AI-Powered Engineering: From Medical Devices to Maritime — with Axial3D and Compute Maritime

Episode Summary

The episode delves into how AI is revolutionizing engineering and product lifecycle management through the lens of two companies: Axial3D and Compute Maritime. Shahroz Khan, CEO of Compute Maritime, focuses on developing generative tools for vessel design and simulations to expedite ship development pipelines. Roger Johnston, CEO of Axial3D, specializes in 3D medical imaging technology that transforms 2D medical images into detailed 3D models for patient-specific applications like surgical planning. Key insights include the potential of AI-driven tools to democratize access to advanced imaging technologies globally and enhance design optimization processes across industries. The episode highlights how these companies are leveraging strategic partnerships and targeted events to drive adoption in niche markets, offering PLM and engineering professionals a glimpse into future technological advancements that could transform their workflows and product development strategies.

---

Full Transcript

Shahroz Khan

Nice.

Roger Johnston

you

Michael Finocchiaro

Hey, we're live. This is Michael Finocchiaro on the AI Across the Product Lifecycle podcast. ⁓ Really excited to you. I haven't done any of these with a customer, ⁓ startups that were dedicated to specific markets. So it's really exciting to have Shahroz Khan, who's dedicated a CAD system to creating boats and Roger Johnston who's working completely differently on the health side on the medical industry. ⁓ Axial3D and Compute Maritime. Welcome guys.

Roger Johnston

Could be here, Mike.

Shahroz Khan

Thank you.

Michael Finocchiaro

Why don't you guys go ahead and introduce yourselves to everybody when I start Shahroz.

Shahroz Khan

Yeah, yeah. Hi, everyone. This is Shiroz. I'm a CEO of Compute Maritime. A bit about Compute Maritime. We're building generative tools for vessel design and simulations to help naval architects and shipyards and design offices to expedite their ship or what we call vessel development pipelines.

Michael Finocchiaro

Awesome. And Roger, how about you?

Roger Johnston

So, Roger Johnson, I'm CEO of Axial 3D. We are a 3D medical imaging company. So we take any standard MRI, CT, X-ray, ultrasound, any standard 2D medical image and automatically turn it not only into 3D, but a rich digital understanding of that 3D. So we can automatically do things like measuring anatomy or identifying anatomy or classifying anatomy. or even doing things like surgical plans for your specific anatomy that are patient specific. So, but we'll get into that over the next time.

Shahroz Khan

Here we go.

Michael Finocchiaro

It's really exciting. before diving into my questions, why is it you guys dove so specifically into one market? Were you passionate about boats or about medicine? Or did you just see a particular opportunity? How did you get this really narrow focus like that? Either one of you.

Shahroz Khan

Roger, you're going to go ahead. ⁓

Roger Johnston

Yeah, let me take it. So our founder, a guy called Dan Crawford, he actually came through medical training and medical biomedical engineering background. So his, when he came up with the idea or the customer need, which was to deploy and be able to realize the benefits of 3D imaging at massive scale, in effect to where any patient around the world can get access. to have there who can get a medical image, or a 2D medical image scan, can have access to 3D. That very much set the industry we were going to be focused on. And we were focused on one goal, which is for every patient around the world, whether first world or third world, to be able to have 3D as part of their care pathway.

Michael Finocchiaro

That's awesome. Awesome. And how about you, Shraaz? Are you a boat guy?

Shahroz Khan

Yeah, no, I'm not a boat guy, actually, honestly, you would believe it. Although I will what I always say that, although I do have a PhD in marine engineering, but I would still not call myself a marine engineer. So I'm still a mechanical mechanical engineer. And I went on doing mechanical design. So when I was doing my masters in mechanical design and a project came up related to CAD,

Michael Finocchiaro

You

Shahroz Khan

So I did my masters at Istanbul Technical University. you know, like the leisure industry or what was called the art industry is quite big there. Exactly. So there was this project that was funded by Turkish Scientific Technological Research Council, something like that. And so the idea was ⁓ because when we were designing, let's say like engineering systems, right, we always focus on, let's say, performance and ⁓

Michael Finocchiaro

Yeah, up and down the bus first.

Shahroz Khan

For the yacht industry in particular, it's not always about performance, but it's also about their semantics, how they look and feel. So for designers, when optimizing ⁓ boats or yachts, it's very easy to integrate numerically the performance criteria. But semantics being very subjective, you can't numerically integrate, because they are very subjective criteria. And there is not a mathematical.

Michael Finocchiaro

Right. Yeah.

Shahroz Khan

that's a formulation or definition to define someone's looks and feels of a certain yacht. So the project was about how we can integrate that semantics into also an optimization pipeline. So that was the very first entry point for us or for me to both for AI and into maritime. So we went on actually building a standalone CAD system.

Michael Finocchiaro

lesson.

Shahroz Khan

We gathered lots of data about yachts in terms of images. And then we went on doing surveys with Naval Architect what they think about what this certain yacht represents. Is it like looking luxury? it chrysomatic, et cetera? And then we trained an AI model, a deep learning model on that one. And we integrated into our software. And of course, that was part of the research activity. And actually, we got a patent. on the technology behind it. So that's what we were talking about 2015, actually. Yeah, so, yeah. And then the journey started to founding a company called Brain Labs. And in the Brain Labs, we wanted to do generative design. But generative design usually, generative design has been there like for ages, right? But in that case, in previously, it was just based on the rules. So we will identify some handful of rules. We would hand code those rules.

Michael Finocchiaro

Well, that's a good segue into the first question, I think.

Shahroz Khan

to create some design variations. But in this case, in generative AI case, we have the data to create those design rules. And that's where the journey started and we are now here in compute maritime developing generative tools.

Michael Finocchiaro

Awesome. Thanks for that background. That's really cool. And it segues very well into this first section of questions, which is ⁓ back in 2022, we had this pivot, ⁓ a tipping point, right? In terms of AI and LLMs and ⁓ suddenly people had, everybody heard of AI, not just people that read science fiction or people that were academics. ⁓ Were you guys, mean, obviously your answer Shahroz is going to be pretty obvious, right? Because you're doing it already, but were you guys like super skeptical or super bullish back in November, 2022 and open AI changed the world? What about you, Roger?

Roger Johnston

Yes, so I think we were children of the time in that in 2016 when we were formed, we were born in the cloud. So there was no concept of us developing anything on a client server architecture or anything like that. And we were actually born in AI because when we started understanding the problems associated with turning 2D medical images into 3D.

Michael Finocchiaro

Hmm.

Roger Johnston

First all, takes a lot of compute power. So it had to be cloud and video predicated, but actually we needed the only way of doing it at scale, that was fully automated, was to train machine learning models that could produce, ⁓ let's say, FDA quality outputs 100 or tending to 100 % of the time. And to get that level of precision, the only route was AI. So we were very fortunate. Good luck, good timing.

Michael Finocchiaro

Of course.

Roger Johnston

Born in the cloud, born in AI. ⁓

Michael Finocchiaro

Nice.

Shahroz Khan

That's a great example. I would say a joke that I always tell that when I started, when I was working on this problem during my PhD, actually, and someone would ask me, like, what you do, what research you're doing. I was like, I didn't know how to explain that. Like, it's really hard. And then to 2022, 2023, January, I defended, Chai GPT came, the language models, what I would say, like language models came to the public domain. So that was so easy to correlate saying that, well, we are doing, I'm doing something similar, but exactly for ships. So I think one thing that I always say that, yes, the pivotal point, we saw it in 2022 because the AI, would say like, or the generative or the predictive models, they came into public hand through the natural language processing, right? So it's exactly...

Michael Finocchiaro

Yeah, they were democratized basically, right? They were no longer academic exercises.

Shahroz Khan

So that came through the language model. So when? Because we can correlate each other very easily through the language. So that's why the public, I would say, concentration grew because the language model came out. And they could see through language that language model have a lot of potential. And now these language models, AI could also be used for other things. So what I always say, even the very first model that was invented, ⁓ which was Boltzmann machine was generative in nature. We stopped using generative models because we didn't have the capacity to do it. Andrew NJ, which is a renowned figure in AI, his entire PhD thesis at Stanford was on generative models. So that's the pivotal point actually happening in 2015 when Andrew NJ, Ian Goodfellow introduces GANs and then the story continued in terms of images and that's where like the 3D shapes and then the language came into being. So yeah, it's quite an exciting time.

Michael Finocchiaro

computing. It's awesome. Sounds like I need to write some historical articles on that too, since I've all the things on CAD and PDM. So I know you guys started before ⁓ cloud code and anti-gravity and well, cursor, if I go backwards in time, ⁓ how has AI changed the way your developers are working on a daily basis? Obviously, ⁓

Shahroz Khan

you

Michael Finocchiaro

It's hard to even imagine a programmer now not leveraging AI at all because you can just go so much faster. ⁓ What's it like at ⁓ Compute Maritime for the developers? Are they in cloud code 24-7? Yeah.

Shahroz Khan

⁓ We are using Claude for sure, but a lot of the technology that was built, I would say, the hand-coded by the programmer instead of like assistance having from the Claude. But now the interesting bit is that we have now agents, those who can understand the back-end code that we wrote ourselves. And by understanding of that one, now we can implement multiple agents that work interactively with our programmers to help them expedite.

Michael Finocchiaro

Hmm. Right.

Shahroz Khan

And that's where an interesting thing that happened within the company that we said, we're not going to be, we're not a company that is building a software, that's like a UI. So we are a company that is building a technology and we are presenting that technology through APIs. But once you have APIs, you don't need to build like as a company, ⁓ UX, because the customers going to design their own UX.

Michael Finocchiaro

Nice.

Shahroz Khan

And that serves two purposes. First, that software or the UX includes only the features that a specific design team or customers is interested in. And the second, the learning process become very simple because the UX is designed only for them. So that's why we say we build back-end technology in the form of AI. We stopped over development in term of all from the UX point of view. We focus everything now in the APIs. And now we, whenever... a team comes up, we understand their problem, we see which components of our array will be used, we plug in that, we see if they have any existing technology that can be used. So we plug in all together, build a simple UX through cloud and give it to them.

Michael Finocchiaro

Awesome. Same with Axial 3D, Roger?

Roger Johnston

Yes, so I think if we look at our core system, the algorithm development which we were born in building for this process called segmentation, which turns 2D into 3D images, we then have further algorithms that increment or included that automatically. identify the landmarks of the anatomy that automatically create measurements of for the anatomy and even ultimately automatically create could be a patient specific implant or some sort of patient specific device selection. process, I guess the AI that we had in 2016 has incrementally improved, but that that was already, I guess, the bleeding edge of certainly medical of medical AI. Where we've seen the huge step change, if you imagine that, those algorithms are wrapped in workflows. We have a platform that moves through the process from image capture through to whatever the surgical plan or device is at the end. That's just a standard web application or web platform ⁓ application. All of the productivity gains that we've seen where we used to have stacked developers and there are now... huge number of tools that have massively cut out that make us much more productive. So let's take a at the baseline. I think the really interesting area here is imagine that we're surfacing medical record or medical images that are used to be hard to understand x-rays or CT scans or whatever. Neither rich 3D images that you and I can understand and there will be an automatic recognition that there's a clot. here's the size of the clot, here's the dimension of the clot. That just gets surfaced as the metadata just automatically comes up. Because every doctor who maybe looks after stroke patients needs to know those information. However, where it gets interesting then is whether it be in terms of the patient understanding of that or whether it's providing broader information to the clinician team is being able to actually go out to the various large language models or models that are being created by whether it's Google or AWS or whoever, that actually could go, okay, for this type of patient condition, you can start getting either explanations for patients and what that means, what their choices are, all of that type of information that exists somewhere in trained models or indeed in the broader global data sources. That is really exciting. So we can start building patient-specific clinical applications that are presenting to them not just here's the information but truly what it means and even what it means for them based on their patient record. So I think that opens up a whole lot of interesting space. Of course one thing which we might come into Michael as we continue this discussion, all of this has to be done in a regulatory framework because that's where this all gets pretty pretty fruity.

Michael Finocchiaro

Mm-hmm. Yeah, I was about to say that you've got I was about to say that's sort of the maybe less of a challenge for Shahroz, but certainly for you is the fact that we're talking about probabilistic models and you need deterministic results, particularly in medicine. ⁓ So is that a question of guardrails? Is a question of just narrowing the scope of the eye and one particular domain to just something very, very specific just to driving it, know, just if you have enough lane to make a decision without ⁓ hallucination.

Roger Johnston

Yeah, so Michael, I'm so old that I was a COBOL developer in my first year. So I'm very used to structured software development. And I guess in many ways, even though we were born in AI, actually, as you say, we can't have dynamic, continually learning anatomical databases, because if we get 99 % pixel accuracy, let's say on

Michael Finocchiaro

Ha ha ha!

Roger Johnston

converting images from 2D to 3D, and then we get some rogue data and that dynamically alters the model. And suddenly, what was FDA cleared no longer meets the bars. So what we have to do is in effect snapshot. Lucky I did all that cobalt programming. So we can take a version one of the database, we validate that, we do all the FDA tests or whichever regulator requirements are. We validate that, we publish that.

Shahroz Khan

Hmm.

Michael Finocchiaro

Hahaha ⁓

Roger Johnston

So very much standard code releases. That then becomes the version one. get our FDA clearances on that. And that works. The requirement is you don't put out an MVP because you've got to get to clinical standard. Once you're at clinical standard, you're at clinical standard. You don't get many bonus points for going clinical plus standard because it's not required. The clinical standard is what's required by medicine. However,

Michael Finocchiaro

Right.

Roger Johnston

What we can then do is go, okay, for whatever reasons, maybe we're adding some extra measurements or something, we want to do a version two of that algorithm. So what we do is do a retrain, make sure all existing performance is the same or exceeded, that where we've added new bits, the performance meets the requirement. And then we republish that as a new version. The difficulty in FDA words or in regulatory worlds is that historically meant fundamentally new apps. application to get it as if it's a new medical device. saw a new medical device and that's a six to nine month process that costs six figures of investment. Where we've moved to and we can maybe pause after this, we have got something that's called a PCCP clearance with the FDA that means they have approved our version control, our work

Shahroz Khan

Hmm.

Michael Finocchiaro

Hmm.

Roger Johnston

flow updates or our algorithm updates, where as long as we stay within our ISO 1345 workflows and our development methodologies, they then go, okay, you're self-approved to self-approve the next algorithm version, which is awesome, because that means we can be like a normal software company in the non-FDA world. We get the first clearance and then, in effect, we self-clear and maintain all that justification and all that evidence.

Michael Finocchiaro

That's cool.

Roger Johnston

that we've done all the testing, we don't have to go back to the FDA again. even though the regulators are slow and maybe not as fast as we would like, what they have done is actually be something that allows AI to be deployed in real world medicine. We don't just have one static system that lasts for the next 10 years. It has to mature and it has to be done in a way that's scalable.

Michael Finocchiaro

That makes me wonder for Shahroz, like you were saying how you've got this agentic backplane. Do you have agents showing up for the Scrum meeting? ⁓

Shahroz Khan

Yeah, I mean, the one thing like coming to the point that this regulatory framework, even in from over model or self like our foundation model, because it's solely focusing on the geometry side of the things. And then creating that, I mean, there's two aspects of that, like, first, like, where this data came from, like, so the data came from a lot of, let's say, like, existing resources through the web.

Roger Johnston

huh

Shahroz Khan

and also a lot of data that went into the synthetic data generation. And that's where the large portion of our work went into. It almost took us five years to gather that data. So finally, in 2023, we were able to release the model. Another important aspect is about this what we called class approval of the design. So the class approvals, because you can consider them as a regulatory body that would approve your design. ⁓ from its development to from its designing towards development towards operation. And that's where like a lot of bottlenecks are created from the classification point of view. So now we're also looking to see because what we did like our model is not a language model, right? So what we say that because both the training data is 3D geometries and the output you also get the 3D geometries and the input that you give it to the to the model is your design specifications. And those design specifications can be in the form of, let's say like a language, or it can be in the form of actual engineering constraints, the way that you define your engineering constraints. And once the designs are there, so a lot of question arises, well, these designs are new, right? So what the class or society got to look at it, like how the approval process got to look at for these designs. So we went on, we already sold a manufacturing layer also because, okay, well, if the design is very efficient, is it manufacturable or not? Right? This is the question. Another point, if we sold this manufacturing, we actually, just sold it with a UK funded project from InnoVate UK funded project. And now we're looking into this classification. Can we satisfy these rules or embed these rules early on the design? Right? So then...

Michael Finocchiaro

Mm.

Shahroz Khan

The design is not only physics informed, manufacturing ready, it's also capable of being approved quickly through a class. So that aspect becomes very important and then that creates this holistic model where everything what we believe ⁓ can happen at the preliminary design stage. So I'm not sure about the other fields of engineering, but what we say in, arguably say in shipping,

Michael Finocchiaro

Interesting.

Shahroz Khan

is that 90 % of the environmental impact is logged at the primary design stage. So that stage becomes very, very important where most of the way that we have trained over designers or naval architects, they don't focus that much on that stage. So the process is only you get a requirement from your customers, you go on exploring your existing database, you pick a design, and then you tweak this design a bit. and then you go on doing the detailed design and building the vessel right away. mean, and another point is that this approach has been worked well, right? But now we have more advanced and efficient tool. We can actually think about now rethinking the way that we design the vessels. And then from there, we can actually rethink how we are training over neighbor architects, right? So this is like, I would say like, we need to look at these problems holistically.

Roger Johnston

Thank

Michael Finocchiaro

Okay.

Shahroz Khan

with the help of AI and without the help of AI. However, we want to look at it.

Michael Finocchiaro

So in the current versions of compute maritime and axial 3D, ⁓ where is the AI sitting? Is it something that the user is seeing because there's a co-pilot and it's telling him what to do, or is it more just ⁓ the plumbing underneath it to make sure that the geometry is correct and the conclusions, the insights that it gains are correct? It sounds like it's probably the latter for both of you, but I'll let Roger chime in.

Roger Johnston

Yes, so to date all of ours is in the plumbing. We have very little user interaction in truth. let's say that we are looking at the result and maybe we give them a little bit of the ability to manipulate. Let's say a surgeon prefers a three centimeter long screw rather than the 2.7.

Michael Finocchiaro

because they're just looking at the result of the analysis of the dome.

Roger Johnston

centimeter long screw and maybe they want to change the angle based on your bone density like something like that. I think going forward where it gets interesting is some of these applications I was talking about where the patient goes in the context of their broader medical record or in the context of the rest of the citizens around the world with their condition, with their same biometrics and their same whatever.

Michael Finocchiaro

Mmm.

Roger Johnston

the ability to actually go here's a set of information, typical outcomes might be this, here's the physio treatments you could have and you could potentially query against that and that would be much more of a gemini like UX for those sorts of applications.

Michael Finocchiaro

Experience. Absolutely.

Shahroz Khan

For us, it's really integrated. as I said, it's not just a co-pilot in a sense that where you're going to say, write it down. I want to have a shape. So this input is just a medium, right? The way that how we're going to feed input to data. In our case, it's the design. It's a 3D geometry. It's a 3D geometry. that is not just a variation of a random product. It's a 3D geometry event at a very complex engineering structure, right? That gonna go sale in the real condition, will be exposed to various risks, right? So in our case, is the model trained on directly on the 3D data? So there's nothing, let's say like, from the language aspect is there, okay? But we can do a connection for for the model that can render or the inputs in the form of language, but at the end, it's the 3D design itself. Another thing to add here that there's a lot of conversation, I would say, that AI for CAD and AI for design side of things is happening. So what is happening at the backend actually, it's the LLMs who are converting the instructions into a G code and that G code is rendered by a ⁓ 3D CAD kernel. So if you look at it, so model has never seen a design itself. Well, you can represent a certain design or geometry in the form of text, but those intrinsic features, the context about the geometry is still hidden in this 3D representation. I think, Roger, that you would also agree here that, for example, in your case,

Roger Johnston

upset.

Shahroz Khan

There's very intrinsic elements of this geometry, right? From the anatomy point of view, from the geometric representation point of view that are hidden within this 3D data, right? 3D scan, right? It cannot be rendered or converted into or embedded through the language. So yes, language can be a medium for your users to use your models, but the model has to see. this 3D spectrum of the geometry, the kernel of the 3D, it has to see to understand the geometry well. So that's where I think industry now lagging that everybody's saying like AI for design, AI for CAD, but that the CAD that hasn't been represented to the model itself. But in our case, that's what we did. The model, that's what we started because the design has to be represented. Another point is that ⁓ The model has to be informed intrinsically about what's the context of this geometry. It's not about just feeding a random data and then it's gonna start creating random ships, right? The ships has to be, it should also satisfy the fundamentals of what defines a ship as a ship, right? So that context has to be embedded also so that what makes a ship as a ship. So it should not spit out a log, right? So that these understanding and then this understanding

Michael Finocchiaro

Mm-hmm.

Shahroz Khan

has to be given through to by creating also, would say like, not just using off the shelf models, but also creating specific models for the specific problems. And that's what we did. And that's what I said, like the progression that we had to commute Mary, but it was due because our model was built around the specific problem. And then the language was open.

Roger Johnston

I would say, just to pick up on some of your points there, I think that the idea of first of all for the actual volume, the geometry of the 3D, understanding what in our case every pixel is inside that volume is really critical. Again, if we look back to an X-ray ⁓ as an example. the question maybe a lab tech's answer, can you see where the trauma is or whatever it is. It's a very manual, can you look and visually see this. We want everything to be derived from, I understand and affect what every pixel in that 3D volume is. That's a ligament, that's a bone, that's ⁓ soft tissue, whatever it happens to be. The second question is the context because If I don't know that you want to see only skeletal structure or it's a back pain patient and therefore I want to see their spine, I want to see their discs, I want to see nerves, without that context, then it's nothing. we have to have that. At that point, maybe the ability for patients or surgeons or whatever to query starts getting really interesting, but it's that the first stage is exactly as you said, Shiroz.

Shahroz Khan

Exactly.

Michael Finocchiaro

⁓ My friend Michael Finning in the chat was asking, this is referring probably to the previous question, if you ever thought about using AI to configure the design software itself? Like, have the AI configured the design software, not just the scaffolding or whatever? ⁓ Like, no code but done by AI? Kind of interesting.

Roger Johnston

Yeah, so let me give you one first. The FDA, and I keep using them because they are sort of the gold standard of medical regulation. Yeah, they prescribe one method of training these large or these anatomical reference databases. Whilst there's lots of ways of doing this segmentation process, there's only one of those maybe eight or nine different ways.

Michael Finocchiaro

you Well, yeah, and they're your gatekeepers, right? So super important.

Roger Johnston

that they say you can get to medical grade. Therefore, actually looking at somewhere where the AI is used to train the AI model and then it does its thing. I think the idea of using AI to actually create the design model, I don't quite know. I'm not clever enough, Michael, but I think the barrier is likely to be that we would need to get it to where the FDA trusted that. ⁓

Michael Finocchiaro

Yeah, exactly. Yeah, I'm not sure that we get back to that probabilism versus determinism problem, Anything to add to that?

Roger Johnston

Amen.

Shahroz Khan

Exactly. No, I fully agree. think that's AI to build AI. But the context, when I say again, is lost again.

Michael Finocchiaro

Well, that's sort of what we're doing with OpenClaw, which is a bit crazy, right? But anyway, ⁓ we're not talking about Facebook, we're talking about boats and bones. ⁓ Well, before I move on to the last question in this section, I wanted to ask about, well, just some advice, because I did look at the demographics. I do have younger people watching the podcast. And I think a lot of people, because we're talking about these changes with AI. ⁓

Shahroz Khan

Yeah. Yeah, that's another story. would say we need another podcast for that then. ⁓

Michael Finocchiaro

I suppose that even with AI, an engineer ⁓ building a boat or someone working for Axial or a doctor still has to have a fundamental knowledge of what is anatomy, what is ⁓ marine engineering. ⁓ So how, what are the kinds of things that the younger generations, the people that leaving university now are very, very concerned that there won't be a job there because of AI? What do they need to focus on to get their chances better and work for an awesome startup like compute maritime or actual 3d instead of going to some boring corporate job.

Roger Johnston

Yes, so I think our world, we aren't a company that has hundreds of software developers. What we need is, and we recognize what AI can offer. We are scratching at the surface and we are just one little representation of where the medical diagnostic imaging market is at. We are at like level one and there's 10 levels ahead of us over the next 10 years. The opportunities, whether it be for graduates or researchers, whatever, this, we're at the start of a journey. And actually a lot of this isn't about being able to write the code or whatever. It's being able to envisage, validate, build, da da da. That whole innovation cycle. We're just starting. Imagine our world started with, for the last 50 years, take a photograph. Take a photograph. Take a photograph. Now we're going, let's turn that photograph into 3D and start interrogating it. But now we're going to be able to automatically create diagnostics, automatically create recommended treatments, automatically produce whatever else. Change your lifestyle, change your diet, inform your relatives, inform your lawyer, whatever that, all of these things, there's just so much potential. So I think that's on one hand. I think if I look at then on our customer side, What this opens up is massive opportunities in our healthcare systems for solving the fundamental problem that virtually every healthcare system in the world. We all get older, fatter, sicker. That's where we're all going. We live longer. The cost of healthcare grows incrementally every year, compound, 7%. There's no economy in the world is growing their healthcare budget, whether that's privately funded, publicly funded or a blend at 7 % compound. So if you're in a lot of the private economy, privately funded or majority privately funded economies like the US, we all hear the stories about healthcare is unaffordable. In other maybe ⁓ government funded healthcare systems, we're in danger of moving to where we only provide the healthcare we can afford to provide. as opposed to what our citizens need. What we've actually got to do is with the same resources be dramatically more productive. And that's what AI, not by itself, but actually with people who understand how to build it in safely into these ecosystems can deliver. So I would turn it away from, there will still in my industry, whether it be from a ⁓ provider point of view, a software or technology provider point of view, or the...

Michael Finocchiaro

Right.

Roger Johnston

on actual service delivery, the healthcare side. I'm sure there'll be different needs in different places. I don't see a fundamental ⁓ reduction in staffing. I think what we've got is we've a combined obligation to deliver a lot more healthcare based on the resources we have and that's the opportunity. So I think for any young person, if you have a medical bent or an AI bent or a combination of the two, This is an awesome industry to be going into.

Shahroz Khan

Yeah, I will answer this question in two folds, first generically and then secondly, more related to the company itself. And I would start with the first one saying that whenever there is a technological shift, right? Yes, one sector loses an opportunity, but a new opportunity is gross, right?

Michael Finocchiaro

And Saraz, you would say the same thing for boats?

Shahroz Khan

So let's go back and look at the era when everything used to happen by 2D drawings, not in only in, let's say like in naval architect, marine engineering, but also in aerospace and automotive, whatever engineering direction, everything used to happen. Then the CAD came, right? The CAD came a lot of, I'm pretty sure a lot of draftsmen, they lost their job, but a completely new field opened, right? Next we went on towards building simulation tools, right? So instead of conducting physical prototypes, went to doing validating everything digitally. And that's where a new set of opportunities came. So when we went on digitally, that means different opportunities from different sectors came. Then there was engineers actually running code to build CFD tools, software engineers learning the maths behind CFD and the physics to help with those CFD tools. So this is one direction. And now I also believe so now coming towards this programming point of things. And I think, yes, a lot of companies going to use, let's say, agents and AI to write code. But then we certainly going to see an era where there will be a new way of coding that would be specifically designed for LLMs to understand better the context of the code. Right. So the way that we have built C++, the way we have built Python, the way we ⁓

Michael Finocchiaro

Hmm.

Shahroz Khan

built have any programming language was built to for the humanese. But now the backend, whatever happened processing is in term of ones and zeros is completely different, right? So we built for humans. Now there will be a completely different spectrum of code writing because code writing is to give instructions to the computer, right? So we will have a completely different way of writing code and that will be better understandable by LLMs. So a new way of writing codes will gonna emerge. So that's how the programmers gonna adopt to that. Academia gonna adopt to train the next generation of propped up programmers. So having a conversation that they gonna left alone, I think this is not the case. Yes, there will be a ripple effect where some gonna see immediate effect, but there's a new set of opportunities gonna emerge. So now coming back to the computer maritime, that's why we focus a lot that. That we what we call is It's on the base of this human AI interaction. For Overtool, for Neural Shipper that we had developed, at the end is the user, it's the naval architect who is the sole decision maker, who is driving the tool towards a certain direction. Yes, it is helping them to expedite their process, it's helping them to expedite the line. At the end, the final judgment will be based on the user. So that fundamental aspect of having a naval architect in your team will gonna stay there. So for another example is that although we in a company were using a lot, let's say Claude and all the softwares, but that doesn't mean that we're not going to have any program in our team. Right? Right. So that's what I'm We will still need someone to understand the logic, right? The way that we code. It's the logic will still need to be there. Yes. Until we achieve a, ⁓ AGI and then we will have somebody and just, just don't care about the human context.

Michael Finocchiaro

Okay. Right.

Shahroz Khan

But then to create that, I would say generic models, then we would need more data. So the way that data come to train those type of models. So I think there's still a long way I had to see that. But I'm saying that while we're moving that there's new more opportunities going to emerge. The only thing I can say to the younger generation is that always be open to learn new things. That's the very important thing. Always be open to the new things.

Michael Finocchiaro

Absolutely.

Shahroz Khan

Just watch a YouTube video how to implement whatever field you're from. Just watch the video how to implement a cloud agent. That's it.

Michael Finocchiaro

⁓ Are you guys both as bullish as you were at beginning and now in 2026 after four years of this craziness?

Roger Johnston

Yeah, absolutely. I think we, you know, there's both card and stick in our industry here. I think that the genuine enthusiasm for what's possible. And we're now starting to see, think, again, just the necessity of behaviour of regulators in our industry. They're pretty conservative bunch. So the healthcare industry in the broad tends to be a relative, a late majority, probably a doctor. But we're now there, and now what the big focus is, where can we get the big returns? Where is it safe? Where can we get real scale outcomes? I think the stick side is we've no choice because otherwise we don't have enough capacity in our healthcare system or it's unaffordable. So, you know, that's a very powerful ⁓ driver as well as the carrots.

Michael Finocchiaro

Exactly, yeah. But you sure as, you're obviously very bullish.

Shahroz Khan

I would fully agree with Roger on that too. Yes, I I think what I said, I used to think maritime industry is very conservative compared to other ⁓ fields of engineering. There is aerospace and automotive. But whenever I talk to someone, even from automotive, they say, automotive is very conservative. And I was at an event.

Michael Finocchiaro

Mm-hmm.

Shahroz Khan

in UK. don't want to name the event, but I was in an event and I started talking to the potential customers and they was like, they haven't started using CFD in their teams. So they are still using empirical methods. And they were like, how do you expect me to use such an advanced tool? Like, I am still not using, I'm still based on those empirical methods that were developed in 1950s. I was like, what? But at the end, were then the customers, those who were very interested.

Michael Finocchiaro

Good. Hmm.

Shahroz Khan

those who are very interested, those who are the ones who actually wanted to push the domain. So I think now there's like as the time process, so they're going to be like this balance where people will try to understand. But of course, like some would be still like, okay, well, let's see, you know, what happens. So I think the time will tell, but opportunities as Roger said is tremendous. Every day we're ⁓ opening up a new, I would say, interesting idea from the team. Well, you know what? We haven't explored that. We could actually use our technology in that direction too.

Michael Finocchiaro

Right.

Shahroz Khan

Well, you know what, we could train another model on the top of that and then actually could solve also this problem too. So I think like this horizon of possibilities is very exciting things that what we can do with it.

Michael Finocchiaro

That's a good segue into ⁓ the next and last section of the talk, which is more on digital maturity. ⁓ In both your fields, ⁓ I think they're pretty notoriously poor ⁓ in terms of digital maturity and discrete manufacturing, where I'm mostly focused. I think of a scale of one to five, one is Excel and email, which is pretty much the state of the art for a lot of people. Five would be a fully autonomous digital twins, right? You do this, an X-ray and boom, you already know exactly what's wrong with the guy and here's the treatment. And in fact, the drugs are popping out of the wall. Here you go, take this and go home. ⁓ We're not there yet. We're kind of far. I'm wondering, ⁓ on a scale of one to five, where do you think we sit now in the medical industry and the maritime industry in general?

Roger Johnston

Yeah, so I think we're at very clear point in healthcare. I think we are a one, two, three, four and five. I think it's a remarkable industry. So I'll give you the ones. So in large parts of the world, the way that orders are placed, it uses this technology that some of the folks watching this will know. It's called FACS. it is still, please, if you don't know what it is, look it up.

Shahroz Khan

You

Michael Finocchiaro

You saw that episode of the the the pit where the electricity goes down in the ER room and they.

Roger Johnston

It's bizarre. It is absolutely still at the core of many of the big healthcare systems. It is staggering. In terms of transferring medical images, let's say from one provider to a medical device company or something, the predominant standard, it's much more modern to be fair. It's called the CD. And so these

Michael Finocchiaro

Sax machines, yeah.

Roger Johnston

medical advice companies have teams in rooms who receive CDs and then stick it onto the network. And you know, but they've learned about this thing called the interweb and they're moving that way. So they're going to move up to two quite soon. At the other end, we see lots of instances of five where the technology is remarkable and whether that's the software or the robotics or whatever.

Michael Finocchiaro

Hahaha! Ha ha ha ha! Hahaha

Roger Johnston

The challenge then, if I go so we've got those two extremes, but in healthcare, having pilot projects or showcase projects or whatever it happens to be, where only those in Beverly Hills get to try it out or only those who can afford massive amounts of money, that's not healthcare. That's proving what's possible. What we've got to do is see where healthcare is for the mainstream, for all of us, whether we've got our own medical insurance. or are health care publicly funded? It doesn't matter. For the 99 % of it, where are we? And I think unfortunately, health care is probably at best of three. It probably depends on which country we're moving towards actually having a standard medical record. So if you live in Bath and move to Bristol, the new GP can still maybe see your medical history.

Michael Finocchiaro

Yeah.

Roger Johnston

And we're even moving in the UK to where you can see your own medical record, which is obviously a huge step forward. But not being so cynical, I think we've got to find a way of adopting technology faster. I don't think healthcare ever gets to a five in the broad adoption. I think we'll always be a laggard because of this over need for regulators, as we say in Ireland, to be sure, to be sure. So we've got to have always airing the side of caution on deployment of new medical devices.

Michael Finocchiaro

and I'm the maritime industry.

Shahroz Khan

I would say it's more or less a similar story. There's nothing much different. would just one thing to add on, let's say an example I would give. So I was talking earlier about this class approvals, right? For any vessel or any board, any yacht that is built, it has to be class validated. It has to satisfy all this sort of, the design has to satisfy all the certain set of class rules and stuff. So whatever happens, so when a design team designs us like ⁓ a ship, and then they can gonna convert that ship into 2D drawings. So they're gonna submit to the class and the class gonna use those 2D drawings to reconstruct the 3D geometry to see to further validate. I mean, so we're still based in this 2D world, right? And another example that I just gave you, like a lot of people are still not

Michael Finocchiaro

Yeah.

Shahroz Khan

they're still based in empirical methods. CFD is like for any design team building any sort and shape of technology, finite element analysis. CFD, like any physical analysis, any digitally driven physical analysis is a must, right, to drive. And then having these low level empirical methods, which are not accurate at all. They were built years ago and new designs and other this and this this tries another example that

Michael Finocchiaro

Absolutely.

Shahroz Khan

The innovation is not happening, right? If we're still validating over systems through a technology that was built years and years ago, that means the innovation in terms of design is not happening. It's still creating a lot of bottleneck. So this is exactly the similar story, even in the maritime too. Maritime covers more than 90 % of our global ⁓ trade. Anything that you see around the laptop that I'm... having this conversation, the phone, everything comes through a ship. And at any given moment, there are more than 100,000 commercial vessels. It's an astonishingly big number. The only problem is the journal public not get to see that number like aviation, like automotive. So that's the reason that this industry tend to be doing what they started doing like 50, 60 years or 100 years ago.

Michael Finocchiaro

ship. Right.

Shahroz Khan

And now they started to become one of the most polluting company. And if shipping was a country, its environmental impact would be similar to Germany. So you see, like, there's a lot of factors that has to be taken into account. But this industry is so conservative, still based on this 100 years ago of techniques and approaches and the way of doing things, it's still very hard to break it through. But but now the regulatory bodies like IMO, International Maritime Organization and EU, has started to to implement strict regulation. And those regulations are driven by financial penalty.

Michael Finocchiaro

Okay.

Shahroz Khan

So, and that's where all of the sudden industry started to shake and a well, you know what, if we don't do something, then then we're going to lose in millions. And I think that's a good driver. Not sure like how the severity. I don't agree with that, but that's a good driver to at least force the stakeholders to look for look for better solutions. And that the search of this better solutions. is driving the industry towards more modernism, I would say.

Michael Finocchiaro

So speaking of better solutions then I like to end, well almost end on the question, which is my thesis is that using leading edge technology like compute maritime, like Axial3D is a better on ramp. It's really a much more effective way of moving your digital maturity towards the right, towards a 304 of a five, then the old ⁓ fart, you know, traditional legacy platforms, right? ⁓ Have you seen that? Have you seen where a customer was relatively conservative and things weren't really moving. And then they implemented compute maritime where they started using actual 3D and boom, there was this epiphany like, holy cow, if I did things right, if I thought about data governance, if I thought about AI and doing things right and not doing the old way, geez, I would do so much more benefit. It would be so much more better. The patients would be better. The boat. So have you ever seen that with a customer, with your customers? Is it really a big, go for it. Sorry.

Roger Johnston

Yeah, I would say in two ways. If I look at the

Shahroz Khan

you

Roger Johnston

incumbents, the big gorillas and healthcare is dominated by as many big gorilla players, whether it be in medical imaging or medical device or whatever, they all are still in competitive markets and therefore certainly the leaders of those gorillas, where they're competitive categories,

Michael Finocchiaro

Mm-hmm.

Roger Johnston

they are really full on trying to work out how they can use AI to fundamentally gain competitive advantage. And they realize that their industry is quite slow moving at times. So actually if they move at a reasonably quick pace, they can get a lot of competitive advantage. However, I think the really interesting thing and probably across many sectors is where it's really, historically has been really hard to break in. and become a tier one player in some of these categories. The opportunity now, if you're born in not just AI, but the latest generation of AI, and that's not just in terms of how you build your offering, but in terms of how you envisage your business, how you build your go-to-market plan, how you segment the market, how you target your first customers, how you scale your business, how you do your marketing, how you do your sales, how you do your customer support. If you're born in that capability and that way of thinking, you can build product that's fundamentally reimagined. You can build a business model that's fundamental. You're not trying to fix a last decade business model or something like that. Or in Shiroz's case, a last decade or a last century business model. So I think that's, I absolutely do see it with with the incumbent.

Michael Finocchiaro

Sit free.

Shahroz Khan

That's true.

Roger Johnston

But I think what we should all be really looking forward to and excited to is genuine companies we've never heard of before that in five years are leaders in these categories.

Michael Finocchiaro

Absolutely.

Shahroz Khan

Absolutely. think I fully second with Roger and what I'm finding it very interesting how I can correlate ⁓ maritime industry with medical. And this is exactly the similar story to your answer, Mike. Yes. The early on customers, we focus on the smaller design offices, design firms. Those are focusing on, let's say like smaller boats.

Roger Johnston

Thank

Shahroz Khan

leisure yachts and offshore industry. because the risk is quite small. And these are the ones, those who are, let's say like are more into their bubbles compared to those, those big ones, the big three, what we call them, which I'm going to come back to later. So yes. So then the problem is one thing is that you have to do a lot of convincing to those, those companies that you can actually drive and the time that takes to actually, bring them that level of confidence that costs a lot of resources for a smaller company. like us and the company that are still evolving to build this lot of technology, where still a lot of day-to-day activity is very R &D focused and R &D driven. once they have finds, that's why we do a lot of case studies with our customers at the primary stage. But when they see the benefits, yes. And then they're going to come back and ask you, can it does that? Can it do that? Can it do that? That's another issue, that, right?

Michael Finocchiaro

All right.

Shahroz Khan

So when the students suddenly see some benefits, then they start asking, can I do this? Can I do this? was like, well, hold on. Hold on. We just show you the proven case. But this also helps us to also to improve our product. And then when I was saying we were focusing on the API and the building of technology, that's also relatively driven by their needs, that we actually see that's more and more features that we can drive to do this. And then coming towards is this big joints, let's say, Hyundai have industries, Samsung have industries, Hanwha Ocean and let's say a few other big shipyards and design firms in Japan and in Asia. Yeah, they have the capacity to do that. But the real benefit we will see like when these companies that we have never heard of, they will come and they will start being the market driver.

Michael Finocchiaro

HHRA instead.

Shahroz Khan

by using this technology, then we will see, I would say, the true benefit and the value that these tools that we're developing or the Rogers company, Excel 3D is developing, that is having some weight and I would say, yeah, some weight to actually to bring something from those big giants, someone there. So there you see the value in it.

Michael Finocchiaro

Well, thank you guys. That was a really cool discussion. ⁓ know, ⁓ I was just wondering about where we can see you guys the rest of the year. I know Shahroz, we're gonna see you at ⁓ Devout 3D live in Threaded, which is gonna be awesome and work. Where else can we see you guys, Roger? Where can people meet you over the next couple of months?

Roger Johnston

Yes, so probably in the UK, most of the events I would go to are with our amazing partner AWS. we're not that, you know, I'm big supportive of the other global cloud providers, but AWS, we're all in on them. And we're building our ecosystem and platforms and solutions. And I think they have a real understanding in this medical category of what it takes to build a highly trusted application. So probably some of the summits with them and in the US we hang out a lot at the big orthopedic and cardiovascular and oncology shows. But always available online. Always available online.

Michael Finocchiaro

So I've got to come to Dublin to see you then. Okay.

Shahroz Khan

Likewise, we're going to all the big events. of them. So our strategic partner is Siemens. We're actually the early validators of our technology. And Siemens is one of the biggest. I was a digital tool developer. So we go a lot with their events. We have our offices in West London and Chiswick Business Park. Just send me an email. Just pop into our office. Happy to meet and send me an email. Happy to meet online. From the events, we're going to be in the CF Fam or DC Fam. There's a CD Fam that is happening. We're going to be there? Yes, in Barcelona. Okay, awesome. That's great. We're also going to be in the Singapore Maritime Week. All the major, I would say, maritime events, we will be there. Also, in Develop3D Live, I will be there and happy to meet anyone from there and answer any questions. Yes, slowly.

Michael Finocchiaro

CD fan, CD fan. I'll see you in Barcelona. Yeah, I'll be there too. Nice. Okay. Yeah, so, you know, here's the shirt. So, yeah, go to threaded.live and you can sign up for ⁓ for the show in ⁓ Warwick or for Miami on the April 13th. It's been a real pleasure having both of you, Roger and Shahroz. I've certainly learned about both the medical industry and the maritime industry. I really appreciate your time. Thanks to the audience. Thanks for your questions. And we'll see you tomorrow. I've got Intop and Neuroconcept who really huge players that are changing the way we think about design and simulation. So thanks everybody.

Shahroz Khan

Thank you.

Michael Finocchiaro

messed up.