What Industrial AI Actually Looks Like | Kriti Sharma, Nexus Black

[00:00:00] [SPEAKER_03] I'm Craig Smith with Eye On A.I. on the sidelines of IFS's Industrial X Conference in New York City. IFS is the leading provider of industrial AI software. It provides AI solutions for asset-centric industries, think energy, utilities, manufacturing and aerospace.

[00:00:23] [SPEAKER_03] I'm interviewing Kriti Sharma, CEO of IFS's new unit Nexus Black, which builds solutions for hard problems. Okay, Kriti, great to meet you. Lovely to see you in New York. Can you introduce yourself to listeners to begin? Tell us your background, how you got to IFS and how you got to Nexus Black.

[00:00:51] [SPEAKER_00] Yeah, sure. I'm Kriti. I grew up in India. As a kid, I used to build robots. They didn't solve very important things. The first robot I built as a kid, it helped fetch Snickers the candy from the snack bar at 3pm every day. Wow. Much to my parents' delight, I do better things with my life now. I'm a computer scientist by background. And then started building AI businesses and products in the last 10 to 15 years. Yeah.

[00:01:16] [SPEAKER_00] And for the most part in my career, I've built AI businesses and products for people who work at their desks. Accountants and lawyers and so on. But I was fascinated by building businesses, products, experiences for people who are not at their desk. And that's what brought me to IFS. We started Nexus Black earlier this year. It is a business inside of IFS that works with the most strategic clients. Boots on ground.

[00:01:43] [SPEAKER_00] If you think about AI with a hard hat on, that's us. We go on the ground, we spend weekends in hangers, we spend our days in refineries and factories, and we travel along with the field technicians so that we build things where failure is not an option.

[00:01:58] [SPEAKER_03] Yeah. And given your background in robotics, I'm just curious, where did you study? Where did you study computer science?

[00:02:08] [SPEAKER_00] At St. Andrews in the UK on the east coast of Scotland. Oh. Lovely tiny small town.

[00:02:14] [SPEAKER_03] That's a wonderful school, yeah.

[00:02:16] [SPEAKER_00] Yeah, thank you. I've always loved building things and I'm so glad I get to do that now in the physical world. Every day is a joy. And you start to learn how different it is to build things for people at their desks versus in factories and hangers and plants and sites and in the field. Yeah. It's much more challenging and more fun.

[00:02:36] [SPEAKER_03] Yeah. And I was talking to Mark Moffitt earlier about some of the work that you guys are doing with Boston Dynamics with the X and 1X on, I guess it's still called Spot on the robot dog and with Neo, the humanoid robot. How, and is that your focus at Nexus Black? Or I know you're also working with Resolve.

[00:03:06] [SPEAKER_03] Was that a Nexus Black product?

[00:03:08] [SPEAKER_00] Yeah. So Nexus Black products are regardless of the form factor. Yeah. So we're very focused on building the right tool for the right problem that needs to be solved. And we have a team of elite engineers. So these are perhaps the most rare people on the planet right now who've built production-grade AI products already before. And we combine that with deep subject matter expertise. And I'm happy to talk about what that really means because everyone comes and says, you know,

[00:03:36] [SPEAKER_00] they're building deep subject matter expertise in our products. So I'm happy to talk about that. But we're agnostic of the form factor. I see. Whether you're deploying it in a humanoid or you're deploying it in a robotic environment or much more digital form factors.

[00:03:51] [SPEAKER_03] Yeah. And on the digital form factor, which isn't really a form, I guess, is you're doing a lot with agents, I imagine.

[00:04:01] [SPEAKER_02] Yeah.

[00:04:02] [SPEAKER_03] And I'm curious about how you see the agent workflows developing because from what I understand, there's a lot of excitement, a lot of promise. But most of the pilots out there are not going into production for various reasons. And how do you see agents developing?

[00:04:27] [SPEAKER_00] So we focus purely on the job to be done in Nexus Black. And by that I mean we go on the ground and we will make a bold claim. And I'll back it up with facts. We start to see value for our clients in about three weeks. Yeah. And the way we do that is, I've worked in AI my whole career since I was a kid, as I mentioned, and I absolutely would be the person who detests AI washing and statements like this.

[00:04:56] [SPEAKER_00] The way we do it is, I don't think there's any other way of solving these problems and making it successful in the physical environment except for being on the ground. Yeah. So we get out of our comfort zone, of our world, we're on those sites and we look at the situation that we're dealing with, the kind of data, the real problems. And I'll give you an example. When we work with field technicians, they tell us, if I have to take my safety glove off to type into a thingy to use your thingy, I'm just not going to do it.

[00:05:26] [SPEAKER_02] Yeah.

[00:05:26] [SPEAKER_00] That teaches us what we need to really use here is voice transcriptions and other sort of form factors and capabilities. Oftentimes, you might think, oh, this is very cool. I can use an image model to scan a thing and solve a problem. And then you realize when you work with the engineers that often the first signal of something's about to go wrong is a vibration or pressure. So all of these are nuances. You learn those things and you refine.

[00:05:52] [SPEAKER_00] You're building something for a site or a plant in the offshore environment. There's no Wi-Fi in the Pacific. You have to learn to build in this hard world. And when you solve these real problems on the edge, on the ground, with the people who are going to use it, then your chances of failure dramatically reduce.

[00:06:12] [SPEAKER_03] Yeah. And that's interesting working on the edge in environments with low or no connectivity. How do you do that?

[00:06:24] [SPEAKER_00] We compress a lot of that knowledge and we deploy it on devices to be able to do that work. And all of these problems are solvable. And so, when you're thinking about the problem, the value really comes in from knowing which problems need to be solved. Yeah. And it's really also no longer about the best technology or the best model. It's about using it in the best possible way to solve a hard problem. And that is where failure occurs.

[00:06:48] [SPEAKER_00] If you don't identify the right high value solution, don't really identify these constraints that our solutions need to operate in. You don't understand deeply what's going through the emotional hopes and dreams and fears and needs of these workers. And it's very hard to build good product. But when we get these things right, when we know the constraints they're operating in, when we know their real emotional and practical fears and challenges, we work with them. That's when magic happens.

[00:07:18] [SPEAKER_03] How big is your team? And when you go out on site, there was the example of the distiller. And I think that was Resolve that you were applying there. You're going alongside with the worker, kind of watching what they do. How many people do you have doing that with one client, for example? Is it a team that goes in? Is it one person that's responsible for absorbing all of that?

[00:07:47] [SPEAKER_03] And then when you go back and build the product or tweak the product for that use case, are you fine tuning an anthropic model? Or is it always with men? I'm just curious how that works.

[00:08:09] [SPEAKER_00] Let's walk through that with an example. Because I did make a bold, outrageous claim earlier. I told you I was going to back it up. So let's go through it. Yeah. Let's take the example of what the work we're doing at the distillery you mentioned. Yeah. The one in the conversation here is William Grant.

[00:08:24] [SPEAKER_02] William Grant.

[00:08:25] [SPEAKER_00] They are the brand behind Glenfiddich Whiskey and Hendrix Gin. They are on the west coast of Scotland, which brings me back to Scotland a lot. I love that. So we went on the ground and we understood just very rapidly within a few days what their challenges were. And in their case, the issue was of production loss. When a machine goes out of whack and something breaks, they were spending about 38% of their time in emergency and corrective fixes. And that takes so much more time.

[00:08:53] [SPEAKER_00] The plant is down, the site, you lose batches, right? Now, that's point one. Problem two, the number of technicians who have the deep domain knowledge, that is quite constrained. And they have to, they're expanding. So they have to do a lot more work with fewer resources. Some of these are challenges that are not just unique to them.

[00:09:13] [SPEAKER_01] Yeah.

[00:09:14] [SPEAKER_00] On the ground, we observe what's going on with the technicians, how the plant is functioning, and then started to build a knowledge about a representation of the factory of the plant and what's going on. And some of these challenges and the data, the knowledge that is represented here, is not in formats you and I would usually hang out in and read.

[00:09:32] [SPEAKER_00] For example, how a plant, a little pump or a seal connected to the pump, is connected to the rest of the environment, is represented in a complex engineering schematic diagram called piping and instrumentation. Now, these things, if you look at it, it's like, ooh.

[00:09:49] [SPEAKER_01] Yeah.

[00:09:49] [SPEAKER_00] How do I even make sense of it unless I'm an engineer on the job with 20 years of experience? So we take capabilities like Claude from Entropic, and then we domain train it and we push the boundaries. And this is what Garvin from Entropic yesterday was saying with our lead team of engineers, which works so closely together. We push the boundaries to make it understand what is a piping and instrumentation diagram. So when something goes off or before it goes wrong, we can see not only that part that needs to be fixed, how is it impacting upstream and downstream?

[00:10:19] [SPEAKER_01] Mm-hmm.

[00:10:20] [SPEAKER_00] And these are the kind of moments where you go from, I did an AI pilot, it was like, yeah, I kind of did the job and not really ended up with a chatbot or some sort of an agent.

[00:10:31] [SPEAKER_02] Right.

[00:10:32] [SPEAKER_00] To, I now have a solution that understands my site, my plant, what's happening here, what's the connection, how does it fit? And then it's helping my technicians working alongside them get an answer and solving the problem. So the result in the case of William Grant's is by their own assessment, they expect to save about 8 million, 8.4 million pounds per year at just this one factory. Now you scale that.

[00:10:58] [SPEAKER_00] So, anyway, long story short, it works when you are on the ground and you're not scared of piping and instrumentation diagram.

[00:11:05] [SPEAKER_03] That's right. That's right. Or you have a model that can read the piping. Sure. Yeah. The, being on the ground, I mean, I talk to a lot of companies that are offering off-the-shelf solutions. Mm-hmm. Mm-hmm.

[00:11:23] [SPEAKER_03] And part of the problem, it seems, in penetration or uptake at the enterprise, particularly physical operations, is a one-size-fits-all doesn't really work. Right. As you say, you have to go in and see that the guy's not going to take his glove off, for example. Yeah. Yeah.

[00:11:46] [SPEAKER_03] How important is that bespoke element to the adoption of AI in the industry?

[00:11:57] [SPEAKER_00] I think more important than bespoke is deep domain understanding.

[00:12:03] Right.

[00:12:03] [SPEAKER_00] So, if you're building, we spend a lot of our time in these industrial situations. We only operate in these specific industries, in manufacturing and utilities and aerospace and defense, in service and telco construction engineering. We don't venture outside of our lane because that means when I go into these industries, the next and the next and the next company I speak with, I understand their needs better. Right.

[00:12:28] [SPEAKER_00] I've solved the ability for AI to read these complex engineering diagrams once, the next time we can do it like that. Right. So, over time you start to build specialization and you start to scale said specialization. The magic now is previously it would have taken a lot of people, maybe possibly even hundreds of engineers, years worth of effort to make some sort of technology or software work for manufacturing supply chain.

[00:12:57] [SPEAKER_01] Yeah.

[00:12:59] [SPEAKER_00] That's not how it works anymore. The timeline has shrunk. So, the people who we work in my team, they are of course some of the best, most elite AI engineers. We pair that with this expertise of what happens in the manufacturing world. Experts and utilities. Just across the room from here, we have a number of our key utilities, customers, the most advanced, and subject matter experts right now building the next product. I can't tell you what it is, but I will. Yeah.

[00:13:25] [SPEAKER_03] Resolve is a standalone product.

[00:13:28] [SPEAKER_00] Yes.

[00:13:29] [SPEAKER_03] Was that developed for William Grant?

[00:13:31] [SPEAKER_00] It was developed with the first set of clients that are in asset heavy environments. I see. In plants and out to the field. And we did this with a bunch of them, so we end up not making something that only works in one environment. Right. And that's a good example of how we can deploy something and create something very quickly, very closely with our clients and then scale it to a much broader base.

[00:13:54] [SPEAKER_00] Now, for your listeners and viewers here, they're probably already quite aware of this new generation of AI companies where a small number of employee base is creating tremendous amount of value, exponential value in revenue and valuation terms. And that has been now possible because of the ability for the creation of technology itself to have gone so fast and so much more advanced. And that is what we do now, too.

[00:14:21] [SPEAKER_00] It's a small number of people who are highly capable and highly specialized who know how to use the AI tools themselves to build sophisticated solutions that are production grade in a matter of weeks.

[00:14:32] [SPEAKER_03] Yeah. And that's important, production grade, because as I said, so much of what's going on right now is piloting and making it. Yeah.

[00:14:42] [SPEAKER_00] I can't make that work in my industry. I'm afraid. In my world, it has to be production grade. That's right. You know, we work with airlines in compliance and air worthiness, air safety. Every time a flight takes off, about 400 lives are at stake. Yeah. We can't mess that up. It would be reckless to do anything that's not production grade, enterprise grade. Yeah. That's what we do.

[00:15:06] [SPEAKER_03] And Resolve, can you just describe what's inside that word? Yeah.

[00:15:12] [SPEAKER_00] So, Resolve is a product we've designed for the frontline workers in the factories and out in the field. The way it works is it understands the issues before they occur. So, it's constantly monitoring systems at a plant in a site and what's going on through sensors. So, it's temperature, pressure, noise, sound, images, video to detect what might be going on and past failure patterns as well. And then using that, it figures out what might be the issue.

[00:15:39] [SPEAKER_00] It reads those schematics I talked about, upstream pressure buildup, downstream production loss, and then helps create a series of detailed instruction sets like an engineer would to give you the tools as a frontline worker to be able to do the job right the first time.

[00:15:54] [SPEAKER_03] But in that you're a frontline worker, this isn't someone sitting at a computer console. So, how are they receiving that?

[00:16:02] [SPEAKER_00] On their device. I see. So, they're just getting it on their device but they don't have to type because of the situations we talked about. They can use their voice and get the response in the right way depending on the form factor.

[00:16:13] [SPEAKER_03] And do you guys manufacture the device or is it?

[00:16:16] [SPEAKER_00] No, it's just their phones. Or is it?

[00:16:17] [SPEAKER_03] Could be anything. Yeah.

[00:16:18] [SPEAKER_00] They can use it on their mobile phones. And usually we deploy it on the kind of phones that are cleared for usage in the factory environment because of fire and safety risks.

[00:16:30] [SPEAKER_03] Yeah. What are some of the other products in the pipeline then?

[00:16:34] [SPEAKER_00] So, another one we talked about yesterday at the event here in New York is something we're really proud of. This is our solution that we've built for airlines, commercial airlines and airworthiness. What that means is every time an aircraft manufacturer or the FAA, the Aviation Authority, they release a guidance for some sort of service bulletin or airworthiness directive,

[00:16:59] [SPEAKER_00] which means things that manufacturers or operators of aircraft need to do to keep the fleet safe. Now, these things, these directives, they come in the form of hundreds of pages long documents. And this is not just words in the documents. They're pictures, they're detailed specifications and parts. And getting it wrong, the consequences are huge.

[00:17:21] [SPEAKER_02] Yeah.

[00:17:22] [SPEAKER_00] Each time you get it wrong, you could be faced with a one to 20 million regulatory fine dollars. If you have to ground the fleet because you couldn't do this in time, you're talking about $140 million in lost revenue per day. And then the worst case scenario, it's a risk to human lives. So, this is failure is not really an option. Getting it almost right, not an option. So, we read through these directives. We compare it against the fleet of information that we already have.

[00:17:49] [SPEAKER_00] So, it's the tail numbers of all the aircrafts that may be in operation. We look at all the parts and then we do the planning and the maintenance work of what you need to do, which part exactly needs to be swapped out. Now, in normal situations and non-technology situations or non-nexus-backed environments, human engineers are spending weeks reading through this information and trying to make sense of what is the action they need to take. What's that part that needs to be swapped out? What kind of information?

[00:18:19] [SPEAKER_00] And when do they schedule it? Because the planes are coming in and out. So, these are the kind of problems we love solving because they're complex and the impact and the value we drive is huge. And I would love to tell you about one more. Sure. And this is really the closest to my heart. This is the work we're doing in the utilities world when disasters hit.

[00:18:40] [SPEAKER_02] Yeah.

[00:18:40] [SPEAKER_00] Disasters and major events are sadly becoming more common. These could be weather-related events like snowstorms as we see here in New York, hurricanes, wildfires in California, or outages like what we saw at Heathrow Airport, I live in London, or what we saw in Iberia with the cutoff and blackout.

[00:19:00] [SPEAKER_01] No.

[00:19:01] [SPEAKER_00] And when these kind of moments occur, there is a lot of challenges the frontline workers have to deal with. They have to go in and solve a problem when a lot of the infrastructure might be submerged in water. And they're having to coordinate. They send a back signal out to all their other crew friends in the nearby utility areas. They'll pool together. They meet in the parking lot of a Walmart to figure out what instructions there are to go tackle the situation,

[00:19:28] [SPEAKER_00] whether it's to bring a power line back up or it's evacuating people from their homes until the situation is resolved. Now, in partnership with the work we are doing with Anthropic on the upcoming capabilities of the models, not only to answer text or reason through it, but take actions on long-range planning. We can now solve these problems in a way that has never before been possible. That means coordinating this complex amount of action.

[00:19:57] [SPEAKER_00] This is the hardest time for a frontline worker. They get a call, 4am, the storm is going to go this way, it's coming here. The fire is escalating. And for me, it gives me goosebumps when I hear firsthand from people who have been responsible for needing disaster response. For example, what we saw in California beginning of this year, the same person called us yesterday, they sent us a note saying,

[00:20:26] [SPEAKER_00] this is so amazing because it's going to help us get communities back online and hospitals lit up faster.

[00:20:32] [SPEAKER_01] Yeah.

[00:20:33] [SPEAKER_00] It really, right now, I'm feeling the goosebumps because to me that's the power of technology. This is why I love doing what I do. You want to go back to your thing?

[00:20:43] [SPEAKER_03] Well, I was going to ask you, when you're on an airline and there's a problem and they need to fix something, I'm always, how do the guys know, you know, some light is not lit up or something? Is that a use case for you guys or you're really in the maintenance center?

[00:21:06] [SPEAKER_00] So we do a lot more on the maintenance planning work and the maintenance execution work and things where we know that there may be a fault with a part that needs to be swapped out and something else needs to get done. And these systems are getting more and more sophisticated. When a plane goes into the maintenance hangar, a lot happens. My team and I have been inside and we spent the weekend at a hangar. We went inside a stripped down plane.

[00:21:30] [SPEAKER_00] And as you walk through the plane, you do see that you're trying not to triple where the legs of the engineers are lying on the floor just doing things. And oftentimes it's managed on pen and paper, cardboard situation. It's kind of, wow, that has to change. And now we're giving them these tools that give them much more confidence. And also there's a lot of tension on the ground. You know, the engineers want to do things a certain way and the planners sitting in their offices, they want to need to do things a different way. And now we can bring the two together.

[00:21:59] [SPEAKER_00] Process huge amounts of data, but also take the right actions at the right time.

[00:22:04] [SPEAKER_03] Yeah. Is your team growing? I mean, where is this going? Because it seems like the possibilities are endless.

[00:22:12] [SPEAKER_00] Yeah, we are growing. If you know people who are very excited about solving hard problems out in the real world and they love the physical environment, and they love the world of making technology solve problems, then we are the people to come and talk to. We're at Nexus Black at Athos.com. That was a little plug there. I do think the importance, the magical moment now is you don't have to be a geek like me,

[00:22:37] [SPEAKER_00] who is a computer scientist and roboticist type person to come and solve problems in this world. You can be a subject matter expert in your domain. You can be someone who is just exceptionally curious person with very low ego. You're all welcome.

[00:22:51] [SPEAKER_03] Yeah. Okay. Great. Great. That was wonderful.