Episode 36 | How AI is shaping Battery 3.0 | Dr Austin Sendek | Aionics

Welcome to the battery technology podcast with me Ken Davis sponsored by Munters experts in climate control systems for safe, high quality battery cell production and R&D. Delivery stable low-due point conditions whilst minimizing energy use. Episode 36 How AI is shaping battery 3.0 One of the groundbreaking technologies that we've all seen make giant strives over the past decade is AI machine learning to all around us now and it's already delivering potential benefits across a lot of different industries. I particularly wanted to learn more about exactly how machine learning and AI is being harnessed in the battery field to accelerate the design and development process. So I'm delighted to be joined by Dr. Austin Sendeck who's the CEO and co-founder of AI Onyx who are based in Palo Alto, California to discuss exactly how data science, material informatics and computational prediction is revolutionizing battery material science. So firstly a very warm welcome to the battery technology podcast Austin. Thank you Ken. Happy to be here. Many years ago I cut my teeth in industry trying to develop prediction models in industrial manufacturing. So it's always been close to my heart this. You guys are a million miles past that so this is why I'm so so interested in this particular subject. It might be worth starting a little bit of background, a kind of origin story of AI Onyx. How did it all develop and how do you present yourself to the market? Just so I understand those aspects. Absolutely. So first question of how it started is really during I would say you could trace the Genesis back to my PhD years at Stanford University. So I was a PhD student there working on computational battery design and my colleagues and I almost want to say stumbled into this field of you know machine learning accelerated or AI accelerated material science because we were trying to solve this very difficult problem of battery material design in particular battery electrolyte design in which there is a huge design space to play with. There's huge numbers of potential materials that all could do the job of a good electrolyte and there's not a lot of really compelling design principles meaning what do I look for when I want to look for a new material that satisfies these requirements. And that led to historically led to a lot of trial and error experimentation, a lot of sort of rules of thumb that maybe would apply in one case but wouldn't necessarily apply in another. And so we found ourselves thinking well boy it would be great if there was some way to sort of synthesize all of the existing data that's been generated up till now, extract some principles from it and have those principles that we could look at to guide our search through the space of materials. And this is about 10 years ago kind of right as machine learning was entering the materials world I would say generally. And we thought boy this sounds like machine learning, what is machine learning? We should probably go look into that. And so throughout the course of my dissertation that was really kind of an exploration of how machine learning could help us accelerate battery materials design in particular in cases where data sets are fairly limited. Right so our friends at Google have the you know the privilege of working with data sets of millions of images, you know 10 billion cats and 10 billion dogs now train a model that can discriminate cats from dogs. Well you know maybe we only have a couple dozen data points of something that is a in this particular case a good electrolyte versus a bad electrolyte. So the modeling was machine learning but machine learning in the sort of small data world that formed a very compelling proof of concept that hey if you can incorporate some physics into your into your models if you know a little bit about the problem at hand you can build compelling predictive models on small data sets. And that really led to market pull around these ideas. So we started hearing from companies that were saying hey you know we we want to apply this method or we have this data and we want to understand how this can fit into this kind of model. We started hearing from investors and we thought well you know we ought to we ought to see what's here. And so in 2020 my two fellow Stanford PhDs and I started the company and now you know we're several years in and we're focusing mostly on liquid electrolyte systems to date and we can get into why that is. But that's that's so really our kind of technology today is oriented around this idea of can I take a kind of can you give me two sets of inputs one is a cell type so this could be a you know what anything this could be a lithium ion you know NCA graphite 4680 cell or it could be a lithium metal NMC multilayer pouch cell or it could be something not lithium at all. And then can you give me a set of performance criteria that you really care about. So hey I need the charge time to be this I need the cycle life to be that. And given those two sets of inputs we can engineer an electrolyte solution that works in that cell to achieve those performance requirements at least that's that's sort of the idea. And I think that's really in some ways the next wave of battery engineering we talk more about that but that's what we said today we've announced partnerships with automotive and well with with an automotive partnership that we've mentioned which is with self-force group which is the the EV battery subsidiary of Portia we're about to announce some more partnerships here pretty soon and it's been after the races. Well that's a great introduction fascinating there's lots in that that I want to pick away at and tease out a little bit more for my understanding really. The first thing I should really understand a little bit more about I think is I do this decision to focus on electrolytes and was that a result of machine learning just lent itself to electrolytes better because I should say it's it's maybe the unsung hero of battery development is the electrolyte. What did it lend itself to machine learning or did you simply recognize that hey this is an area that is not being developed in quite the same ways as the other battery elements. That's a really good question and I can give I think the sort of a top-down answer and a bottom-up answer I think which are the same answer. The top-down answer is as you were saying that the electrolyte is essentially sort of this unsung hero of battery innovation and I I see this as essentially being I think we are now entering what I would call batteries 3.0 which is sort of the third era of battery innovation. Batteries 1.0 is hey lithium ion works and it's good enough to power an electric car right this was this was a major innovation of course if you ever have the pleasure of watching of walking through the Porsche museum in Stuttgart you can see the first car that for NAND Porsche ever worked on which was a fully electric basically looks like a horse carriage with a battery in it and that battery weighed a ton and it had 50 miles of range and it took a week to charge right so we've solved the the Ferdinand Porsche problem right with batteries 1.0 we have a couple hundred miles of range 2.0 is where we start to engineer those electrodes as you were saying engineering electrodes looking in alternative things like lithium metal silicon you high voltage cathodes some of you know some of these alternative chemistries as well sodium ion we can start to engineer the electrodes that's mostly targeted at improving energy density so more energy per weight or per volume in the eb context that's pushing up range those are expensive improvements to make these these electrodes often take you know billions of dollars of investment to make its scale and lots and lots of time to of course scale up processes and build factories but again I think range and energy density is now for a lot of applications approaching good enough right now into batteries 3.0 which is essentially catching up all of the other metrics to the improvements that we've made on energy density so I'm thinking here about about you know peak power about charge time about lifetime about safety and these are all essentially innovations on the electrolyte the interesting thing about the electrolyte is that there's a again a even bigger design space to play with than in in cathodes because you can mix molecules and a nearly infinite number of mixtures there's 10 billion molecules that you or I could procure commercially within you know a couple of week a lead time and you really cannot traverse that space without some. kind of computational assist, right? And by the way, the marginal cost to improve the electrolyte is essentially zero because it's a drop-in solution, it's a liquid that drops into a cell. And synthesis and scale-up of molecules is much, much easier than synthesis and scale-up of these solid components like electrodes. So anyway, so the top down answer I think is to say, "Hey, this is, as you said, this is where the straggler of battery innovation, the bottom down, sorry, the bottom up, I don't want to go bottom down." And bottom up approach is to say, this is a really interesting engineering problem because you have the ability to generate a lot of data, which is again tied to the fact that synthesis is relatively easy. If one data point is a different formulation, well, you take 10 chemicals on the shelf and you mix them on different ratios, now you have 100 data points, right? So there's that's a great way to generate more data for your models. And the intuition around how different mixtures will affect these different properties, this multi-property optimization is really murky. And so it's a really interesting scientific opportunity for machine learning. And we've worked on electrode systems and there's a lot of interesting science there. But the reality is that you're often very limited by small data sets and the kind of inability to quickly iterate, right? We had cases where we found really interesting crystal structures that should have perfect properties. And then you go into the lab and you try to make it and it takes, you know, a week and then the synthesis fails and then you try to tweak the recipe a little bit and then, hey, you know, okay, well now it looks a little bit closer, but we have this defect. And that's a much slower road. So I think there's a lot of both kind of market forces pushing us towards electrolyte innovation and then a lot of kind of interesting scientific is well positioned from a scientific perspective to be impacted by machine learning and AI. And now a message from a sponsor is moist to control a bottleneck in your battery production. Join a webinar by via Sala to learn how fast response dew point measurements can enhance your manufacturing efficiency. Register now at visala.com/battery and optimize your production line. One final thing on that before we move on to the AI side. So are the effects that are possible through this research and development on to on electrolytes? What's the significance of the effects in terms of performance? Are we talking about something which is fundamentally changing the performance of a battery? Or are we talking with electrolytes something that really will only ever have a kind of marginal kind of effect? Just give me a sense of the scale of the improvement that can be made when you get it right from electrolyte development perspective. That's a very interesting question. And it's interesting also because there are physical limits on the positive impacts of electrode design. So there's like theoretical maximum capacities that are associated with these structures which are you know it's how many lithium can you put into a box of a certain size? You cannot surpass some limit. There's some maximum capacity you can unlock. With electrolytes as far as we know there aren't these hard limits which makes it very interesting because it makes you wonder if there is sort of unbounded potential here. This isn't something that the scientific community I think has been able to answer in a compelling way yet is like where are those limits? But you think about the core problems that are that come down to the electrolyte which cycle life is a big one right? How many cycles can you get out of the cell? Some applications you know a thousand is okay for other applications you might need ten thousand. Is there any reason why you can't get you know a hundred thousand? Well, up priori you can't really say no. You can look also at safety right? Will the electrolyte burn? It's very possible to engineer electrolytes that just simply don't burn right? And now what's the what's the value of that? Well okay you might have other safety issues associated with the electrodes but you sort of solve that electrolyte problem. What is the maximum peak power that you could get out of a cell? You know that might be a case where you could put a limit based on kind of limits on viscosities of materials but I don't know that we necessarily know exactly where that limit is. Anyway all of this is to say how much impact you can have really depends on sort of the use case because you could be able to completely solve the problem at hand for one particular use case. You know if your biggest concern is this electrolyte is burning well engineering electrolyte can solve your problem completely. If you're you know if you have a cell phone then maybe you only need two thousand cycles. Maybe you only need you know a certain number of cycles and engineering the electrolyte can get you there. So it's hard to answer I think it's without a particular application in mind but I think in particular applications solving the electrolyte issue can be as impactful or more impactful than improving the electrodes. Interesting. Now I want to get into the physics of this or the mathematics of it really but let's start at a really basic level in the sense of you know how does the machine learn in AI change the game. Obviously things move quicker than with human calculation. I kind of get that but I'm sure there's a lot more to it than that. So if you just give me a kind of basic kind of introduction to just why this is such a powerful tool. Absolutely. So the first question is well what is machine learning and or AI for that matter and one of the definitions that I've heard that I really liked is make sure I get the quote right. I'll probably say the quote wrong but to paraphrase the idea is it's the science of getting computers to act on a task without being explicitly programmed for that task. And so it's one thing to say I have this calculation I want to run just run it faster right that's that's not really machine learning that's more standard computation. The benefit of a machine learning approach is the ability to extract some new insight on what you should be doing in the first place right. And so that that means basically if you have some data a really good machine learning or AI algorithm can identify some pattern in that data that might have eluded your your eyes right. And so one way to interpret this is to say hey I did so I mentioned the cats and dogs image image classification problem so to go back to that if you say hey I took you know a hundred formulations and put them into the cell and they worked and then I put a hundred and put them into the cell and those hundred didn't work what is the differentiator between good and bad that's very similar to the what's the differentiator between these million cats and million dogs. The power of machine learning is to be able to sort of suss out those really subtle differences in this group of things versus that group of things and once that's identified then you can apply that much faster to some big broad chemical space right. And so I think really where there's maximum value to be had here are in cases where the mapping between what you can control and what you get out is unclear and in cases where you know basically your your the benchmark is some some degree of scientific intuition or maybe some degree of models that are okay but maybe not great. And I think of the machine learning and AI as kind of giving us the capacity to supercharge scientific research and scientific intuition because we might be able to notice patterns that we otherwise just might you know might we might not notice at all. So it's not just about moving faster it's actually about uncovering new physics or new science or new principles that we would have never found you know in a million years otherwise. Gotcha thank you for that. There's a couple of things that kind of lead from that which I just like to understand a little bit more. Firstly I guess if we're dealing with things at an algorithmic level everything has to be expressed mathematically or must have the potential to be expressed mathematically. Now in comparison to humans you we don't go through a process of expressing things mathematically to come up with insight it kind of appears right magic doesn't it? How easy is it to express the things that you are trying to capture mathematically? Does that provide a significant problem to people using AI machine learning? That's a really profound question because I think it goes to it goes to the question of of what do we do when we identify patterns in data right? If you were to give me a cat and a dog and say which is which I would extract pieces of information of course I wouldn't be thinking about this I would my brain would just do it. I would extract pieces of information of each of those images and I would say okay well this one has floppy ears but this one's got 20 ears and you know this one has big teeth and this one you know is meowing and I I'm extracting these features. of these images that you've shown to me. And then I'm essentially mapping those features onto some existing or newly trained model of what corresponds to what. Similarly, in the battery science context, the scientists is asked to do the same thing. So, going back to the 100 formulations that are good and 100 formulations that are bad, I might look at that and say, "Well, and the ones that are good, "it looks like the concentration of salt "is a little bit higher. "It looks like molecules that have carbon rings "all mostly seem to perform poorly." Right? So, we're extracting similar pieces of information and then just in our minds, we're trying to validate whether those are actually the right relationships. A core part of scientific machine learning in AI is to automate and expand that feature extraction part of the puzzle. And that's a big part of the, for one, it's a big part of what makes a model work or not is how good your feature extraction is. And features can run the gamut from fairly low information to really information rich. But it's also a main part of where a lot of the research and innovation is in developing new feature sets. And that's a whole rabbit hole that we can go down. There's a lot of really interesting innovations, particularly in deep learning, where we're trying to kind of map this process on to actually how neurons in our brains are connecting. Right? So, you take pieces of information, but then you convolut them together and you can multiply them and you can, you know, add them and this is kind of represents the complexities of our connected neurons. It gets really, really interesting. But to answer the core question, the models are only going to be as good as the information that you extract from them. Right? If a core piece of information is, I made this on a Tuesday and when I made it, there was a full moon, right? You're not going to, well, actually, you know, who knows? (laughing) Far be it for me to speculate about the power of a full moon on scientific research, but you're probably not going to get any signal if you compare that to, you know, what was the humidity in the room? That's much more likely to have an impact on the measurement that you took. Thanks for downloading the battery technology podcast, I hope you're enjoying the journey. 2024 has been a thoroughly enjoyable year on this and with lots of new episodes in the pipeline for 2025, make sure you subscribe to receive the battery technology podcast delivered to your devices. And a big thanks to everyone who's contributed, everyone who's listened and of course to our sponsors and advertisers for their support. We've had an 83% growth in this ship and each episode gets on average 2,700 downloads. That's a big number these days. So I'm very grateful for your participation in this enterprise. Now, as you know, I'm also involved in developing the Gigafactory and Battery Manifactin Expo in Huntsville, Alabama in June 2025. That is looking like it's going to be a very serious event. So if you're interested in the future of battery manufacturing, that will be the place to be in June 2025. It's free to attend and I hope to see you there. Right, let's get back to this conversation. Well, that's interesting. It brings me back to my original days during this 100 years ago, which was trying to decipher between correlation and causation. I mean, my favorite, spurious correlation is the one about there's a very strong correlation between master's degrees in theology awarded at the University of Texas and the number of tile installers in Texas. I mean, there's a remarkable correlation. I don't think anybody would play causation. I mean, interested in how you guys, within these models, build into the model some mechanism for seeing causation rather than correlation. How does that possibly work? Yeah, that reminds me of a, this was a few years ago now, but there was an image that was going around the internet that was showing a really, really strong inverse correlation between the number of pirates in the world and emissions of greenhouse gases. And it was saying, hey, if we want to reverse climate change, we just need to bring back pirates. [LAUGHTER] It's an interesting question. What's interesting, too, though, in the scientific research context-- or maybe I should say in the materials research context, specifically, is you do want to know what is causal versus what is a correlation. But in some ways, if something is really truly correlated, if it works as a design principle, then maybe you actually don't care if it's causal or not. Actually, I can give you a real example from our work. So we were building a model recently for a specific case where we are mapping the electrolyte composition to cycle life. And so this is kind of the canonical problem in electrolyte design is how does the electrolyte affect the cycle life? It's a really difficult one, because the electrolyte can impact the cycle life in a lot of different ways. There's a lot of ways that an electrolyte can break down, that it can induce other parts of the cell to break down. And so you can imagine essentially a flow chart where, depending on the cell, depending on the electrodes, depending on the electrolyte, you might follow some particular kind of failure mode. And there could be signals that are correlated with that failure mode, which are not directly causal to the failure mode. One thing that we noticed was, in this case, the presence of a particular element in the electrolyte seemed to correlate with the performance. And we started looking into this, and we realized that the presence of that element itself is not the issue. But the presence of that element is correlated with other effects that can be brought about by other elements that are similar. And if you're only looking at that kind of myopic description, then you could say, well, the best way to fix this is to never put that element in any of our electrolytes ever again. Just take that descriptor down to 0. And the cycle I should go up to infinity. But it turns out that that's a correlation that's not causal. The causal piece of this is really complex electronic structure facets that show up. So I guess I'm dancing around your question because it's a hard one to answer. Sometimes correlation is all you need. Sometimes not. But it's a tough thing to tease out. And I think ultimately, there's only so much you can say about what is truly causal. But the ultimate question I think for us is does the design principle reproducibly get you better results when you listen to it, whether it's correlation or causation? That's the key. Reproducing ability-- I definitely understand that. When I first came across AI within industrial manufacturing sets in it, it was seen drug discovery. It was seen pharmaceuticals. And they were honestly really early because they had so many compounds that they needed to process to work out which ones were having any kind of efficacy. I'm just wondering if the work you do within batteries is in any way informed by some of the work that we've done back in the day and on the pharmaceutical side? Absolutely. 100%. And I think the reason why the battery field hasn't gotten into this area of exploration earlier, I think comes back to this sort of battery 2.0 versus 3.0 idea, which was there was a lot of work to do on the electrodes. And as you said, the most pressing issue in the drug discovery space for a long time was around molecular engineering. That's only more recently emerged as a key bottleneck in batteries now, which in some ways has been an advantage because we have all this infrastructure that we can use from the pharmaceutical world best practices for model building. We actually regularly use some of these models. There's a framework called QSPR, the quantitative structure property relationship, or QSAR quantitative structure activity relationship, which sets guidelines for how to build these models originally from the drug discovery world, but is very useful for us. There's also-- there's tools, there's models. And I mentioned this 10 billion number, these 10 billion molecules that you or I can procure. So much of that infrastructure has been set up around drug discovery. So these molecular manufacturers are not really there to sell electrolyte molecules. They're there to sell drug molecules. And bioactive things and solvents and all these things. And so there's huge tailwinds for us in this space. And it's a very similar problem, not exactly similar, but very similar from the idea that I have a complex chemical system. I want to put a molecule into that system to solve one problem, but not to create others. And so it must be targeted at the problem at hand, but I must have some way to know that it's not going to break this thing down in this way, or it's not going to cause some reaction within the system. And so there's a lot of-- there's a lot of process over there. And that really works for our business. And when we explain what we do to investors and to partners and companies, we rely on that kind of analogy, right? To say, hey, you look at the development of the COVID-19 vaccine from from Biontech and Pfizer. And Biontech was essentially the company that designed to discover a lot of the the core IP, but Pfizer was the company that scaled it up. And happy factories and brought it to market. And in that comparison, really, we're the Biontech. We're sitting on our computers in Palo Alto, right? We don't have a factory, but we can discover the core IP. And then we work with our partners to actually scale these solutions up and make sure that there is real commercial impact. Well, that's a great analogy. So in terms of the actual work you're doing, in terms of the electrolyte developments, and I appreciate in the work you do, there's going to be a lot of stuff which is not for public consumption. For obvious reasons. I mean, can you give us some kind of maybe examples, just maybe some general principles in terms of the kind of work you're doing, the kind of benefits that you're able to achieve through the through the very interesting science that you do? Absolutely. So there's there's a lot of different sectors where this work can can benefit in each sector or needs something different. Automotive is an interesting place to start because that's where most of the market is today. If you look at the electrolyte composition of many of the today's commercialized electric vehicles, you will see that the majority of those electrolytes across those cells, across those companies are more or less made of the same 11 molecules in different ratios. So there's essentially this kind of playbook that we know. It's been successful. It's worked time and time again. But it's essentially created a cell that is sort of good enough, but not particularly great on any particular, particular access. So in automotive, there was really a drive for a long time to increase the energy density, to increase range. You know, the range is maybe still an issue in some particular cases, but I think really now what we're seeing is that range is becoming good enough and what is preventing us from reaching cost and performance parity with ice vehicles are the other parts of the experience. So in particular, it's a very interesting time. No one must go on a road trip and sit for 30 minutes to charge the car two times in one day or something like that. Class, of course, is also an issue for us to improve as well. But in the charge time case, what we've seen is that improving or changing sometimes even small amounts of these, the ratios or the identities of these molecules in the electrolyte can really change the fundamental properties that can allow for much faster charging without decreasing the overall performance of the cell in other ways. The goal of the industry, I think, really is to get down to like a five minute charge. Right. So about the same amount of time that it would take for you to get, to fill your gas tank or petrol tank. And we're really, I mean, we're really on the way to doing that, I think, as an industry. Those tweaks, again, to the electrolyte are essentially drop-in solutions. So you might have a couple of solvents in there that maybe one is problematic. So you can take that one out, you can replace it with another, you can add, you know, additives here and there. And actually what we're seeing is that electrolyte design, because it is a drop-in solution, can be an exception to design freezes that are really common in vehicle design, because you don't have to retool the factory, right. You just essentially drop a different liquid in from one day to the next. Anyway, so I think in the automotive front, we've been able to see, and others have seen, you know, big improvements in these kind of or physical properties like viscosity and ionic conductivity and, you know, peak power output through engineering these molecules in their concentrations. And you can see that across other sectors as well. You can make these improvements without, again, killing kind of the other aspects of the cell, which is historically been a big issue, you know, you improve property A, and now property B is, you come to the back, and you go back and fix property B, and then now you've changed property C, the kind of revolution around machine learning and AI, and it will just consider all of those properties at the same time. There's a couple of things, quick things, you should come out of that. You mentioned these kind of 11 molecules, essentially, what you're working with, these are the principal molecules. It's an actual limit there, or within the work you do, are you also looking at things that don't molecules that are within these 11? Did it actually outside that, where may, may offer something completely revolutionary? I mean, is that part of the landscape too, or was sensitive to really narrow down to no, we're going to, we're going to get the blend of these 11 right, and that's going to get us where we want to get to. Yeah, this is a very, very important distinction to make. So I should have been clear that we actually are intentionally looking beyond those 11. Right. Right. Right. And I often tell people, hey, we have about 30 properties that one can want to improve with batteries. And we have essentially 11, so 30 diseases and we have about 11 drugs. Right. So we should probably find some more drugs. And so we found some really interesting things. For example, we in our office, we have a little vial of perfume. And this perfume is actually a very electrochemically stable molecule that we discovered through one of these broad searches of chemical space. And we said, hey, this thing is, this would be great for high voltage cathodes. Let's order it up. Let's test it. Let's see how it does. And then we turned it turned out that it was on the market as a perfume. So these kinds of things happen, right. There's been a lot of really interesting cases of molecules that have been explored for some pharmaceutical application or another. We're working with one that I think has been tested as like an antidepressant if I'm not mistaken. Apparently smells really bad. I was told not to smell it. It's not like the perfume. This one smells terrible and it's toxic. So like don't open the wild smell it. So it's a really exciting time. And we haven't talked about this, but I should also mention a piece of what we're doing is around the generation of new molecules, which had never been made before. So there's about 11 molecules that have kind of tried and tried and true and tested. There's about 100 million or so that can be procured in reasonable amounts, you know, reasonable volumes for testing. There's about 10 billion that can be procured in any amount, even small amounts typically like, you know, maybe small micrograms, which is for biological testing and things. But then there's a nearly infinite number of molecules that could be stable and could be designed, which have, which is aren't in that bucket of 10 billion. And that's a really interesting area, particularly around generative AI where you can actually generate new structures that hit certain performance specs or certain criteria. And that's kind of separate from sort of modeling, you know, property modeling and screening. But it's all a part of the, I think of how AI machine learning will revolutionize this field of battery research and design completely. Well, that's been fascinating. It really has. And I suppose I should finish this by just asking from an AI on its perspective over the course of the next five years. I mean, where do you see your business developing, changing, metamorphosizing to take advantage of these enormous opportunities? Yeah, it's a great question. So really our mission is to be able to completely rethink, you know, electrolyte broadly. And to be able to design these high performance custom solutions for different chemistries, different customers, different use cases. And right now, a lot of the industry is based on traditional with the Mayan, although we are starting to see some silicon metal. And so in the short term, you know, our plans are around really getting our formulations into cells across these different kinds of chemistries. I think where it gets really exciting is where we see more chemistries come onto the market. And that's not just sodium ion, although, you know, that's probably coming and there's probably similar. But if you take a broad view and you look at how we are going to decarbonize the economy generally, there's tons of applications where the only way that we have to make something is to get it really hot. You know, for example, factories that make cement and steel, these are all based on burning things, getting a kiln, you know, really hot or oven getting really hot. There's really interesting work around essentially supplanting that process with an electrochemical process that can run at room temperature. And if you can do that, then suddenly you could take carbon out of every facet of manufacturing. And so you have to design all of these new chemistries and these new cells and some of which we've never seen before. And for example, there's a lot of work now on electrochemical cement where you're essentially running a calcium ion battery. the you know the breakdown of the battery which you know only the one avoid in this case is actually producing You know calcium silicate which can help you make cement and make it without burning in CO2 so The future for electrochemistry. I think is very bright and it's very exciting and I think You make you may accuse me of hyperbole but I think electrochemistry is how we're going to save the world Right and so our vision is you know we want to essentially turn the electrolyte design piece of this into something that is You know so easy to do you know you pair two electrodes to solve your problem and you got your electrolyte right there And now that cell is going into production and it's doing the job that needs to do so if we can get our our machine learning and AI and calculations to work right and then hopefully we can help bring that that future bear I could talk about this all day But I have to acknowledge that you've got a a trillion molecules to process I have to respect that I've really enjoyed this I really have Explanations have been really clear. It's allowed me somebody is not an expert in this field to really Understand a lot more about the kind of work you're doing and also the potential Particularly machine learning within the electrolyte space. That's been really fascinating. So Thank you Austin so much for being on the battery technology podcast. I really appreciate your time. Thank you. It's great chatting with you The battery technology podcast is a copy-rided GSC media limited production For more details and to reach us you'll find our contact details in the show notes or at our website www.batterytechnologypodcast.com , you