Jonathan Nistor, COO Blue Wave AI Labs

There's a way, a way such a better way today, today. The measure for us tell the world there's a better way, today there's a better way. This is Rod Adams and it's time for another Atomic Show. My guest today is Jonathan Nister, the Chief Operating Officer of Blue Wave AI, which is probably a pioneer. I'm sure there's a pioneer in using AI to help nuclear energy prosper. Jonathan, welcome to the show. Thank you very much for having me Rod and thanks for the kind introduction. Now you're welcome. Hey tell me a little bit about Blue Wave. We'll start with your company first. Okay, Blue Wave AI Labs, I guess not to be confused with Blue Wave AI up in Canada. Is an American company we got founded in 2016. The behest actually was one of the founders had a friend at the Department of Energy. And basically, he gave him some advice to quit his day job and he would introduce him to some utilities that were having some problems. Recurring problems with operating their plants. And the idea was to look into all of this immense stores of data that operating plants have to collect for a sundry of purposes, whether it's compliance for NRC or operating guidelines or a poor monitoring system. The point is they logged a lot of data for specific purposes. And then it just kind of got archived and siloed and sat there. But the truth is, there's so much information in that data. And so we really got started with the idea of helping the existing operating fleet leverage that data to unlock potential increase efficiency. And, you know, drive the cost competitiveness of the existing nuclear fleet way into the future. All right, so that's Blue Wave AI. Tell me a little bit about Jonathan Nister. What got you into the business of doing AI? What have you done before you got here? Well, you'd think that if I'm doing artificial intelligence in nuclear, that I would have a nuclear or AI background. But the truth is, is that I am a classically trained theoretical physicist from Purdue University. The founder also went to Purdue. We actually happened to have the same thesis advisor just separated by about four decades. And that's how we knew each other. But the transition from, you know, physics to really any engineering field or AI is not as far as one might think. So you got into AI from physics. Blue Wave got into AI from a national lab. What is it about the nuclear industry or nuclear technology that lends itself to having artificial intelligence, at least the artificial intelligence that we know today as a valuable tool, not just a new market? Well, you know, actually it is, it was a new market. So when we got started in 2016, you know, it was a different landscape at the time. AI and nuclear and neither were cool at the time. So it was a little fortuitous to get into this field at that time. But the truth is, is that nuclear has always been, I'll say, kind of behind the curve in adopting innovative technology. That could really streamline a lot of their cumbersome processes they have to do. I mean, there's a good reason for it. It's a very, it's a safety culture. They're very centered around safety and, you know, a methodology that's tried and true. But, you know, that comes with its downsize as well. Because there's, you know, a lot of advantages to, forget about AI for a second, wireless technology, having wireless sensors throughout the plant. You know, these sorts of things didn't exist when these machines were built decades ago. And they're, they're costly to implement now. And so that's a buried entry, costly because of needing things to be nuclear grade. But really, the truth is, is this industry has been behind the curve in a lot of respects, and they've been suffering or they had been suffering from, you know, the treatment given to some other energy sources that, they just, they were floundering and plans for shutting down. Now you know today, the landscape's completely different. People are talking about restarting plants. But we, we really thought that, you know, we could bring in, you know, AI to help optimize and to give increased visibility into a longstanding challenges that were, you know, costing these companies a lot of money in the past. I might not have worded my, my question with the leading nature that I was looking for. But when I talk about AI being a new market, the big news in today's world about AI and nuclear, is that AI data centers are consuming such vast amounts of electricity that they are causing the electricity demand in the US and around the world to grow, which is something that hasn't happened very much in the last 20 years. And so without growth in a market, there's really not much need to build new stuff, especially when there's a bunch of new stuff that was being heavily subsidized for, for it to grow. So there just was a room in the market for nuclear, which is actually the reason why nuclear plants were shutting down, not so much that their costs had gotten out of control, but it was that their costs were relatively fixed. And the price of electricity wasn't so fixed. It was being driven down by oversupply. But that's being said, artificial intelligence, as we know it today, often is machine learning, or based on consuming large data sets and using that to start drawing inferences. One of the things that nuclear industry has done, as you noted earlier, is that it is a mass enormous quantities of pretty structured data over the years. And that data has not been that easy to access. Seems like what you and some of the other AI companies are doing is changing that situation, at least making some big nibbles at it. There's all you have a lot to unpack in there. I agree with your characterization. The truth is, there's a synergy there, nuclear power in AI, and then AI also helping drive efficiencies in nuclear. And really, this is just the start. I mean, the large data centers, the hyperscalers, eventually, I mean, the electricity demand is only going to grow as we introduce new technologies, like personalized robots that need to be charged every day. The growth in EVs and stuff. The demand for stable, always on power, clean power, is only going to grow. And there's not a lot of energy sources that fit the bill, nuclear being the primary one. But you mentioned types of AI. We got started before CHET, was the household, nine before it existed. They had to give large language models, a Gentic AI. Now you hear superintelligence. You guys started when AI primarily meant more classical machine learning, artificial neural networks, transformers, technologies of that sort. And the truth is, is like you said, these plants that have been operating for 40 or more years have a huge repository of structured data. Having said that, it's not big data like what LLMs are trained on. It's not, you know, every word ever written on the internet. But it is a large set of data that's, you know, been untapped for so long. And one of the challenges when we got into this industry and actually still is a challenge is really finding out where that data is siloed and getting access to it. Not because our utility partners don't want to grant access, but because often they don't know exactly how to retrieve it or where exactly it is. And unfortunately some data going way back, you know, has been lost. Over time, you know, the data historian archives data different, you know, sampling rates and down selects it as time passes. So very old data sets are less dense than the newer data sets being being collected. But really the applications in nuclear are limitless from predictive maintenance to, you know, fuel cycle optimization, work order optimization, better predictability of outage dose rates expected going into the outage. All of these things matter a lot of both from a safety perspective or workforce perspective and cost perspective. I mean, knowing how to accurately assess the dose, the radiation dose that the outage personnel are going to see going into an outage is very attractive idea because then you can properly allocate equipment shielding PPE material have the right number of workers there. And you don't have to, you know, introduce all this rework when your estimates are way off. And that's kind of true across the board. The rework is what gets these guys. And so reducing uncertainty, eliminating margin and that rework is the game that we play. According to your website, one of the tools that you have successfully introduced has been some things that help utilities optimize core reloads. Can you tell us a little bit about that? I'm not sure that everybody really understands how important it is for a reactor operation for each cycle to be calculated uniquely to make sure that the fuel is used. As efficiently as possible. Yeah, we haven't talked about that. I don't know exactly who your audience is. I'm going to pretend like some people don't know, but I assume everybody knows the difference between blood and water reactors, BWRs and PWRs. They are different, you know, machines entirely. And, you know, we've done a lot of work. And, you know, currently in the majority of the BWRs and the US fleet are customers of ours. But boiler water reactors are unique because they, you know, the moderator, the coolant, which is water, is allowed to boil, you know, when it's being pushed up from the bottom and the quarter of the top. And that steam that's produced in that boiling is directly what, you know, drives everything. But that two phase flow between water and steam is, you know, I'll put on my physicist that for a second. It is not easily modeled. The physics of two phase flow is not well understood. And even high fidelity simulations can't get it quite right. We just really don't understand the physics of two phase flow, turbulent flow, things of low nature. So, every boiling water reactor in the US is operating on a 24 month fuel cycle. Oh, Creek is the latest to join. They were on an 18 month fuel cycle. But basically what that means is, every, you know, for 23 months, the reactors expected to run it and 100% of its rated power, except perhaps a short one to two month one to two month cost down period at the end. And then they shut down for a month, you know, some are very efficient and do their reviewing outages in two weeks and some. a little longer, but say they're shut down for about a month. In that month timeframe, a third of the fuel is discharged, the oldest fuel, and a batch, a third of it is brand new fuel that's loathed in the core. And all the bundles are shuffled around in a very specific way. But it doesn't happen by magic. The process of designing that core starts about 12 months before the outage. And it's a choreograph between, depending on the utility, between their internal core design team and the fuel fabricators who do some of the design, or license, and work, safety analysis. And that 12 month process is answering a bunch of questions. What does the fuel going to look like? So pin by pin in the fuel, what's the enrichment? Where do I put my gadolinia? How do I arrange those bundles? Once those fuel bundles, once I lock in the designs, amongst all the other fuel that's going to be in the core, how do I operate the two-year cycle? Unlike a PWR, BWR uses control blades quite actively. They have maybe half a dozen primary sequences with control blades, and they change them every few months. And they adjust core flow. These are all the sorts of design parameters they have to work with. But going back to my thing about two-phase flow introducing uncertainty, the tools they have to do this design work, these neutronics codes, these core simulators, don't have the fidelity to give them accuracy to infinite precision. They have maybe plus or minus 10% in some key variable, or worse than others. And so when we got started, there were several parameters that are either not calculated or calculated with a large uncertainty band on it. That alone introduced a lot of additional margin that had to be engineered in. And how that translates to is basically loading more fuel into the core every two years than you otherwise would have to. Or if you don't load enough, you run into operational challenges throughout the cycle. And you may not be able to deliver the power you've committed. And so there's a power generation hit. And in either case, these uncertainties were amounting to multi-million dollars of fluff, access fuel, or loss in generation revenue. And so when we got started, we started with a problem moisture carryover, which for some plants is it was a limiting factor. They want their steam that they're producing to be as pure as possible. No water droplets mixed in this steam. They want it to be dry. The problem with water is that it carries over things dissolved in the water. They call it 60 or chlorine, which would form hydrochloric acid. And these things can be tough on the internals of the machine. They can erode internal surfaces of the turbine blades or the valves. You can get a efficiency loss in the electrical generation from water droplets. And then obviously the call was 60 and introduces excess dose throughout the balance of the plant. So you want the steam to be as dry as possible. Some plants are very sensitive to excess moisture being carried over with the steam. And the one fail-safe way to eliminate that problem is load more of fresh fuel every two years than you otherwise would have to. And so you're essentially spending money to burn away that excess moisture. And so one of the first problems we tackle is getting a model that could predict what the moisture carryover would be as a function of all these design variables that the core designers have at their disposal during core design. And in this way, they could actually design an efficient core from the perspective of moisture carryover while trying to minimize or mitigate raising the number of fresh fuel assemblies or the enrichments that they had to load into that core. And then after that, while we had several additional models that do different things, but they all have one thing in common, which is, it was very costly from either design perspective or from fresh fuel or an operating perspective if they got these parameters wrong. Initially, blue way that spoke is done, I'll say technical use of the artificial intelligence tools. In other words, actually using them to make machinery perform better. Have you gotten into any of the uses of AI to get people to perform better in document preparation or reviews or those kinds of things? Yeah, that's a great question. So on the generative AI side, there is an incredible potential, not just in nuclear, but for our entire economy, to really boost productivity and efficiency. And we have an internal mantra or motto, which is minimize the mundane. And generative AI and LLMs have the ability to really do that, especially in nuclear. And so I'll give a couple examples. Every, and I'm not trying to pick just on fuel, but in the fuel space, every core that's designed, once they get in that process, the fuel fabricator will send a multi-hundred page technical document that basically is the instruction manual for what that core is going to look like, the specifications of the fuel, how things were modeled, what the results are for all these different limits and things that have to be tracked. And some human has to review this entire document, not just skim it, but actually have to read all 250 pages, have to go back and reference all their internal calculations at the plant, and reference their internal methodologies, and check that what the fuel vendor is sending them correctly aligns was what they designed their core to do and everything that they had taken to account. This is the long process. It might take a senior engineer several weeks to go through that review process and basically accept that document and certify that yes, this is what we intend to do, everything in here is correct. An LLM with the proper amount of prompt engineering and technical logic baked into it can do this in a matter of minutes. Where a human basically just has to look at the answers, look at the citations that the application is giving them from where the answer was found and check those. And so you can eliminate three weeks worth of work within juicing just this tool. And that's just one very specific use case. In licensing space, there are so many things. I mean, the cost of licensing activities is huge because it's very time intensive. It takes years to assemble the proper document to go back and forth with the regulator and get some licensing application submitted. A lot of that time is doing research. And so if you look at what LLMs are good at, it's good, they're good at sifting through large volumes of data and pulling out relevant information, doing search, doing research, and answering questions that an engineer may have that otherwise would take them hours to find the answer to. And so these are the sorts of things that we're finding an appetite for within nuclear. And we're trying to build the proper products and lightning speed to address them. On a use case by use case basis, we don't want to turn over a generalized chat bot that doesn't have any guardrails on it. We really are trying to do this intentionally so that every use case is well understood and the proper logic is engineered into the full pipeline. And you're not just given an unfettered and well done to talk to them. How do you handle things like cybersecurity and those kinds of concerns? Many people think of our design with models as something that runs on the cloud. What posts a little bit about how you handle that? Yeah, so obviously that's cybersecurity is a huge concern. I mean, it always has been points of vulnerability at our growing as new technologies introduced. Large language models being another one. There's a lot of, especially for nuclear, there are a lot of specific considerations to take into account. Their information is ECI information, export control information, you have to make sure you're not sending it overseas or putting it in the hands of non-US persons. And so where you house the information is very important. And so there are clouds, like AWS or Azure, and then there are Gov clouds, government clouds that give the compliance necessary to store some of this protected information. The big players like Amazon, quite frankly, the security that they bring with the services to provide is substantial. What you have to make sure is that you're not using some open, large-language model that's freely connected to the internet, where the data you feed it is available and open to other people to use for training or for what. And that's the big thing, is making sure that customer data remains their data and isn't exposed outside of an ecosystem. So there are things like private clouds, kind of an oxymora, but there are private clouds that allow for lighter versions of these large-language models to be deployed on, so that you still get the benefits of the cloud computing infrastructure, but you are dealing in a private instance where you don't have to worry about data, leaking out. You all have learned blue wave has been in business and AI since 2016, so you've got a big old grandfather beard growing here. That's nine years for a technology that's changing every six months or every two months or whatever. Have you run into any other competitors in this space or any other potential partners? Using AI in the nuclear world. Yeah, that's a great question. I had say that we were spoiled in the early days because we had no competitors, but we're also an island and I'm very keen on collaboration. So you mentioned both sides of that. A competitor is just another way of saying a potential collaborator. And quite frankly, this space is large enough to handle and really benefit for many organizations doing this sort of thing. And so in this space, there's three classes of competition or organizations doing this sort of work. There's other private companies and there's a few startups that have come out in the last year or so. And then there's one more established company, nuclear and that's been around a little longer, definitely more mature than some of these other startups. I would say classically we're not competitors. There's different niches. We are very heavy in engineering, machine learning, core physics and what AI can do for the primary side of the system, of the machine, whereas a lot of these other companies are exclusively focusing on licensing or corrective action programs, that hasn't been our bread and butter. So the space is large enough. But getting back to what I started, other private companies. There's a few out there that I'm sure more that are going to be coming out of the woodwork every year because this is a growing field. But the other classic competitor would be the traditional large fuel fabricators, the fuel vendors. There's only a few. There's a Westinghouse, GNF, and a frame of thumb for the most part. They have a huge advantage in one respect in that they typically have access to or can get easy access to the data that is needed to develop a lot of these models. And that's always the challenge. Nobody can... The average Joe doesn't have an operating licensed reactor in their backyard to go generate data. But the fuel fabricators typically have access to a lot of the data. And they're increasingly trying to get into the services business because that's seemingly where there's more margins and more revenue. And then the third class, I would say, I don't know what to call them, but the national labs and these organizations like EPRI and POO, all of them have large resources and also significant access to the relevant data. And depending on their structure, they're not supposed to directly compete with private organizations, but doesn't mean they don't. So yeah, it's a competitive landscape. It's only going to get more competitive. But my singular focus has been to deliver very accurate and services, very accurate models that also teach and give a level of explainability that is expected from a nuclear perspective. Everything that's done in nuclear is intentional, like I said in the beginning. And there needs to be a basis for every decision that's made. And so if you're going to start doing AI, you need to really bring to it a level of rigor that can let you explain the inference or the predictions being made. That's key. Yeah, one of the things that people complain about would publicly not known AI systems is they don't really know how it makes its decision. They don't know why it made the recommendation it made or what input gave it the stimulant to go do that. And as you say, it's nuclear. You've got to have the basis. I'll never forget the operating manuals that I was familiar with when I was on submarines. They actually had a two column format. One column was these are the things they're supposed to do. The other column was this is why you're supposed to do it. And pointed you to the right place. You mentioned how well AI tools can can do some research. Another factor that comes into play for nuclear documentation is tracing the references and making sure that things are documented as to where they came from. Sometimes you have a reference that pointed to another paragraph pointing you to another paragraph. And finally, you get to the statement that you're trying to adhere to or the requirements that you're trying to adhere to. How do you set up tools to do that? That's a very good question. I went back up one second. You made a comment about decision making. And I will say that I believe that, you know, a gentle AI, you know, things making decisions on their own. These sorts of systems are going to be prolific in our everyday life in the, you know, not too distant future, whether it be like, you know, personal robots or robo taxis or things driving a car. But, you know, I don't think that that's going to take place. And nor should it. And personally, what we do is our belief is we are providing assistance and additional information and additional visibility to human decision makers. I call it, you know, assisted or augmented decision making, but the, the onus is still on the human to make the decision with all the information and all their existing processes in place with the additional supplement provided by our tools. And I, you know, we've, we've, we've broke this subject with the, the NRC and I believe that their, their comfort zone eventually will be in the same area where the various levels of automation introduced or autonomy introduced with AI assisting human decision makers is going to be the comfort zone. And, and where you have a completely autonomous system, you know, maybe one day with next generation of actors, the small and the smaller ones, the micro reactors, they can be control autonomy. But these, these large machines, the existing, the humans in the loop is the key. Okay. Fast forwarding to your, your question, which at this point I forgot. Oh, is it? Okay. Now you make a mean remember question was that's not the way to go. It was. I remember now I remember I want to hold you to. Okay. So your question was, how do you deal with the, the, the progeny of the information, you know, it needs to be correctly cited. And like you said, you may only have like a five page technical document, but really what it does is has like 30 spots where it has some mentioned to references incorporated their end or whatever. So that takes you to other tones that are much longer that also have to be appropriately reviewed or, or at least at the disposal of an AI to make a decision. And so in the licensing space. So you started to get, we've, we've, we were taking these, these projects one use case at a time. And, and the one that's most mature is this tool to help with 50 59 screenings. So what 50 59 is, is whenever you know, you apply it, you know, someone is proposing to make a change in the plant. It could be a physical change like replacing something or it can be a change in method or methodology. So whenever it changes proposed, there has to be a, by an assessment done if that change is going to affect the plants operating basis operating license. And so what the 50 59 screen allows plants to do is study if that change will affect their licensing, their license or not. And if it's not, then the screen out and they can perform the change without a license amendment. And that's something that, you know, everybody wants to be able to do a license amendment is time consuming and costly. And if you can avoid it, you would like to avoid it. But you have to make sure that the change you're making would not affect your licensing, your license. And so this requires a lot of research. You have to research your entire licensing basis for the impact of that proposed change or activity. And you know, you don't just look at one document. You have to look at many documents. And so this is where like I'll call it licensing logic comes in. The approach is, you know, the latest and greatest large language model can be the engine to do the kind of interact with the system. But and you and rag can be used to look up information from a dedicated set of documents. But you have to make sure you have the right documents available. And you have to make sure that you're not you're properly linking the prompt or what the engineer or the user is asking the question that's being asked is actually generating appropriate set of queries to go out and search your license basis. And this is something just I can't tell you there's a, a prescription to how to do this. It's it's a try and fail and try over again and incorporate logic as you go and having expert reviewers and testers available to you to really give the feedback you need along along the process. So it's very, very much involves SMEs such as better experts and their ability to assess the efficacy of the result. If it was easy, everybody would be doing it right. Very true. One of the things you mentioned was that nuclear is only one of many activities humans do that may be assisted by artificial intelligence. My view is that nuclear is one of the highest payoff places because worse at such a risk averse do nothing or may change as hard as possible culture that AI seems to be really suited for breaking through some of those historical slow moving and many human hours for doing the smallest thing. What do you think? Yeah, I agree. It's on one level of nuclear is low hanging fruit because like I said, they are well behind the you know on the adoption curve compared to other industries. It's not like inventing new technology. It's just getting the existing technology that's you know well proven properly incorporated into the nuclear workflow. And there is so much room for improvement. And quite frankly, there's there's a couple of challenges nuclear has one of them is a relative basis of shrinking workforce and aging workforce. And this is probably not a PC term, but there's a term for like the old retirees. They're called grapebirds. And when they retire they they take with them 30 years or more of of OE of operating experience and simple things like where certain information is located can take a junior engineer weeks to read you know find out whereas if that person had retired they could have pointed into it right away. And so codifying a lot of that mass exodus of of expertise into AI does assist with the junior workforce is important. Being able to be lean and do everything that needs to be done in a plant with the same number or fewer people is very important. And I'm quite frankly nuclear engineers that are that are graduating nowadays you know are finding other things more sexy they're finding advanced reactors a lot things more sexy than the existing fleet. And so there's a huge huge potential still left in nuclear to bring AI into it. And one area we haven't talked about is predictive maintenance that's probably one of the biggest areas there is there's a challenge there which is nuclear doesn't like failures. And generally it doesn't run a component until a failure it'll be replaced or maintenance will be performed on a well before a failure occurs and so there's not a lot of failure data. And so it's a lot of study distribution. But there's ways around that and I think eventually and not to just feature you'll see a huge improvement in efficiency and cost associated with maintenance and reliability based off of using AI in that space and pretty reliability and condition based maintenance. And so at the time that the NRC gathered the industry together for the regulatory information conference I think if I remember correctly AI was a pretty big topic. How does the NRC view the use of AI and is the NRC doing anything itself. Yeah, this is a two part question part one is how does the NRC view their licensees adopting AI and the other one is what is the regulator doing to bring AI in house to help you know with efficiency on their end. And honestly I've been rather impressed with. forward-looking approach for everybody we've interacted with at the air seat. There are internal efforts at the inter seats to streamline a lot of their time-consuming review processes with AI. They're slow to adopt things, they have to evaluate things, and I think it will take some time. But I think it will happen. And as an example, they actually contracted with us with Lou Wave to build them an AI system to do autonomous BWR equilibrium core design. Because they're not in the business that do important designs, but they need core designs to do subsequent analysis, thinking of us as a one-night. And I was taking a very limited team a long time just to get to that starting point of having an equilibrium core design. And so we've been under contract with them for a year and a half to deliver a weight of autonomously autonomously designed equilibrium core for generic boiling water reactors. There's tons of areas where the energy can kind of adopt AI more than just coal pilot. And I think that they are serious about evaluating what options they have and doing it. Now on the other side, they have put out some public papers. They've done some studies to understand where the industry is at and to try to be ready to receive an application that has a huge AI component contained within it. They're not there yet. They're not ready at the moment, and at least at IIS Nation, to properly evaluate an application that comes in their door that is AI based. But I don't think the licensees are also yet ready to send that application their way. So hopefully the RSE does get get ready in a time frame that makes sense about the end of next year or so. Because it's coming. I think the area where AI like I said, we consider our tools to be providing assistance to human decision makers. But there are places where you could interject AI and it's a little more integrated into existing systems. And that that additional integration is going to require NRC approval. And we're actually in the process of working through some of that now with a couple of new products that are in kind of the beta stage. And so, you know, we will be in a position by early next year to want to integrate them into existing core monitoring systems. And that that might require NRC involved. And they're going to need to be prepared to evaluate that request. Are there any working groups maybe under the auspices of the nuclear energy institute or the American nuclear society or effort even working to help the NRC develop procedures or policies associated with using AI and some of the addressing some of these maybe the slowness or the map of slowness. The reluctance that you talk about. Yeah, the the NDI has an innovation task force. And one of the one of the big things on that task force is AI. And you know, they've done some industry polls to see where their members are at with AI projects. And there's been some you know that that's been presented to the NRC to make them aware of all the use cases and areas of interest for the utilities. I wouldn't say that there's any single you know grand industry initiative that's like kind of you know become the standard to go through. But there's been overtures to the NRC from organizations like Blue Wave for many I and I know that there's interest by by these organizations EPRI, EPRI, to bring AI into you know a tool into place that they can give to all of their members. Now that wouldn't be something that needs to be regulated but it's really getting all these organizations thinking deeply about everything you asked. I mean the cybersecurity considerations what would it take for you know their members to be comfortable with any or of their data being fed into these systems. On one level this is still the wild wild west and still early days. Utilities themselves are going through a learning process of spinning up their internal AI strategy what they want that to look like. Where they're going to introduce AI within their organizations what they're going to do themselves what they're going to rely on vendors to do and that really started kind of an earnest middle last year and is going through this year. Probably I would say the beginning of 2026 is where you're going to start to see a lot more real adoption of stuff because I think the leg work would have done at that point. Double specific questions about Blue Wave how many people do you have work in there? We are 22 right now you know when we got started we were like four we are actively hiring and growing we're in a mode right now to try to really acquire specific expertise in different areas whether it be licensing or you know engineering services or something else so we're less active in hiring AI data scientists and more active in hiring experienced nuclear engineers at the moment but right now we're 22. Okay the next question is I saw on your website that your headquarters is in Naples, Florida or is that where most people are you distributed working from home kind of company? Well you know we're incorporated there are our main working headquarters for the engineering services is actually in the West Lafayette in Vienna. It's right outside of Purdue University but like I mentioned one of the co-founders went to Purdue I went to Purdue and in fact we have a growing number of folks now that have gotten a degree of Purdue in some engineering or physics or math background and quite frankly being that close to universities is great for a couple reasons I mean we actually collaborate with the nuclear engineering school at Purdue it's a great it's a great resource for for good talent coming out of school that that's our primary workforce but as I think with most organizations we're becoming more and more hybrid and dispersed as time goes on and was one of the great things about working at Lou Wave is we're very flexible we're a small team we're a close team we want you know as we grow you know that flexibility will remain and so people need to move somewhere else for spouse for spouses opportunity or something else but we work very efficiently even even if we are distributed. You mentioned to a attentive focus on boiling water reactors what is your do you service all the boiling water reactors you serve some PBRs what's going on there? So right now I think roughly 60% of the boiling water reactors our customers of ours obviously we're eager to get all of the boiling water reactors in the US on board with with our services on the on the pressurized water reactor front we were taking those you know generally speaking the way we work is we we we go in with a pilot project it may not be none of our none of our solutions are kind of off the shelf in the sense that every plant is it is like a fingerprint they're very unique and you know one PWR another PWR not the same machine at all and same thing with the BWR and so the problems at each place are very unique we've done some pilot projects and some very you know deep root cause analysis at some of the PWR is out there and if some of these things turn into to you know a tool that can be productized and sold to other organizations great if not then we've solved at a pain point for one of our customers and we're very happy with that but yeah we are moving into the PWR space the lowest annual fruit when we got started with BWR is because there's a lot of room for efficiency a core design but when you get to the licensing activities when you get to the balance of plant stuff a quick and reliability condition based maintenance everybody everybody you know is very similar in the sense of having same at the same there are short problems at hand points and so the things we're developing are broadly you know going to be technology agnostic as far as what kind of machine they're applied to and what kind one final question for you the NRC has been given marching orders for the last five or six years to become more efficient and better at doing their job you are in a fairly unique position and servicing a number of different licensees have you heard much commentary on how things are changing I honestly I can't say that I have I've heard a reasonable amount of commentary that things haven't changed much you know but I do you know what I love about nuclear is a very small community and on one level everybody is aligned with the common goal and you know I think the regulator you know is sincere and wanting to become more efficient and I think the licensees are hopeful that that's true and eager to help if they can but I don't think there's been a lot yet in fact I mean I you know if anything the trends are still you know certain things are taking longer power upgrades license a better requires for power upgrade you know just over time is taking longer to do the same sort of thing you think that in most industries repetition translates to increased that's not true for the for the US regulator but hopefully that will change all right Jonathan we've been chatting for a while I really enjoy learning about what you guys do and learning about some of the improvements that you're bringing to the your customers your licensees that need a help and getting more efficient do you have anything you'd like to talk we didn't talk about well you put me on the spot now like I did before do you well you're supposed to be on the spot you are in fact the guest here so guess get to have their own say now and again I guess I'll just end with saying that honestly this is a very exciting time it's an exciting time for nuclear because they have found a champion for that's hungry for power really you know the landscape that was dreary questioning whether or not the resources and investment to do a subsequent license renewal I mean these sorts of questions are starting to dissipate and all I can say is that's a very exciting time because we have a host of tools already deployed and and many more under development that aim to just continue the snowball effect and the direction of efficiency and popularity of nuclear and so I hope that all the other organizations that are getting into this business can find the same sort of success in driving this industry forward. Terrific thank you for your time and enjoyed chatting. Yep thank you there's a lot of fun. I've been speaking with Jonathan Nister the Chief Upper Inaugher of Blue Wave AI labs a company that focuses on using artificial intelligence and machine learning to process large set of data for the nuclear industry and their nuclear customers to try to make required activities more efficient and less costly. I hope you enjoyed this episode of the Atomic Show this is Rod Adams I've been your host for the Atomic Show for more than 15 years as the publisher of Atomic Insights I've been speaking with experts in analyzing nuclear energy for more than three days. About half a decade ago became clear that investing in advanced nuclear developments could provide exceptional returns. Successful investors, based on Silicon Valley, agreed. While I'll continue to produce new content, Atomic Insights is now a part of the Nucleation Capital. A venture capital fund that specializes in nuclear and nuclear adjacent emerging companies. As a managing partner at Nucleation Capital, I'm expanding my access and digging even deeper into nuclear energy companies. My partners and I are working hard to select ventures with extraordinary promise and success. Their building the advanced nuclear sector and helping to expand our clean energy options. We're investing our own money in diversified portfolio of ventures and enabling investors like many of you to take advantage of our diligence and selection processes. We use an established software platform that allows us to offer full featured investment subscriptions starting as low as $5,000 per quarter. A four-quarter subscription will get you exposure to between four and eight ventures. We encourage longer subscriptions with greater diversification by offering a discount on our fees for eight quarters or more. If you are an accredited investor who wants to learn more about how you can participate, please visit our website at NucleationCapital.com. That's Nucleation Capital, all one word.com. Our fund and all of the information you need to subscribe is available online. You can also subscribe to our newsletter, Nucleation Insights, and join our pro-nuclear investor network where you get information about syndicated investments in selected companies. Limited partners in the fund get discounted access to syndicates, but they are open to other angelless members. If you would like more information, we're happy to help. Please spread the word. Thank you.