have 50 plus years of like complete records failures and successes of, you know, tens of millions of geoscientists. And I thought I could build what if I build a learning system that could absorb all that knowledge. Learn from these successes and failures and make far better predictions where the sort of next critical minds, critical mineral minds of the future will be.

I thought that would be an awesome quest and a great exploration journey. So I dropped out of my PhD and started Earth AI. That's sort of my background, know, condensed the past 10, 15 years.

Dustin Olsen (03:30)
Yeah, that's impressive and you've been all over the world to learn a lot of this. let's talk about Earth AI. What is your mission then? So you're obviously taking all this archive information, creating your own LLM of mining data. So now with all that in mind, what is the mission of Earth AI?

Roman Teslyuk (03:53)
Well, the big mission of Earth AI is to create the abundance of critical materials to build the future world. think there is, one, it's just the shortage, but two, it's sort of like when you compare, which you recently did this analysis of like who produces not even critical, but even like super critical strategic minerals.

People that talk about you know, where's germanium other things? If you look at the you know, China Russia and their allies they produce 70 % of the world supply then the rest like the 20 % is produced by sort of these neutral undecided countries the Western world is only produces 7 % of the world supply of these strategic minerals I know us government has us government has been investing in

in western mines but like even if you you know help all of them there's only 99 western mines or western alliance mines versus 900 of china russia neutral like the only way to win this this this this race this critical minerals race is to find new discoveries new mines in the hundreds in the thousands and that's sort of what our technology

does it, I think that's sort of in the next five, 10 years, I think that's like the critical mission of Earth AI is to just overtake China and Russia in the number of new strategic and critical mineral mines. That'd be sick. While doing really sick work, exploring remote years, drilling down, you know, 3,000 feet, just a lot of fun.

Dustin Olsen (05:33)
Yeah. Okay. So I want to, I want to go back to something you had said earlier, commenting on the decline of projects that are being discovered, announced, what have you. What is your opinion on why that is? Is it a talent thing or we're just losing the talent and we don't, or have we already discovered so much that there's just less to discover?

Roman Teslyuk (05:57)
⁓ In short, the low-hanging fruit has been picked and you need to do much more work to find new mines. I'll explain why. First of all, mineral deposits are formed by forces of magma and water. All of the water we have on Earth on the surface, the oceans, the lakes, the rain, the clouds, all came from magma originally. So the process happens like this.

It deepened the mental, know, hundreds of kilometers deep, have this parts of the mental melts, goes up, rises and intrudes the earth's crust. And that magma soda fires into rock, but at the very end, you have always just few percentages or a few, you know, parts of percentages of water and other fluids as incompatible metals and materials that escape that magma.

gush into these surrounding rocks, create this convection cell, because you have really the hot magma called a crust, convects, and that's what concentrates the metals. And the water escapes and flows back to the surface and the oceans, but that's what concentrates mineral deposits. This process happened since the births of the earth. have deposits that are billions and billions of years old. The rocks don't age, they look exactly the same, but…

You know, one story short, is our crust, Earth's crust is full of mineral deposits. Now, how did mineral exploration technology evolve? You're taking a big step back from prehistoric times till, 20th century. People were just walking around stumbling on these rocks. They were shiny, colorful. The oars, they were just outcropping right there from the soil, from the Earth. Just, you see them.

walking around, driving, flying, they were like visually identifiable. We picked all that fruit. Then we decided, okay, if this is a body of rock, it's full of metal, it should have striking physical properties. And we could map that by doing surveys. So we flew surveys all the 20th century and we keep doing more and more of those surveys. But essentially as we covered the globe, we were identifying these

Striking off the chart anomalies in those in that survey data. There were just huge peaks that are caused by these shallow ore bodies say 1 to 200 meters or 3 to 600 feet under the surface You know the it becomes like this massive red dot or you know red blob on the map They just go and drill boom because that that rock is so anomalous And we pick that fruit as well. So now

When geologists just rely on the survey data and we have covered the whole world, there's only a few places remain that have not been surveyed. There is only these low level anomalies left. So it's kind of like, was the way geologists use the data when they just kind of eyeball it and do a rudimentary data analysis, they just sort of, it's indistinguishable whether this anomaly is caused by mineral deposit or it's just a weird rock.

So the success rate right now is one in 200. So one success in 200 attempts at the status quo. And because we put real anomalies without knowing whether they're going to be related to mineralization or they're just a weird rock. So what we did differently is we, our model, the science behind our AI system is that for every anomaly we add the context.

So when we drill, we don't drill just any random anomaly. We only drill those anomalies that our model has predicted they have relationship to this hydrothermal system and to mineral deposit. So when we drill, 80 % chance that's proven on five projects and four discoveries, 80 % chance that we're gonna hit something that's gonna be mineralized. It's gonna be a new discovery.

Dustin Olsen (09:44)
That's really awesome. And in that contrast of traditional methods versus what you guys are doing, which kind of begs the question, you guys can't be the only ones out there, right? You have competitors, right?

Roman Teslyuk (09:58)
Yeah, yeah, totally. We have ⁓ Cobalt Metals. They are the best funded competitor in our ecosystem. We have Ver.ai that do also some more type of prediction. Then we have Terra, Terra.ai I think. They're also doing some sort of prediction and modeling. Where we differ is…

We are more kind of big scale training and prediction. We train on the whole continent scale data set and we'll learn a lot of things. We have built our own science over the years to detect these hidden signs of mineral deposits. And then two is we are vertically integrated. So we have our own geology team, operations team. We design our own drilling hardware. We run our own 24 seven drilling operation and

We're also building a lab, so a kind of complete cycle that we are independently picking land, developing it, drilling it, analyzing it, develop these, make discoveries, develop resources, appraise them, see how much are going to be worse, are they economic to mine, and then we're going to sell them off to mining companies. So we're different technology, more content of scale training, and we are vertically integrated.

Dustin Olsen (11:11)
Okay, I think that's a great distinction. then, so let's talk about the business itself. You guys are based in Australia. As you said earlier, Australia is very mining friendly. Is that the only country you're in or are you operating in other countries as well?

Roman Teslyuk (11:25)
So we are just a sort of a quick technical detail. are a US company. We went through YC in summer 19, but then it doesn't make sense to start in any other country because of a few reasons. Australia is the biggest mining market in the world. My money is like a critical mineral exporter in the world. The permits are

really quick. You can permit a mine within a year. The quickest time from discovery to production is three years. Sanfire Resources developed the Grasse in like three years from when they discovered it and then started pouring the metal. It's copper. On average, like so average median time to develop is like four to six years. So it's really fast to operate.

really mining friendly legislation. It's already a very, like very well established ecosystem and process. So it makes a lot of sense to be here. we, you know, it just doesn't make sense. Secondly, for us kind of our specific company is vertically integrated. So we need to predict then really quickly we go send a geologist to the ground to test. And then we send a drill rigs to

to prove that the metals are there, make the discovery, I'll give you the definition. Discovery is when you drill out and take a sample of rock that has metals in equal or higher grade than currently producing mines. So you know you can make money by mining this ore, but the question is how much of it is there? Then you have to go do resource definition and feasibility, which means drilling an extra 20 holes, then drilling an extra 100 holes to have this really well-mapped subsurface.

deposit that you can like guarantee there's no geological risk remains and you can you know build a mine and make a lot of money on it. So when you do all of this in-house it makes a lot of sense to be concentrated. So we work in New South Wales which is a sort of state where Sydney is on the east coast of Australia and Northern Territory which is like a really wild state. Three times size of Texas.

And the population is 180,000 people. I think 100, 150 of that or so, like 80%, 90 % of that is in Darwin in like the capital city. It's like nobody, it's just desert. And if you've like farmland, like big, big, big ranch lands, big farms. So fun to work there. And then New South Wales, we are the biggest landowner in the state. In total, we have in our…

ownership and partnership 3.1 million acres. That's like size of Qatar or size of Connecticut. Big pipeline.

Dustin Olsen (14:06)
Yeah, that's incredible. So how many people work for Earth AI? How big is the company?

Roman Teslyuk (14:11)
Thunder 40.

Dustin Olsen (14:12)
Just under 40. what, break that down for me. So you got geologists, scientists, I'm assuming, obviously drill operators. What does that look like?

Roman Teslyuk (14:20)
Jewel operators, geologists, operations team that manages the process, secures mineral rights, gets the permits, hardware engineering team, AI team, and the executive team. That's it.

Dustin Olsen (14:31)
Cool. So to recap some of what you just said. So Earth AI specializes in finding the ground that might have the minerals that we need. You drill it, you observe it, you determine viability for a profitable mining operation. At that point, you sell it to a company that then has the resources to actually do the mining.

Am I correct so far?

Roman Teslyuk (14:57)
Yes, so that is the main model. We do think that if we found a mega mine, a gigantic resource, then we might as well just contract a miner to develop and mine it for us. That's a pretty common practice in the industry. There are miners that don't operate their mines, they just use a contract mining operation to pay them the fixed cost.

take all the profit above that. So if it's a mega mine, it would be stupid to sell it. If it's an average mine, then we'll sell it and keep maybe a small amount of ownership, royalty.

Dustin Olsen (15:30)
Got it.

Okay. So then just to be clear, your ideal customer then are the mining companies such as Aerofura, they're in Australia, Linus, is there anyone else that you're targeting when it comes to selling the land or is it just companies like that?

Roman Teslyuk (15:47)
I think that that is typically who you would sell or partially sell, they can operate it. I think the way we think about it, we will just run a competitive process when the project gets there. I think there's two more things that is emerging and something we want to look at. One is, you know, selling the off take earlier. This could be a potential to kind of get some earlier revenue.

or you know kind of direct the we want to direct make sure we direct all our material into into the US for kind of support the new wave of reindustrialization so we could sell the offtake before that and three is I just see that there is a lot of new funds that you know emerge that want to fund critical mineral projects so that might be another avenue to sort of sell or partially sell

the projects too. As you know, there's massive trend now for critical minerals and as well as new explorers and new tech companies and a lot of new capital is pouring into the industry. So that's another probably third avenue that we think about.

Dustin Olsen (16:52)
Okay, so you just said something interesting, right? You kind of highlighted ⁓ potential investors, financiers to be part of this. So let's dive into that. Do you have relationships there? You've already secured funding for Earth AI. Do you talk to them to say, do you wanna be a part of this? Or what does that look like for somebody who might be interested?

Roman Teslyuk (17:14)
Well,

Mining industry is sort of interesting. It says there's a lot of parallels to say what property development where we have to pretty much find the project, assess the project, proved it's profitable to be mined, then to get the honest kind of the fair price when you sell it. So the way it looks is that, if we, if you're developing it and you're

You're sort of like before the house is built and it's livable or you can rent it out. It's like a bunch of walls and concrete and it's like just has such a massive, massive discount to what its actual value is. When you complete it, boom, it just restores to the normal price. So what we expect is to, you know, get a project to the point where we have like 20 to $50 billion worth of metal in it.

and then you sell it for a couple billion dollars. That's like a, normally in industry sell it for like 5 % of the value of the asset. But if you don't get to that point, then you really have to sell it for tens of millions, which what we're really looking for in this business model is that asymmetrical upside where you find one big deposit, a lot of parallels to venture, where you develop say 30 projects and one or two of them will be like fund returns.

⁓ So that's what we're aiming for. So back to your question is like, there's no point of talking to people and trying to sell something that's unfinished. We develop a few projects in the parallel and then once the time is right, then we'll do a proper sell process, run a proper selling process.

Dustin Olsen (19:01)
Okay. Out of curiosity, what do you do with the land that you come across that wasn't as abundant in minerals as you thought? Do you just leave it? Do you sell it off to just, you know, a housing development? What do you do with it?

Roman Teslyuk (19:14)
I think you just have to leave it. know, I think it's, it's, have, you know, as I discovered the sort of, as I, as I, you know, described to you the levels of how technology has evolved. And now we have this sort of new map of, of anomalies we can detect. We can find these heterothermal systems and host mineralization.

But maybe there's a better technology later in the future or somebody else has a different variation of how they detect and it's gonna work differently. They can find something. In terms of industry and how it operates, there's a lot of people that just kind of like take that land and sit on it and they don't have anything on their licenses, but they just keep it just to prevent others from doing your work. And I think that's just the…

Massively detrimental to the industry because then you know nobody can explore and then you do nothing. It's just the liability So I think with our pipeline and how we you know, take this exploration business Like how we how we think is the best way to run it is you you you you stake the land? Then you have to you know, do the work there actively you have to drill it actively and then if you not finding anything just just let it

let it go and then keep moving along your pipeline. We have like 750 to 800 pre-qualified targets across Australia. That's like one kilometer square each. We're detecting some more and there's so much for us to go and explore. I don't wanna just hold back to something that didn't work out or.

When we test it properly, we think it's not there or it's not going to be big enough. We just move on because there's so much more to to to assess and test.

Dustin Olsen (20:53)
I like that approach. admire that you notice that someone else might have a better idea, a different approach. So why not give it to them so we, I guess can all benefit from their discovery, hopefully. That's a very mature outlook there. So, but speaking of doing things a little bit differently, give me an idea of just some challenges you had in.

with the technology and really leveraging AI to qualify the deposit, what barriers did you have?

Roman Teslyuk (21:21)
Well, I think one is like the work, the science behind doing what we do works really well in the lab, but doesn't transport well to the real world. So in a ⁓ simplistic example, the mineral deposits are formed of

you know, minerals. So there's rocks, which are composites of multiple minerals. Mineral is a chemical compound that has a specific crystalline structure. And you can detect different minerals through light absorption in shortwave infrared. You can see the different minerals because of the different crystalline structure have different light absorption patterns. And then also, you know, the chemical

formula, chemical compounds, you know, gives them specific concentration of different elements in the chemical sample. So in essence, we want to use the chemistry as sort of label, labels for our training. And then we use the that sort of physical data to run the model and learn to connect that physics with that the chemistry.

So it works really, really well in the lab, but when you go out to like real exploration and data that's been taken of the fields and deserts was all the, how to say, all the impurity of it, like the vegetation, the clouds, the mixture of all the rocks, it was hard for us originally to kind of build the right way for the model to look at the data and to transform it into like extract that.

the signal in so much noise. So that was like a big problem, but you know, over the years we build our better and better and better, with higher and higher success rate, higher precision, higher recall. And, you know, coming back to why it's good to be in Australia is that we were able to go in the field all the time and run this iterative experiments. There's even like Northern Territory, as I mentioned, the state we operate in the North.

you can go and visit land without owing mineral rights. can just give a, like there's what's called preliminary exploration. You go and you visit it, you take samples without needing to do anything else. So that was really helpful for us in technology development. So that's one, the technical challenge. Two is the industry is very, very, how to say it, conservative.

dogmatic. There is no kind of like acceptance of anything new. It's sort of, you know, a lot of criticism. People bring up some old examples of failed AI research projects as if like because once one time somebody tried it and they failed that means that nobody else should try this ever again because that's done, solved. That's hard, you know, because you're sort of like…

you know, unwelcomed. that's, and it makes the barrier of entry really high for new technology companies because, you know, traditionally you would sell the software, sell the services. We ran, when we started, we ran 80 pilots. We found that there isn't enough uptake. There's a very small price people are gonna pay for software. There isn't really much recurrence in the work that we do, because it's like, once you give them the predictions and they just use it.

So we thought that, you know, that's a problem. And the problem three is the logistics. We go, we are in this, we have to explore in this remote areas that nobody's been there before. Nobody found anything there before. There's like an open landscape, no support. You got to be there on your own in the desert. know, there's, there's water is really far away. Help is really far away.

So that precludes normal explorers to do much work there. They would rather do something like redevelop the brownfield projects, redevelop something that's been already found in the past. They would come back and drill extra holes around it. There's a bit of infrastructure, try to focus in areas that are near some existing operating mine. It'll be easier for them to sell. There's also more infrastructure, more support.

But that's not what the best opportunities are. That's something that will solve the real problem, you know, at scale. So I still see today that like all of our company is this big logistical exercise, you know, being able to build that sort of organization that has all the intertwining functions and processes that can go and predict software initiates it, operations team licenses it, then

gets the permits, geology team comes in, then the drillers move in. They run this drilling operation that's quite logistically intensive in middle of nowhere at really high uptime, which is sort of like, nobody's really taken that problem of mineral exploration and looked at it as a logistical exercise. People think of it as like a…

like a lottery, right? They'll come in, try one thing, work like they won or they lost and then they retreat. Whereas for us, it's like, we gotta build the pipeline of projects, we gotta run this pipeline, deepen it, widen it, just keep going. And then eventually this approach will guarantee success, you know, will eventually be like produce huge amounts of repeatable, asymmetrical outcomes. So that's the way I see it.

Dustin Olsen (26:49)
Yeah, I think that's really fascinating. just listening to you talk, there's an area of efficiency involved with your approach to discovering a lot of this. And I don't think we've come out and just said it really, but it costs a lot less with your approach compared to traditional methods. Because it sounds like the traditional methods, those that are already established, they're still trying to go for that.

low hanging fruit opportunity to keep their costs low but still be in operation. Is that a fair assessment?

Roman Teslyuk (27:22)
Yeah, so for us, the cost of discovery is like under one million. For industry, as it works out, was accounting for all of the failures, 300 million per discovery. So much cheaper. The other thing, when you look at kind of like from a strategy and tactics point of view, status quo explorers, they want variable

that are really high, but they wanna switch it on and off. They drill once every, like once or twice a year. They spend a couple of million to organize all this drilling. They pay really high costs by outsourcing all the contractors. So their variable costs are really high, but then they can like switch it on, drill for a few months, switch them off, go back to the kind of small fixed cost. For us.

vertically integrating obviously our fixed costs are higher, but our actual unit cost is so much lower. Our drilling cost is a quarter of what others pay contractors. And it's also this kind of element of permanence added to it. We drill all the time. We fail once, who cares? We go drill tomorrow. There's no like, ⁓ damn, the drillers are leaving the site. They can't drill for another six to 12 months. They have to organize the site, organize waste management, organize environmental.

logistics, all this stuff is like, ⁓ for us it's just, well, thanks to our hardware, so we built our hardware to be environmentally friendly, there's no disturbance involved, it's really flexible, resilient and self-sufficient. for us, I, now, drillers send us pictures every day of what they drilled that day, and then if I say, all right, I don't think this whole, you know, makes sense, and now geologists also think, yeah, there's no point of drilling this one, let's move on.

we just say, okay, let's stop this hole, let's move to that place tomorrow. And they pack up and move. There's no extra need to rely on anyone else. No extra wait times. I hate inefficiency. I like things to run smoothly. So if I could build an operation that runs smoothly and there's no kind of these bottlenecks or setbacks or like waiting for lead times, then why wouldn't I do it?

Dustin Olsen (29:31)
Absolutely. So as we're coming to the end of our time here, one more question I would like to ask is, where do you see AI's biggest impact in mining critical mineral sector over the next five to 10 years?

Roman Teslyuk (29:43)
I think so I give you an exploration analogy. So Say I read this book really recently about spice trade, which is really exciting But like it's sort of adjacent to it, but Portuguese, you know, they wanted that they were buying spices from from from the Arabs You know, they would they would move that from Indonesia for a long time And if they when they found out they can sail all the way around Africa

to spice islands and bring it back, that could make 10x profits. And they were doing this for a long time. Spanish got involved, sales from the other side, which was really exciting, awesome book. But eventually we decided to, let's just do the Suez Canal. And then you have massive efficiency gains because we figured out that we can do that and the technology allows this massive shortcut. I think there is all these shortcuts we don't know exist.

in our kind of logical way of thinking, it leads us back to the kind of thought pathways that we knew before existed. What AI can do for us is to kind of reveal these new shortcuts that we haven't thought existed and therefore, you know, massively improve the efficiency, the abundance of mineral resources, the way we mine, the way we explore. And I think it's it's,

It's coming so fast. And I think that in 20 years, 10, 20 years, I think it would look so different. The industry will be so different to what it is now, what it was historically. So I think now is one of the golden ages for critical minerals and mining. So we're lucky to be alive.

Dustin Olsen (31:17)
Yeah, absolutely. I agree with you. think AI on many fronts, in many industries is really going to change the landscape. But I liked your thought process of how it will probably enable a lot of humankind to revisit these neural pathways that we thought were impossible. And we just, we have these breakthroughs. So that's really exciting. So Roman, if someone wants to

Connect with you, be more involved with Earth AI. Where should we send them?

Roman Teslyuk (31:45)
You can just email me at roman at urs-ai.com.

Dustin Olsen (31:48)
Easy. So yeah, if you guys like what Roman's doing, please connect with him. And for those who are listening, watching this show and you enjoyed what you heard, please give us a thumbs up. That helps the show get more visibility and can help promote Roman and his company to other people who are also interested in this topic. And don't forget to subscribe to future episodes. Roman.

Thank you so much for taking the time to be on the podcast with us. We've been so much in the trenches of hearing traditional ways of doing things that bringing in some modern technology and modern approaches, efficiency to the whole industry is kind of refreshing. So thanks for your time.

Roman Teslyuk (32:26)
Thank you Dustin, was a pleasure. awesome.