Transcription
Welcome back, everyone. Uh, so the next presenter is, um, two very fine, uh, software engineers with metallurgy backgrounds, geology backgrounds, um, Mikita Lipsky and Hayden Brighton. But before they come up, I have to tell a bit of a backstory.
Okay, so I think we're all familiar with this, uh, National Instrument 43-101. That is, um, you know, used, I guess, to ensure that what is being reported as being under the ground is actually there, and to validate, you know, the, uh, the presence of mineralization, etcetera. So, so these documents are filed by companies that, you know, are getting serious about opening new mines. And, you know, for us in the innovation space, you know, one of the things that we're trying to do is find the connection between the NI 43-101 and what we do in terms of developing technology. So, so what we did is that we, we commissioned, um, MBR to do a study, not a study, to create an application, a software that will enable us to be able to identify those opportunities, those synergies between what we're doing to produce innovations and what is inside the 43-101. It's really, really unique what we've done. Uh, so without any further ado, I'm going to call upon the team from MBR to give us, um, a glimpse of what's possible when we can mine the 43-101 for the insights that we need on the innovation side. So, guys, come on up.
Yeah, thanks for the introduction, Shamai. So, you know, we estimated in 2023, there were 175,000 pages of, um, NI 43-101s released, right? And I can almost guarantee that no one has read all those pages. So the, the project was, can we ask and answer very specific questions from these reports? And the first question that we got commissioned to ask and answer was, how do we help the MIC members commercialize? You've already done all the hard work, you've developed your technology. Um, if you are, um, treating selenium contamination, you know, there's 2500 listed companies on the TSX to do mining, but I guarantee that not all 2500 care about your selenium project. So which ones do? Um, and that's the, the question that we answered.
I'm just going to jump through these, um, so this is MIC, great organization, they're growing. Um, this is you guys, you're at some point along your journey, and one thing I notice that a lot of you, some of you have clients, some of you don't. Um, so you're at the commercialization stage, you're asking for some money. Um, in some cases, it's a lot of money, and, um, you're asking companies to take a risk. So how do you identify these companies that are willing to do that? And, you know, I said 2,200 mining companies, but not all mining companies look the same. That's Grasberg, massive mine in Indonesia. And then we have, you know, Gold Rush Alaska. These guys don't have the same challenges. Um, it's not the same environment. Uh, they don't care about the same things. So really understanding who your ideal client is, um, will, will help you, will help you with your business development. And, you know, you really need to get this right because mining, although in terms of GDP, it's quite large, in terms of companies, it's really small.
So this is from the Government of Canada. Um, there's about 150,000 incorporated construction companies in Canada, and there's 2500 mining companies listed on the TSX, and most of those have a market cap of sub $30 million. And so they're not even in a position to take a, a chance on your tech because they don't, you know, they're keeping the lights on. Um, yeah, so this means that you're trying to focus your sales efforts at these very few companies. You need a very targeted message because you don't have the opportunity to have a shitty closing rate. You need to, um, have a very targeted message to a few number of companies about why your technology is the best. And finding these opportunities is very difficult. You know, you guys are PhDs, you're this and that, you're in the lab, you've developed the technology. Do you have the time to sift through all these technical reports, all these press releases to determine which projects and which companies are interested in adopting your technology?
So this is where our platform comes in, and Mikita is going to give a demo in a few minutes. But basically, we've done the hard work for you. So we've loaded all the technical reports from 2008 through to the present, minus a few weeks. Um, and we've used large language models to read them. And, you know, hallucinations. So we've, um, incorporated a human validation aspect to this. So all data points are validated by us. Um, but we just use large language models to generate that first pass, and that's 90% of the time. And then for the most part, it's just us saying, yep, that's true, that's true, that's true. And so that's how we're able to generate so many insights so quickly with a smaller workforce because we're not reading these reports ourselves. We're just reading the quotes that the large language models use to generate these insights.
Um, yeah, so the, the question was, which companies are facing which risks? And so we've categorized, we've created 10 categories of risks based on the MIC members, and we've identified over 5,000 instances of these risks just from 2024 projects alone. So that means that there's 5,000 instances of companies potentially requiring your technology. And, yeah, this is, this is how we do it, basically. You know, this slide is supposed to show that, you know, we're smart, we're switched on, um, and the green box means that you should trust us because, you know, our eyeballs have been on the data as well.
So looking through the member list, we decided to pick on P-MAP. Um, so they seem to be an environmental company. I'm not sure if they're here or not, um, and they're championing a new technology for treating acid rock drainage. So, um, for our little internal demo, we just said, okay, is our, you know, did Shamai waste all his money on our project? Um, so we looked at acid rock drainage in 2024 and we identified over 100 projects that listed it either explicitly or implicitly, um, as a risk using our technology. So that's then potential 100 leads that P-MAP could go to, start annoying them and seeing if they're interested in their technology. And so, so this is just the start.
Um, risk and opportunity, that's where we're at. Um, we're also doing the same work on press releases. So, um, you know, we're extracting drill results, um, from the press releases. You know, we have a client that wants to update their Leapfrog models and all their neighbors. And so we've created a pipeline where they can do so. And, um, also detailed cost and operating parameters. You know, if, if, um, somebody releases a tech control report and they're saying, oh, we, we think we can mill at, um, you know, 16 kilowatt hours per ton, but everyone else in the region says, no, it's probably actually 25 based on the technical reports, then you have the opportunity to say, as an investor, um, you know, that's, that's probably not true. We should update your power cost assumptions. But, yeah, so this is, this is just basically the, the start. And, you know, our vision is to become the leaders in mineral intelligence.
Yeah, so I'm going to kick off and give the platform to Mikita, and he's going to show what the platform actually looks like.
Welcome to our platform. We, uh, very honestly called it M-AI. Um, basically, this is the landing page where any user of this application would land. Um, as you can see on the top, you would, uh, you would see the latest reports that we pulled out of SEDAR. The date effective does not really count, they're all sorted by the date that they were been uploaded to SEDAR itself. Um, you would see the company that has prepared the report and so on, so you can quickly assess, uh, what's happening there. Um, in the bottom, you would see the companies and the projects, uh, that you already investigated and you're interested in, you're following. So as NI 43-101 reports coming, there's amendments to them, there's press releases, we're following them, you would get notifications, you would see what's happening and what new insights that we have extracted from them.
Um, so M-AI itself is a database of extracted information from 43-101s. We've currently identified a couple sections. So as Hayden mentioned, we've got the companies, TSX, ASX, and some other companies, sometimes private that are have parent companies. Uh, we identify mining companies and also consulting companies that help write those press, press releases and NI 43-101s. Sometimes it's small independent companies, sometimes it's big companies like SRK. Uh, the qualified people that actually have happened to write those NI, NI 43-101s, so we extract them, their designations, their positions, sometimes they change, sometimes they not, but you want to keep the information on who actually wrote which report, so we have the history of that. And of course, the projects that have been connected with the, uh, extracted for NI 43-101s.
And as new, uh, pre- not press releases, 43-101s come in, um, the stage of the project and their mine type will be updated. Sometimes it's too early to know if it's an open pit or underground. You would just know that it's, um, basically we identify if it's too early, and as time goes on, we'll be able to update this information. But as, um, my name, MIC member, you would be a, you would want to know your potential leads, your customers. So we've, uh, selected the risks which we extract from 43-101s. As Hayden mentioned, some of them are inferred, some of them are explicitly mentioned by the qualified PE persons. And if we take an example of P-MAP, um, you would probably want to search, uh, through the list of 10, uh, designated risks that we have identified. The first one being acid rock drainage.
Um, so currently we have 10 types of risks, but as more people and more companies start using it, we'll be able to develop more finite, uh, more targeted risks. So, um, there's a bunch of companies, a bunch of risks that have been identified from different reports. The reports had can be downloaded, investigated. Explanation of why each, um, risk has been created is mentioned. But for example, we want to, uh, get the company that is in production stage. So we would go and identify somebody like Ore Zone, that has production in Burkina Faso, where Hayden and I just came from. We would go and see that they have a bomb-bomb project, and from that information, would want to go and investigate the risks that the system has given to us.
Going to go quickly switch something quickly. So basically, we wanted to show you that, um, the risks that we identify, it could be acid rock drainage, it could be all of them, some of them are inferred, some of them are not. We get from, uh, the report itself. So currently it's the latest report. Each risk, so we have identified acid rock drainage, um, in reported explicitly, risk in the section risk analysis, with a quote given that the main environmental risks associated with this project is acid rock drainage. So for you, you know immediately that it's potentially your client. As you go further, you can start seeing the patterns that there is multiple risks for this because it probably explicitly has been mentioned in a couple sections, uh, including ground contamination and so on.
Um, so as I mentioned earlier, um, this information, we do generate it through, uh, large language models, but we do need to validate it somehow. So we've also developed, uh, a model for a team of validators to go through this information. And as an example, just wanted to show you one of the later reports, how the validators go through it. So the AI generates the information, they've got the snippets of information that are just extracted from the actual report, and the people, the validators, aka can go through it and ensure that all the information is correct, validate it, improve the information, add new fields if needed, highlight whatever we find as the quote, and that will be passed to you as the final user.
And so finally, as the, as the MIC member, you would be interested in following this project and seeing how it develops over time. So, yeah, that's like a little short demo of the project right now.
Hey folks, if you have any, anybody, if anybody has any questions, feel free to ask, but please grab the mic so that we can hear your question loud and loud. Thank you. We've got about five minutes for Q&A before we go to the next session.
Questions. One comment while we're fetching the mic. Um, the qualified persons to me is also, um, like a really important aspect of this, um, understanding who, you know, the most important QPs are in the industry. You know, and then we don't have it here, but like filtering by commodity, by geography, you know, if you're developing or trying to target, you know, like South American copper or something like, like that, you can very easily see like, okay, you know, um, um, George Clooney is the, is the most common technical author for the metallurgy sections for, for that type of project. Um, and then eventually correlating that with actuals and you can say, but he does a job, like his grade reconciles poorly. So that's kind of where we're also heading.
Yeah, okay. Thank you very much, uh, for the presentation. I, I have met these gentlemen before just a casual social event, and this is the first time I've seen the product. Um, I, I'm a QP and I specialize in these 43-101 type reports. Um, so I'm just very interested in the metadata that you select for sort of summarizing. Is that still a work in progress? Is that something that, you know, you continue to develop? Because I think ultimately, as a QA tool, we might be able to dig down and, as you mentioned, who, who are the firms, who are the reputable, more reputable, less reputable, more trustworthy data, those that are perhaps telling a little bit more of the truth, and those that are painting a somewhat rosier picture. So is that, is that one of your, I guess, goals is to pull that into the metadata and track that?
Um, yeah, absolutely. Um, so this was kind of the first step of the project, and this was meant more as like a BD tool, but we have all the project, sorry, all the QP information, and one of our next steps is gathering cost assumptions, you know, like met test work, even, you know, the QAQC on the drill, drilling databases. So we want to be at a stage, and we think, you know, later this year, early next year, um, you can say, Brian, okay, this project came out, they said a 16 kilowatt hour per ton, um, for milling costs, but, you know, their neighbor has been operating for five years and they put 22 for their expansion project. So, so can you justify that 16? And then also with the reconciliation aspect in terms of like the mineral resource, yeah, absolutely. It's, so that that involves connecting with like annual reports and other sources of data, which we've, um, already half done with the press releases, but being able to say like, okay, um, you know, the mineral resource estimate came out at 1.5 g per ton, but they've been operating at 1.2 the last few years, like something, something happened. So, yeah, yeah, we're absolutely looking into that.
I, I have a question. Um, I, I can see the value of mine data information and save time and identify right opportunities. But as you mentioned, you need to validate. I've had some experience in industrial databases with, uh, 70% in average of accurate information. So one out of three, one out of four, you always had in your mind, if this is accurate, if the phone number is right, or is the right person, etcetera, etcetera. So you mentioned that you validate. So by the time you have a report and the time that you validate, so this is someone that will spend time. If you have multiple clients, how does that work to have that? Is it two hours per request, or is it half a day per request? So this is not clear in my mind.
Yeah, so, um, by the time the product goes live, which should be in a few weeks, we'll have, you know, all the reports in the system will be validated. And there's roughly two reports that are released a day. So, and because, because of the way the system is set up, it, the AI runs automatically. So, you know, we wake up in the morning, or whoever wakes up in the morning, and there's an alert, you have unvalidated data in the system. Um, on the user side of things, that data is already in the database, but it's flagged as unvalidated. And then we go in and we, uh, manually check each data point. Just the, you know, originally we wanted the large language model to generate everything, but very quickly you realize that hallucinations are too prevalent. So, we're around 90, 95% accuracy on the large language model, which is pretty good, but it's obviously not good enough. So now it's just become an exercise of speeding up the data collection validation, where 95% of the data coming in is correctly modeled, and then we just click one button saying, yeah, that's right, instead of having to read the whole report. So we find validation to be pretty quick.
Thank you. Thank you, MBR. I can already see how as a business development tool for a lot of these solutions looking for customers, so that could be a very valuable tool for them to use.