Subscribe
Share
Share
Embed
With World Cement’s Optimisation 2026 virtual conference on the horizon, it seemed like a good opportunity for this episode of the World Cement Podcast to takes a look back at the AI panel discussion from EnviroTech 2026. This wide-ranging discussion sees Senior Editor, David Bizley, put questions to three AI experts: Vivek Juneja, Head of Engineering – Cement, alcemy; Noah Miller, COO, Gigaton, and Scott Ziegler, CEO, CemAI.
So settle in and enjoy the episode as our expert panel explores the role of AI in transforming the cement industry.
Creators and Guests
What is World Cement Podcast?
The World Cement podcast: a podcast series for professionals in the cement industry.
Hello, everyone, and welcome to the World Cement Podcast with me, your host, David Bizley. To celebrate the recent launch of ticket sales for EnviroTech twenty twenty seven, which is taking place in Madrid on the seventh to the tenth of March, I thought it would be a good opportunity to revisit some of the interesting discussions covered earlier this year at our London show. Over the next two episodes, we're going to revisit the AI panel discussion, which featured expert insight and commentary from Vivek Juneja, Head of Engineering in Cement at Alfie Lloyd- Noah Miller, COO of Carbon Re, which is now known as Gigaton and Scott Ziegler, CEO of SemAI. Settle in, relax, and enjoy the discussion. I just wanted to take a moment to remind you to register for World Cement.
David Bizley:It's free of charge and gives you access to the latest issues of World Cement, both in print and online. Every issue comes packed full of regional analysis, technical articles, project case studies, and the latest industry news. Simply head over to worldcement.com, click the Magazine tab, and register today. It's as simple as that. Happy reading!
David Bizley:Let's kick things off with a bit of an introduction. So starting at that end, Noah, tell us a bit about, well, who you are and what Carbon Re do in the AI instrument world. Good afternoon, everybody. Thanks for sticking with
Noah Miller:us as we get closer to network drinks at the end of the day. I'm doctor Noah Miller, COO at Carbon Re. We're building fully autonomous AI control systems for the pyro process, and that's backed up with five half a decade of experience controlling the pyro process in our customers across the world, reducing cost and carbon. My role as COO is really implementation, making our control systems work in the control rooms across our customers. But before that, I was also head of machine learning, so I've been involved in building the whole stack of the control and modeling systems that we implement today.
David Bizley:K. Thank you. And, Vivek, same question to you.
Vivek Juneja:Yep. I work at Alsumi. I lead the cement engineering at Alsumi. I represent the team here today. Alsumi, if you have already, heard about it in our the last presentation, by, mister Podcast, who we really work closely at the Simpler OWAG group.
Vivek Juneja:We work from cement to concrete and help optimize the strength and reduce the clinker factor over time, and, I think that is the biggest contribution that we are doing for our cement producers to reduce CO two emissions. Our main role is turning reactive optimization to proactive optimization through machine learning, and I think we're just getting started. So I and my team have been working at different stages of our cement product line, so working from ingesting large amount of data that our cement producers generate, making sure they are trained with our machine learning algorithms, and then they get really high quality and performant predictions from our side. Thank you. And Scott, same question for you.
Scott Ziegler:Yeah. Thanks, David. So SemAI offers really off the shelf solutions for process optimization across the entire plant and predictive maintenance from the quarry to the to the load out. We are and have been from the beginning cement focused. We are born, bred, raised through the cement industry.
Scott Ziegler:The developments have all been done with the cement industry as well. Myself, I've been in the industry for twenty years and all of our employees as well share in our background on the cement side of things.
David Bizley:Excellent. Thank you. Well, going back down to that end of the row, starting with you, Noah, I mean, Berkham mentioned or sort of described AI and digitalization as almost like an unsung hero in the the decarbonization process. How fair do you think that is to say, and what would your estimate, I guess, would be of the impact that decarbonization can play in enabling the decarbonization of the cement industry?
Noah Miller:I think it it's critical, and maybe the nuances, I think people underestimate the process challenge that decarbonization is, all the things that we need to do to decarbonize, whether that's alternative fuels, whether that's different raw materials, whether that's clinker factor reduction, those are all process control problems that increase a huge amount of instability. And so where AI comes in is by reducing that instability and allowing those strategic objectives to be hit through those processes.
David Bizley:David, saw you nodding along. Do you have
Vivek Juneja:Yeah. I mean, we as humans, overestimate in short term and underestimate in long term. Right. And I think the same thing applies for the greenhouse effects, you know, reducing that over time. I think to Noah's point, one of the things to look out for is these some of the initiatives need to be started having a long term goal, but having clear short term milestones.
Vivek Juneja:So a lot of the, opportunities where cement producers have is to have a clear RONI that often our customers have gotten under for five months, but keeping the long term goal consistent because a lot of it, as you heard mister Burden saying, is culture transformation, making sure that people are comfortable with that change, and that takes a really long time. But definitely we have seen from our customer conversations and observation that this is slowly affecting the reducing the emissions on carbon dioxide.
Scott Ziegler:Yeah. I think I'll I'll put it succinctly. I think that AI is the lowest hanging fruit out there right now, for for decarbonization. From cost to time of implementation to ease of implementation, it is by far the I'll call it the easiest one to take the first step in. It is not capital intensive.
Scott Ziegler:And again, I think it is has been demonstrated enough, across different, parts of the process to to really show the value very quickly.
David Bizley:Okay. And in that case then, speaking of low hanging fruit and what has been one of the main barriers holding producers back from adopting it and reaching for this low hanging fruit?
Scott Ziegler:Yeah. If I'll, if I'll, I'll start because I'll take, some of the things that, that, that Birkin mentioned, which is the tremendous amount of data that is available right now. It seems to be a concern that am I prepared? Am I ready? Do I have enough information?
Scott Ziegler:Do I have enough data? The short answer to that, and this is not a selling point, but the short answer is there is enough data. There always is enough data to start. Even if it is something small that seems to be a big pushback. We do compete against some of the capital, of the spending when it comes to whether it's process optimization or maintenance any of the other items that you look at as far as AI goes at the Cement Podcast, we're competing for that same funds.
Scott Ziegler:To Vivek's point though, the return on investment is so quick and it is such a fast turnaround. I think it's almost doing yourself a disservice not to get started.
Vivek Juneja:Absolutely. Think I could add to Scott's point. I think, the infrastructure required to make sure the data is clean and the health is quite high, used to be very high maybe a few years ago. I think with all the advances we are seeing on language models, agentic AI, we have seen some great improvements where machines can help and augment the improvement of the quality of data. So the barrier is only getting smaller and smaller over time.
Vivek Juneja:I would say one other thing, the barrier that comes from our experience has been the mindset change. Building confidence, these algorithms, especially when you're building closed loop systems. As humans, we take a lot of pride in our expertise, and that's never gonna go away, but I think these, the closed loop systems and our implementation has shown that, the systems are get smart getting smarter and really powerful on a day by day basis. So having the kind of the mindset that the shift is happening pretty quickly and being able to build capabilities over that as humans will become more of a decision makers and augment those capabilities to over time. I think that is what is, I will say, another barrier for adoption.
David Bizley:Okay. And sort of following on that human point, have you found, you know, any ever faced any sort of skepticism potentially from plant teams that might be worried about things like job security or the results that the software and the process is going to produce? And if so, how do we overcome that?
Noah Miller:I I think that was what what I was gonna talk about for the last question, so that's helpful. I think what Vivek was saying, that trust aspect is really key, in two ways. We saw there was the first question on the last panel was where should humans be involved? That's always, part and and, you know, write down to the operators. They want to understand exactly what the system is doing and why so they trust it to run for longer.
Noah Miller:And what we found is that explainability is so important. We can show them why is the control system making the decisions it's making. They can feedback on that. We can build that trust very quickly and go from they want to turn it off if it's difficult to they want to keep it on if it's difficult. But I think the other aspect of trust comes, unfortunately from a lack of trust in AI providers.
Noah Miller:A lot of people at this point have had a failed project, have been burned by someone who didn't understand cement, who didn't really know what they were getting into with industrial data, who came into their plant promising the world and didn't have any results and experience to back it up. So I think the other side of that, there's trust for the operators, how's the system working? And then there's trust for either management in plant or management across the group of, are you working with someone who has proven results in cement, who deeply understands your cement, understands cement as a process, and your nuances, and that can build the trust in both sides.
Scott Ziegler:I don't think it's as much a concern of displacing folks either. We saw earlier and started to use this human in the loop kind of concept as well. You know, we're not, you're not always displacing someone. That's never the end game. That's never the thought process here when it comes to everything from model fine tuning, new equipment, new sensors, updates to the process.
Scott Ziegler:On top of that, as our colleague from Sempra was mentioning, this is attracting new talent. This is new talent that's not been in our industry before either. And that's a very, very, very positive thing. When we think about the retirements, we think about the aging out of a lot of the workforce. It's still such a good opportunity to take that next step to use a lot of that knowledge that's already there as well.
Scott Ziegler:You know, if you're on on on a process optimization, the opportunity to run it in open loop and use more information and and let your control room operators make the decisions on the maintenance side of things. You're still going to have to replace equipment. You're still going to have to do the basic maintenance that's there. So this is less of a reduction of human, but keeping more of the human in the loop and more engaged from my perspective.
Vivek Juneja:I wanna quickly support that. I think, I wanna quote one of the comments we heard from our customer, Cecil, where they talked about having a lot of these tools in their plant team's hand, almost having them superpowers, And the role of the plant teams is augmenting a lot of the capabilities we are building by extending it to the, for example, the unsolved problems around that. So when the problem of strength predictions is being solved through closed loop, the staff is retrained and upskilled to tackle the other hard parts of it. So again, we will start seeing a bit of shift in the type of work rather than complete displacement. And I would also second what Noah mentioned, a lot of the trust comes through transparency.
Vivek Juneja:So again, at Alsemi, we sometimes take a lot of time to explain how, the machine learning really is working behind the scene to build that confidence, and we have seen explainability and transparency really helps to build that trust for a lot of the people in the teams.
Speaker 5:Yeah, maybe
Noah Miller:just Does anyone here know a process engineer who says they have too much time in their day or that running their plant is easy, right? There are every process engineer I've ever spoken to has a list, the length of their arm of things they need to do to make their process better that they haven't been able to do because they don't have the time, they don't have the human resources to do it. So the more we can help them and augment them, the better they can make their process.
Speaker 5:Scott, did I catch you?
David Bizley:Oh, no. Okay. So with Vivek, for you, if a cement producer wants to build their own AI stack, what do they typically underestimate most about establishing that?
Vivek Juneja:I think at the fundamental level, the base of these AI systems is making sure that you're able to collect high quality data from the parts that you're optimizing and setting up a kind of a direction, your KPI on what you're trying to optimize for. So a lot of times, as Noah mentioned, a lot of these experiments don't run to the right outcome is because the initial KPI is not kind of well thought of or and, the measurements are all over the place. So some of our customers start by measuring samples across different, stages and different devices, but the more we get the fidelity of that particular data, the better these systems are able to develop predictions. So I think from a capability perspective, the biggest bang for buck is having right source of truth systems for data, making sure that there is clear KPI on what they're trying to track, having a close feedback loop on building trust and confidence in these systems, and that's basically a five flywheel that over time lets you explore more part of the ecosystem. Again, for cement producers trying to build all their own, they need to start thinking about what are the aspects, which is basically in their, things that they want to build differentiation on.
Vivek Juneja:Is it right for me to build an LLM based chat versus trying to kind of work with a provider to kind of solve the first fundamental problem. So again, you need to start where is your starting point, figure out that starting point, but build an experimentation mindset, and that comes by investing in really high quality data sets, source of truth, and building a clear understanding of what you're optimizing on.
David Bizley:Did Okay. You want to add to that? I wouldn't would Sorry, you next. I don't mind. Yeah.
David Bizley:I would tell
Scott Ziegler:you that the one thing that starting from scratch to build your own is the time. There is a significant amount of time, a significant amount of resources. You know, we actually heard, again, let's reference back to Sempore, he was mentioning if we took, you know, four to five years of building it up, that's a typical thing from starting from scratch. There are solutions. There are solutions that have been proven out over from our perspective.
Scott Ziegler:We've got 40 plus process optimizations going on across assets around the globe. You know, we're in 22 cement plants running predictive maintenance. These are already proven assets and proven results. It gives the opportunity to take those first steps. It also gives the company an opportunity to look and see what do they like?
Scott Ziegler:What do they, what do, what maybe they want to improve upon? So again, I think it's a big step to take to do it on your own and quite frankly, whether you're a large multinational or you're a single plant, getting those resources and investing in that time and that amount is a significant ask.
David Bizley:And Noah, you want to turn it?
Noah Miller:I think, we we look at it to build, you know, the AI PC of the future, you need process, you need control, you need software, and you need machine learning. And you need all four of those to be sort of really, really world class and you need them to play well together. And often we see, particularly in the bigger groups of our customers, people we work with, that they have one or two and maybe they're trying to push for a third but if you can't do all four, that will really, really hurt you. And maybe two specific examples of that, the sort of machine learning ops and infrastructure, right? How do you keep a model online automatically retraining, automatically assessing whether data is out of distribution, needs to trigger something else.
Noah Miller:Like that's very unglamorous work that people don't think about, but that will derail control just as much as a controller that didn't work. On the other end, you know, to to give a very simple process control example, every process engineer will tell you that LSF should have a, you know, massive impact on your freelance. But if you don't understand how to use the data, feed it into the right sort of model in the right sort of way, you can I know people who have spent months trying to build that model with no success because they had all the process control knowledge, but they had none of the machine learning knowledge or not as much as they needed to make that really successful? So it's gotta be all of those things together.
Vivek Juneja:Let me just quickly add. I think, what Noah was also pointing at, there's a lot of the 80 to 90% behind the scene machinery that is not visible to customers who are trying to do this all by themselves. Yes. You there are a lot of things that are kind of opportunities right now to build quickly, but in the real world, you're constantly upgrading, maintaining your datasets. There's a lineage around the data.
Vivek Juneja:You are also trying to make sure the models are improving over time. And as you scale, those problems multiply. So the operational challenge also, needs to be considered. Right. So, yes, there are places where, we have seen cement producers trying to think about testing it in a light mode way, but as soon the scale comes in, the complexity of this 80 to 90% behind this in machinery becomes quite complicated to manage.
Speaker 5:Great. Thank you.
David Bizley:And, Noah, I'll start off with you, with this one. How do you see the rise of things like agentic AI and LMMs within LLNs within the sphere of industrial production? You know, for a cement plant, how do you make the decision about whether to build
Noah Miller:or buy? So I think, you know, there there may be a couple things to separate. There's AI LLMs for sort of operator support, training, data analysis, Birkin gave some great examples of that, and then there's AI for like process control. And on the first one, like, I think there are great use cases we're seeing already. We have a bunch of tools that can take, you know, scribble down operator notes and turn those into data that our models can learn from and train to be better.
Noah Miller:On the other side, I think that's what's the most exciting thing that we're talking about. The current set of LLMs are not gonna run aplomb because they are very good guess the next word machines for a bunch of things that there's lots of data on the internet for. But GPT-one was in 2018. 2023, 90% of Fortune five hundred were using LLMs in their workflows and now the world has changed. So we are really excited for what is GPT-one for cement, that's what we're building, we think.
Noah Miller:You know, these sort of understanding the latent space underlying all cement processes and using that to build a truly autonomous control system.
Speaker 5:Excellent. Vivek, did you
Vivek Juneja:Yeah. I could add to that. I think one big differentiation is the support use case that Noah referred to. A lot of it is, I think, the most accessible way people, start using a large language models in an environment. I think as soon we give tools like through the agentic system, this becomes interesting.
Vivek Juneja:So again, in our customer base at LSME, we have a lot of closed loop implementation. I see this closed loop implementation as almost like layer zero of an agentic setup, where agents have a goal and you are given guardrails and the agents can then figure out what's the right solution to have. So I'm quite ambitious on what Noah said. The GPD one moment of cement is coming very soon, and I think we all are working on each part of that equation. I think aspect is to think about one I think we sometimes, I think, overestimate in short term and underestimate in long term.
Vivek Juneja:So as soon we give lot of the tools access to these agents, it's built to perform set points. Again, we are referring the last presentation a lot. So in Shimper OAG, they were able to directly set the give the set points to the systems deployed locally on-site, and they can take actions on it based on conditions. I think this is where the future is heading.
David Bizley:Scott, and Lovely. Yeah.
Scott Ziegler:I think the, as far as where this is going, the, the, you know, the LLMs are an opportunity for us to expand even more AI solutions in the industry itself. When I think about logistics, when I think about sales, when I think about all of the other items that are outside of a cement plant per se, that's a big opportunity. That's a big opportunity for us at Same AI and what we're looking at as far as our total solution. But I think again on the agentic side, you know, is enough inherent knowledge of the employees, we're talking about the, you know, the control room log books, OEM manuals, there's enough information there for us to be able to pull together a lot of these things to really take it to the next step. So we're enhancing, we're just enhancing the basics of what we have there.
Scott Ziegler:You know, do we get to GP for one? I'm excited about it. That's certainly one of the phases that we see there, but it's all building on the fact that a full solution and and going digital across the entire industry itself, it really gets to the whole decarbonization aspect of
Speaker 5:it. Okay.
David Bizley:Well, we had a question come in here from the audience. So who wants to take a stab at this one? Where do you see the role of first principles in the AI age, for example, thermodynamics for process or product optimization?
Noah Miller:I can definitely start. I think it Oh, please. It comes back to those four pillars. So purely data driven approaches don't understand thermodynamics and there are, you know, everyone here is aware there are hugely nonlinear, complex, difficult relationships to understand in the data. So you need to build models and control systems that bring those two things together.
Noah Miller:They come from both what you can see in the data and they also come from the understanding of physical systems. And even more than that, it can help operators right now. So to give an example, we were looking at blockage detection in one of our customers in South America and the operators were convinced that it was their magnesium levels that were driving their blockages. And we built a blockage model that did not use magnesium. We did a whole load of data and we showed them that it wasn't actually that, it was something else, and we were able to both model, blockages to to prevent in the future and help their understanding of that right now.
David Bizley:Vivek, I see you nodding.
Vivek Juneja:Yeah. I do. I think, I may have kind of a general answer to this question. I think what Noah mentioned is absolutely true. Lot of the inherent learning happens if the relationships around the first principles, you know, understanding of physics is very inherent in the data.
Vivek Juneja:Whenever that is not inherent in the data, especially the question that you asked about large language models in agentic, there needs to be a feedback loop. So, you know, machines learning through feedback. Again, we will see very soon, real world, you know, world models, robotics in the real environment where things which are not in just data that is available in the training set, but something that they're interacting with allows them to experience these first principles first, know, head on. So I think, there is a role for first principles in the AI age. We are still at the beginning of what it could look like.
Vivek Juneja:A lot of these will come from machines trying to feel the physical environment, but also learning through human feedback. And I think that's where I see a potential of, you know, expanding into different areas.
Speaker 5:Okay. Thank you. And and, Scott, did you
David Bizley:have anything to add?
Scott Ziegler:Nothing to add.
David Bizley:Okay. Excellent. And but for you, Scott, now where do you see the sort of the balance of value when it comes to adopting AI in in the cement? Is it more a case of preventing downtime or optimizing process performance? Or is it a sort of fairly even balance?
Scott Ziegler:For me, it's both. Happily because those are the solutions that we actually we have out there in front of us. Personally, you know, this is we like to use the analogy of the self driving car and we actually use the analogy of a car in a lot of things, right? We think of the check engine light as the predictive maintenance type of side of thing. We think of the self driving as the, as the process optimization.
Scott Ziegler:You know, the old, I think one of the old adages is you can't optimize something if it's not running. Well, you, if you don't know if it's running, you need to, you know, can we get everything we can get out of it if it is running? It's a good balance to have there. You know, when we think about where that leads to further down the road, when we think of whether it's decarbonization or process optimization, whatever may be there, the ability to have the availability of the equipment is key. That's the first thing, right?
Scott Ziegler:And so if we can improve the availability, reducing the meantime between failure, having that overall equipment efficiency, kind of the basics of it. If you have equipment that's running and you really need to go squeeze that next 10% out of the machine, you need to go get some more energy reduction. If you feel like the plan is already in a situation where it runs fairly well, that's really the opportunity to go and get some process optimization, whether it's around specific energy consumption, whether it's around throughput, is it on, you know, quality control? Each of those in their own way are going to drive specific savings. And so I wouldn't necessarily put one in front of the other, but I would say that if the plan is, if your efficiency of your planning, available, if your plan is high enough, get with the optimization.
Scott Ziegler:You'll have bigger impact possibly on cost control, on CO2 emissions. But again, if your plant's not running efficiently at all, you will have everything from emission problems to everything else. So there's a balance there, and I think it's very dependent on the customer themselves.
David Bizley:Well, that's all for this episode. If you enjoyed it, please make sure to like and leave a review. And, of course, stay tuned for part two coming up soon. Oh, and if you'd like to meet experts like these in person and discuss the decarbonization of the cement industry with leaders from around the world, then head over to worldcement.com/envirotech to buy your tickets for EnviroTech twenty twenty seven today. I just wanted to take a moment to remind you to register for World Cement.
David Bizley:It's free of charge and gives you access to the latest issues of World Cement, both in print and online. Every issue comes packed full of regional analysis, technical articles, project case studies, and the latest industry news. Simply head over to worldcement.com, click the Magazine tab, and register today. It's as simple as that. Happy reading!