Next-Gen Manufacturing: AI on the Shop Floor — with Manukai and Productive Machines

Episode Summary

The episode titled "Next-Gen Manufacturing: AI on the Shop Floor — with Manukai and Productive Machines" delves into the integration of artificial intelligence (AI) in manufacturing processes through the lens of two innovative startups, Erdem Ozturk from Productive Machines and Pascal Weber from Manukai. Productive Machines specializes in developing digital twins for machining processes, which helps streamline operations and enhance productivity in various industries, particularly aerospace. Manukai, on the other hand, focuses on leveraging cutting-edge AI models to revolutionize CNC machining, aiming to transform how products are manufactured today.

During the discussion, key technical insights emerged, highlighting the potential of AI-driven digital twins for optimizing manufacturing processes and reducing waste. Both companies emphasized the importance of transitioning fundamental AI research into practical applications in engineering and science. Strategically, they discussed their plans to participate in major trade shows such as the MAK exhibition in the UK and the Hannover Fair, showcasing their technologies and engaging with potential clients.

For PLM and engineering professionals, the episode underscores the transformative impact of integrating advanced AI technologies into manufacturing processes, offering new opportunities for efficiency and innovation. It encourages listeners to explore how these cutting-edge solutions can be applied within their own organizations to drive productivity and competitiveness in the ever-evolving landscape of product lifecycle management.

---

Full Transcript

Pascal Weber | Manukai

effect.

Michael Finocchiaro

And we're live. This is Michael Finocchiaro on the AI Across the Product Lifecycle. I don't know why I stumble over that every time. And I'm here with Erdem Ozturk of Productive Machines and Pascal Weber of Manukai. Two pretty awesome startups, as you're going to learn about in a minute. We're very happy to join you today. I'm a little stressed because I just got off a train. was late. And Pascal has a video problem. So we're going to take a nice breath and we'll get started.

Pascal Weber | Manukai

Hi.

Michael Finocchiaro

⁓ So let's start with Erdem because he was the one on time. He gets to do his introduction first. So tell us about you and tell us about Productive Machines.

Pascal Weber | Manukai

you

Erdem Ozturk

Yeah, of course. I'm Erdem Öztürk. I'm the founder and CEO of Productive Machines. Productive Machines is going to be a five-year-old business in March. It's a spin-out from the Advanced Manufacturing Research Center at the University of Sheffield. I worked there for a long time, more than 10 years. We developed a lot of technologies that helps many aerospace companies. But there was one particular technology, which was a digital twin for machining processes. And we took that technology out to the... world of commercial world and since then we have been growing the business. That's a short intro for me.

Michael Finocchiaro

And we'll hear more about what you're doing as the podcast goes on. I just left your country, Pascal. I just came out of Geneva, just arriving in Paris. Zurich's not right next door.

Pascal Weber | Manukai

Yeah, I thought you could have caught came by here that would be have been easier because Well, but trains are on time here in Switzerland at least

Michael Finocchiaro

Well, for those that don't live in Europe and maybe don't know that it actually turns out that ⁓ ETH Zurich, well, I think there's two technical universities there that are turning out these amazing startups. I've already had two or three on the podcast. Manukai is yet another one. So I just wanted to point that out that if you guys didn't know that, I mean, there's like in Germany, there's a group in like Western Germany around Karlruhe and then in England you have Imperial College, Royal College, Sheffield and Warwick. France, well you've got France and then in Zurich is a hotbed and then in Germany it's more or less Munich and Berlin where all the startups are right so pretty interesting. think you're in Germany right Erdem?

Erdem Ozturk

No, no, we are in the United Kingdom, England.

Michael Finocchiaro

The UK in England, okay. Well, ⁓ so Pascal, tell us about Manyakai.

Pascal Weber | Manukai

Sure. Yeah. So hi everyone. I'm Pascal Weber, co-founder and CEO of Monokai. Our story is a bit different. So we studied like fundamental research in applied AIs, AI. So how we can really use AI models that come fresh out of the frontier labs and then basically kind of misuse them for engineering and science applications. And yeah, so after doing all this studying and researching during the PhD, me and my co-founder decided that we want to do something in the real world.

Michael Finocchiaro

Hahaha.

Pascal Weber | Manukai

It turned out manufacturing builds the real world. So that's a good start. And then found our way to CNC machining, where we believe we have the greatest potential to actually ⁓ change how we produce things today.

Michael Finocchiaro

Yeah, but there's a lot of startups digging in that very rich mine of CNC. That's great. So we started the podcast looking back a couple of years. And when you think about, you know, 2022, 2023, OpenAI announces this chat GPT thing. And obviously the people in the know knew there was a one and two, but all of us learned about three. And, you know, the world seemed to change. It seemed to be a bit of a pivot point. Were you guys skeptical or really bullish ⁓ about the possibilities of that? was change engineering and change manufacturing at the beginning.

Erdem Ozturk

Who's first?

Michael Finocchiaro

Go ahead, Pascal, you want to do that one?

Pascal Weber | Manukai

Sure, sure. So, I mean, being kind of an AI researcher, was initially very sceptic. think I still am because I think in manufacturing, language and imagery is not really the kind of modality. Of course, for quality inspection, you can use the image recognition capabilities, but language, I think, is still a bit of a far scratch for being used in manufacturing.

Michael Finocchiaro

Of course you knew about it though.

Pascal Weber | Manukai

So I think still today, right, there is still, I think, big steps needed to actually adopt and create AI models that can deal with the type of data that we deal with in manufacturing and CNC machining.

Michael Finocchiaro

And what about you, Adam?

Erdem Ozturk

Yeah, large language modules were very useful in daily lives. I started using it in my personal life, like get some advice, like review this, review that. It can consume documents very quickly and then give summaries. But yeah, initially, productive machines, technology was not going to be influenced by AI with a large language module. What we have is we have a physics engine. So we basically simulate. the mechanics and dynamics of the process, like cutting forces, stresses on the tool, torque, vibrations, and we work on a very specific vibration type called chatter vibrations. That's the bad vibrations for the sector. I didn't see that AI would have any influence on things. ⁓ For this reason, yeah, we didn't use it, but yeah, as it gets better and better, of course we are using it in the company, especially in the product development to accelerate the development times. But yeah, we have our own solution. I don't know how to call it, but it's basically a physics engine. I think it can be called agenting AI, although it may not be really correct. We have a simulation engine, and then there is a algorithm running that engine. So it's agenting AI we had in the past, but it's not like the agenting AI that is now ⁓ promoted by the companies like OpenAI and Claude.

Michael Finocchiaro

Okay. and N8N. So you both were sort of skeptical and kind of a wait and see mode. However, I imagine that you guys are both using AI in terms of a development tool, for maybe automatically writing PRDs, probably a lot of code also. So first part of that question, so how are you using...

Erdem Ozturk

Yes.

Michael Finocchiaro

or how the developers at Manny Kyn productive machines using AI on a daily basis.

Erdem Ozturk

Yeah, all the developers have access to it. So we have the enterprise version of the solution. So we use Chess GPT, and the guys are using it to accelerate the development time. I can give a very specific example. We started with cloud. So we were cloud native. We thought the future was cloud when we started. And then we developed an autonomous optimization software.

Michael Finocchiaro

Which one, Cloud Code ⁓ or Cursor, Chachipiti? ⁓ Vertex, right? No.

Erdem Ozturk

which is basically enabling any cam programmer to optimize the processes by pressing a few buttons. But then we faced the reality of the market. Market didn't like cloud solutions, so we had to shift everything from cloud-based deployment to on-premise. So in that ⁓ transformation, I know that our guys used ⁓ AI tools a lot because it was standard.

Michael Finocchiaro

The cloud to the pin.

Erdem Ozturk

⁓ processes that the guys followed and without AI tools probably we would have the on-premise installation in a longer time frame other than like three months.

Michael Finocchiaro

Okay, so it's useful to reduce the time it took to get the application out. Pascal, what about at Manukai?

Erdem Ozturk

Yeah.

Pascal Weber | Manukai

I mean, we also use it for the development, especially when it comes to documentation or full request reviews, all these kind of, think, cumbersome processes that you somehow don't want to do as an engineer. And I think that's generally how I think about AI. I think it shouldn't be used to replace, let's say, the core and established work. ⁓ that we know it works and we have been developing and improving it for a long time, but it can be used as an enhancement to kind of maybe, well, do replace some of the bottlenecks or do replace some of the repetitive and boring things that no one wants to

Michael Finocchiaro

But are you guys doing some form of vibe coding where you're really just talking to a chat bot and it's generating the code? Or are you guys holding off on that? Okay, go ahead.

Pascal Weber | Manukai

No, I mean, I mean, my opinion is right. And the same applies, I think, for applications in manufacturing and CNC machining in more detail. Right. I think if you want to produce something production ready, there needs to be a senior engineer there to think about the architecture to fix and create something that really scales because I think wipe coding for us is something that we use for prototyping for user testing, right? You can basically throw at it like Figma design and some description of a use case. And the thing is going to generate a nice, let's say websites that we can then give to the user to actually test and give us feedback on the ideas we had. But I think after that, if you really want to make it into production ready code, it's not yet at the stage where AI can be taking all of it. At least not in these very specialized regimes we are working because I mean, most of the code, most of the data is all proprietary. It's not like standard websites that we are building.

Erdem Ozturk

Yeah, I think it's similar. We don't have AI-generated code in the product. It's more about guidance. The guys in the development team are using it as assistant when they are wondering how to do this, how to do that, so that kind of questions they get. But yeah, we are quite strict about not having the AI-generated code in the actual production code. Maybe it's changing the creation code at the moment. No, it's not in the production code.

Michael Finocchiaro

you Adam. Also in this case, sort of a, it's a. So it's more of a sidecar and in copy paste kind of thing, okay, rather than a full blown environment. That's because it's some of the other founders I talked to, it was almost like having another couple of developers on the team and that made the joke that maybe the scrum will be between a couple of ⁓ agents instead. And I'm wondering too, like in the fact that... It looks like it's not the case for YouTube, but I'm wondering over time if ⁓ waterfall and agile, these processes are not going to morph into something else because we have AI doing a lot of the heavy lifting that you would use a scrum to get the crap out of the way, To fill out the backlog. mean, maybe that's where AI agents will be useful to kind of work the backlog while you're working the new stuff. I don't know, just a thought. What do you guys think?

Erdem Ozturk

I have a specific example on that. in the UK, there's a competition called AI Pioneers Award. We passed the first stage, Expression of Interest stage. Now we are in the second stage. It's a competition available to all the UK-based companies. And basically, one developer is working on it now, and he's using AI tools, agent AI tools. But it's mainly to demonstrate what is possible. It's not going to be a product. Basically, we'll send a demo and there will be an interview in March. We'll present the solution, how it works, to the jury. And then if we are successful, we will get 250k prize money, which is quite interesting. But for that kind of demonstration and prototyping, I think it's great. But it will take a few months to convert it, maybe more even, to a product. Then our guys will be coding. Because the code... We want to have traceability, explainability. Basically, those are the things that we are seeing missing at the moment with the AI tools.

Michael Finocchiaro

Do you have any comment on that, Pesco?

Pascal Weber | Manukai

Yeah, mean, to me too, right? I think in the end, it's garbage in, garbage out that also holds true for these AI models, right? I the degree of specification that you would have to do in order to actually maybe get meaningful output is more or less the same if you... than just writing the code. So I think, yes, it can be useful for summarizing things, or for really doing repetitive things, like things where you have all the information there, but you kind of want to restructure it or present it in a different way. But I think it's not yet there. And I think also even the big labs agree on that, that when it comes to creativity, when it comes to actually ⁓ creating something really novel. And out of the box, it's not in the nature of the training algorithm that it can capture these kind of tasks.

Michael Finocchiaro

⁓ In the chat, Naga Reddy ⁓ had a question. He asks, is anything you're using AI for in terms of validating if the tolerance is applicable to the parts is correct?

Erdem Ozturk

Sorry, can you say it again? I was distracted.

Michael Finocchiaro

He said, anything you're using AI for validating if the tolerance applicable in parts are correct. So I think he means like are using AI to validate tolerances and the machining process, ⁓ using ML or kind of a regression algorithm to determine whether the tolerances are correct.

Erdem Ozturk

OK, maybe I can start to answer. So we have a solution that runs in the process planning stage. We, of course, get some data from the machine. So we have a method called tap testing. We tap the tool. It's basically simple tapping on the tool. And then we identify the initial frequency systems and damping properties of the machine. And then we run our software called SenseNC, which gives it, basically, sense to the NC code. And then. We do an optimization and then we create new operations in the CAM environment like NAX Mastercam. And then we send the code to the machine and the machine does the manufacturing. Current product scope is there. But in the future, of course, we will have a solution that gets the real-time data from the machine. At the moment, we rely on the customer telling us, OK, you achieved X percent improvement in productivity, Y percent service quality improvement. But at the moment, we are not getting real-time data from the machine. We only get ⁓ basic data from our customers ⁓ verbally or by email.

Michael Finocchiaro

But

Pascal Weber | Manukai

Yes, so maybe since we are a bit earlier in the process, I can elaborate directly on that. So for us, it's bit different, yes, because we get basically the technical drawing, so the PDF with the tolerances and in the end, the step file, right, of the part and that goes into our engine. And of course, in order to then choose or suggest appropriate production strategies we need to ⁓ understand what tolerance requirements are given, what material requirements are given, et cetera. this goes in and yes, is AI being used. But also here, think tools today are really optimized for more, let's say, mainstream tasks, as I said initially, maybe understanding documents with written text and we see challenges of using out of the box tools to actually understand more complex technical drawings. And also, mean, the technical drawing is something that is not very forgiving. So if you misinterpret the tolerance, well, good luck ⁓ because you might generate scrap and the worst case, you might even demolish the machine.

Michael Finocchiaro

Interesting. Thank you for that. And thank you, Naga, for the question. ⁓ I'm wondering, too, like another thing I've been thinking about, and sorry, it's not one of the scripted questions, but I think it's still an interesting one, is that it's kind of, I'm thinking like the next generation of PDM, know, design engineers, CAD guys, they use PDM and they have all these CAD models that eventually you guys have to have in order to do stuff. And yet the PDM doesn't understand anything about machining, doesn't understand G code, doesn't understand tool paths. All it understands is CAD. So how can we get it so that the PDM actually understands what you guys are doing? mean, it seems to me it's a little silly that there are two different systems. It just should be the one. There should be an intelligence system that comes in like menu CHI or productive machines to work on the CAD data and give an output. But then whatever the output is, it should be in the PDM system so I can get to it really easy, right? I mean, it's not rocket science. I don't understand why we're not there yet.

Erdem Ozturk

Yeah. Yeah, exactly. think whoever closes the feedback loop will be really successful because I manufacturing is well known. It's like very segregated industry. Like CAD is different, CAM is different, you post-processing. You have machines and then machine controllers weren't very open maybe until last few years, until like five years ago or so. They were really conservative sharing data with the outside world. And then there's cold assurance. like CMM machines, like measuring the dimension quality. Yeah, so at the moment they are all siloed data. I think it needs to be all... Another problem is standards. There is no standard of communication. There are some standards coming, but again, it's not at a state that everybody...

Michael Finocchiaro

Right. Well, that's what agents are for, right? They're the ones who figure it out. Just give it a multbook. We'll create a multbook page and let the agents spattle it out, right? Pascal, you seem to have something to say.

Erdem Ozturk

Yeah. Yeah. Yeah. Yeah. Yeah.

Pascal Weber | Manukai

Hahaha. Yeah, I mean, in the end, right, I think it's a common problem. It's not just in manufacturing, right? You have the designer and then you have the engineer building the thing or the production realizing the thing, right? You have it in architecture. The architect draws the house and then someone has to build it, right? It's the same in manufacturing. have basically the designer and then you have the machine programmer setting up the thing for running. And I think it's just in the nature of the thing. It's kind of two different... ⁓ task somehow to think about how something should look in order to be functional for the design task. But it's kind of detached from whether you can or how you can actually create the thing in reality. so maybe AI can help that a single human can be capable and perfect in both worlds. So being a perfect designer and at the same time being the perfect engineer. But yeah, I believe there is still long, long ways to go until we are there. Because it's just a little kind of.

Michael Finocchiaro

And I didn't even throw in the other piece. I didn't even throw the other piece. I think it's required as simulation. I mean, when you're doing FENCFD, why are they different systems? In terms of the file management, why are they different? There's another question from Dennis Bealgilly. I hope I spell that correctly. He says to you guys, which is a nice company, he so fresh in this year, both very intentional with your approach and language models for using for the sake, and not just using it for the sake of using it, right? You're not just using AI because it's fun. So his question is, ⁓ if there are currently four to five cam automation optimization, one click done types of solutions emerging, what do you think will be among the crucial differentiators? I think he's talking about cloud and C, limitless, digital CNC. ⁓ What do you see as the crucial differentiators in a market getting rather crowded? In your experience, is the direction of product offerings aligned with what the customers need? Or is it just, again, AI for AI sake?

Erdem Ozturk

Yeah, I can give a very specific example on that because it's part of our history story. So we had a product that was developed in the university environment in my team. And that product was implemented in big companies like Renault, like in France, and then Mazda Aerospace in Spain. And then we continuously developed it further and then we implemented this technology in the partners of AMRC and many aerospace companies. And then we thought, okay, now we have the product, go out. So we got investment from Boeing and Boeing said, okay, go and sell this solution to my suppliers. And then my suppliers will be more cost competitive. And then I will gain, my supplier will gain because they will be competitive. And then you will gain because you will have customer. And then when we talked to the suppliers, the recurring message that we were hearing was we talked to many suppliers of Boeing. They said, this is great technology, but we don't have the engineers that can run this. We said, what do you mean? We don't want to train engineers on it, and then we don't want those engineers to disappear to another company. And then even if they don't go to another company, they get promoted, and they become a manager, and then they need to train another person. And basically, they said, give us something that any program can use. Like ease of use was the main thing, because there is also a skills gap problem in the UK. I think it's general for the whole. modern nations or manufacturing nations. And basic ease of use is, I think, a critical factor in terms of differentiation. And of course, trust. ⁓ We learned that trust is very important. So whoever has the most trust of the manufacturers, then I think they will be more likely to win.

Michael Finocchiaro

Very cool, I like that answer. Pascal, do you want to field that one?

Pascal Weber | Manukai

Yeah, I mean, in the end, yeah, I agree with a lot of it. think interpretability, safety is super important. This is going to be, I mean, it's a must have. And then I think privacy is another thing, right? It's, think, steel manufacturing is an industry and it also makes sense, right? Because, I mean, everyone can steal a design, but if you don't know how to produce the parts at scale, then it's useless. So What you really want to protect is maybe your really production processes. So I think having a solution that doesn't strictly preserve privacy will be a no-go for large scale adoption. And I think the third one also goes into the direction of ease of use. think A, needs to be very tightly integrated into the existing system. So I think engineers don't want to shift the workflow they have been using for 20 plus years. And I think also. It needs to be ⁓ kind of addressing all the intricacies of the different industries, right? So manufacturing has very many of these very specialized niches and the solution needs to work for basically all these niches because otherwise you're still just going to be kind of well ending up as a solution for like the broad masters, which is probably not never going to allow you to climb the Olymp to the top, Where the kind of large... really large OEMs and leaders in the market live.

Michael Finocchiaro

I'm that's a great response to it. It just makes me ⁓ think that maybe well, I know there's a ⁓ guy named Andy Fine has I think all the fine physics consortium where he's ⁓ helping ⁓ simulation companies. It chooses very carefully as becoming sort of a block. So you've got a group of partners you can work. Maybe we need that in the manufacturing space too, so that you guys can go and have a sort of complete solution so that you don't everything just doesn't go to NX cam and to tell me. All right, because that's Sort of, ⁓ that's more of the competition, I think, than anything else. It's gonna be ⁓ Sandvik, ⁓ and now Cadence, because they bought MSC. Anyway, it's gonna be an interesting couple of years, I think.

Erdem Ozturk

Yeah, I think your comments made me realize that distribution also is very important, like partnerships and distribution, because it doesn't make an impact. If you have a great product, it's very easy to use and your customers trust you. But if you are not distributed globally, then of course, who will know your solution if they are in a different country, for example?

Michael Finocchiaro

Yeah. And you won't have much to say in the growth phase of your VC pictures. Sorry, that's good.

Pascal Weber | Manukai

Yeah. Yeah, I mean, and also, think not just distribution, but even access, right? I mean, here we are again, we are talking with CAM, we are talking about proprietary data. It's not a text file where there is, I don't know, a written piece about how the thing should be manufactured, but this is a binary file that you cannot understand and read if you don't get API access. And so it's really, I think, quite a big determinant whether you get to partner with the big CAM. players and they give you access via their API, give you support to, while also make your product robust and not just a kind of prototype. this, yeah, I completely agree that this is a very, very big component to the game.

Michael Finocchiaro

Okay, so let's switch to the next question, which is, ⁓ in your product, ⁓ there's obviously a little bit of AI somewhere, but is it something that the user of product machines or menu guy sees because he's got a copilot and AI thingy chats to? Or does it underpin more, is it more the plumbing of this is how I'm moving data around and making inference and figuring out what the next step is going to be?

Pascal Weber | Manukai

I mean, I would disagree with a little bit of AI. would maybe, I think, put it a bit in perspective, right? I mean, the word AI is by now ⁓ very much misused. Most people think about large language model when talking about AI. I think AI is much more than that, right? And that's what I also said initially, if we want to tackle manufacturing challenges, we need models, specific models, specialized models for dealing with the type of modalities you find in manufacturing, which is 3D models, which is technical drawings, which is in the end machine data that can be super noisy, very high frequency, a lot of ⁓ data, maybe Adam you can elaborate on that, but that's different challenges. This is not something that LLM will capture. And I think all these components are all, I think, fairly put under the umbrella of artificial intelligence. And then I think what we believe is crucial in any use of AI and is something that I think we also see when using chat GPT, but it's a feedback loop with the user. mean, you need to somehow validate whether your output and the things you generate are correct. And you need the human to tell you that because only the human expert can determine whether a machine program makes sense, whether, I don't know, chosen strategy in a cam makes sense. And then AI cannot, I think, do that charging because you cannot use the system itself to charge itself. So, and I think that's a critical component and that I think the user will also kind of see and get access to the actual learning of the AI because it will be able to actually tell the AI, Hey, this thing is not as it's supposed to be. Please go back and train and get back to me once you figured it out and they are able to actually solve the challenge that I just gave to you.

Erdem Ozturk

Yeah. Yeah. My answer to that question is like, machine is really like a complex ⁓ process. Like there's a lot of parameters that affects the outcome of the ⁓ process. ⁓ Pascal mentioned that like, you need expert to make decisions. I agree with that. Where my view differs a bit is, humans are great in some things, but not great in other things. For example, I have...

Michael Finocchiaro

Interesting. Erda?

Erdem Ozturk

22 years of experience with machining research and also development since 2003. If I go next to a machine and then if there's a new part on it, I won't be able to tell which parameters will work best for that machine. I can only use my experience, which may have accumulated in 22 years, but that experience may not necessarily translate to a new part, new material, new machine.

Michael Finocchiaro

Or a new machine, right?

Erdem Ozturk

At that point, physics comes into play because physics can guide an expert engineer or any expert engineer in the same way. So when there is uncertainty or unknown, humans are not great in making the best decisions. And in our experience, usually they do a few iterations and then they say, okay, this is good enough. Let's continue producing this part like this for next five years. Of course, they do continuous improvement every year. They improve the process maybe 3%, 4 % every year. But after five years, they get 15 % improvement. And what we have demonstrated is in a week, they get 25 % improvement rather than five years. So a human cannot say, OK, these are the best parameters. Maybe they will be satisfied with the local optimum, but global optimum is quite difficult. It's quite a difficult challenge for a human.

Michael Finocchiaro

So, just staying on the topic for a second, do you think that we're gonna start developing some large machining models, large toolpath models, like specific AI models that are going to learn from all that experience, all that stuff? Because I think the fear that a lot of people are figuring out in manufacturing is that the guys that know how to use those machines, the women that are on the thing that know the right parameters, they're going to retire, they're becoming gray haired and they're leaving. And when they go, these young, a young kid, a millennial who's only looking at his iPhone all day will have no clue how to use that machine. And then what are you going to do? Right? So what do you think? Is that some of you guys are working on? Is that something that you're hoping somebody else will come up with? You know, where do we sit on having models that are specifically trained on that kind of data?

Erdem Ozturk

I can, yeah, so we did a project called Coroma. So Coroma was before Corona. So there was no Corona before Corona. So it was about developing cognitive robots for machining applications. So we had the task of drilling. Robots are good for drilling processes. And our task was basically to increase the cognitive capabilities of the robots for drilling applications. And imagine robot is drilling in aerospace applications, there's hundreds.

Michael Finocchiaro

Go ahead.

Pascal Weber | Manukai

Ha ha ha.

Erdem Ozturk

maybe thousands of holes you need on the parts in an airplane. And basically our idea was we will drill a lot of holes, but every time you drill a hole, you of course collect data like torque, power, vibration behavior, and then you quantify the quality of the holes. And then you will have a database of good processes and bad processes. And then the simple approach was If robot does the same hole in two months, robot will disagree with the operator to do that because he says, you did this like two months ago. You shouldn't use the same parameters again. So there is this self-learning robots or self-learning machines. It was very popular in the academic world 10 years ago. Still, I think they are popular. But in the commercial world, in the industry, we don't have that solutions yet. But that's going to come. Like, for example, usually, The process level of information is just basically, as we discussed, siloed. They just check what's happened in the past. So in the past, by looking to the dimensional qualities or surface quality. But there is a lot of data that's coming from the process, in process. Like we have a lot of data post-process, which is like dimensional data, surface quality data. But in the process, we have a lot of data as well. So that data is not utilized well enough now. But I see that there's a big opportunity there to utilize this data. so that machines can learn. As I explained, having talk data, power data, vibration, that will give us lot of information about the cost of the process at the process rather than after the process. And there's a big opportunity there.

Michael Finocchiaro

Let's go.

Pascal Weber | Manukai

Yeah, so I actually that leads me to start telling the story from my research times, right? I think it was one of the projects I really liked a lot, right? So what we we we what I looked at is basically the question whether you can like if you have a system that can be trained to do ⁓ separate tasks, let's say it can do three tasks, right? And then kind of make sense that you can also train the system to do all the three tasks. And the question was, okay, which one now performs better, right? It's the situation that because you know now three tasks, you kind of get some information from all of the three tasks. And then you kind of also become better in the other tasks or is it different? it because you have like kind of lost some capacity because you only can kind of cognitive abilities only limited or also the ability of the model to capture. information is limited to only a certain thing. And if you focus on one thing, you maybe get better in this particular thing compared to being able to do multiple things. mean, it turns out it seems to be the case that the specialized model always outperforms a generalistic model. I think it's something that we see in the topics we discussed before with the designer and then the programmer of the machine and or the architect and the construction ⁓ worker. I think it's something that also holds true for these kind of algorithms or AI models that, OK, yes, you can maybe train a model that will unify all these NC code and then can maybe understand and interpret a lot of NC code. But I think a model that will kind of be trained specifically for creating NC code for a specific application, for example, in aerospace or any other domain, will always outperform this generalistic model. So while I... It would be great if manufacturing opens up a little bit and we get more public data in order to do more research and develop better AI capabilities. I believe, especially in a specialized kind of domain like manufacturing, of all these very specialized verticals, it will not be in the end the winner to have this more generalistic mode.

Michael Finocchiaro

So now that we're sort of four years into the LLM chat GPT kind of world, and we've seen the ⁓ MCP stuff last year, and then a few months ago we had this insane clod bot, mold bot, the hell, open claw now, right? ⁓ Are you guys as skeptical as you were? Because you guys were both more or less skeptical in the beginning. Are you guys skeptical? You think that we're waiting, know, there's gonna be a forest fire as one of my podcasts people said, or, you the bubble burst. and a lot of these startups are going to disappear. you think that AI is here to stay? Where do see it's heading? What do you think is the next phase?

Pascal Weber | Manukai

So the article I always reference when it comes to this topic is, I think the analysis comparing kind of the big like dot net bubble we had where I think the Cisco stock was kind of the leading indicator. And I think the argument that was made is basically you saw, the Cisco stock going through the roof, but the actual revenue. So the actual value creation of the company not really growing at that pace. And I think it's something different that we see today because Nvidia stock, okay, is going through the roof, but their sales are too, right? That's actually infrastructure is actually installed and all this compute will stay, it will still be there even if let's say there is only 20 % of the AI companies remaining, right? And then you have infrastructure, so you need to do something with it. And then, I mean, I think also add them with the physics models, right? If it could run, I don't know, 10,000 evaluations in parallel, probably the output will become better. So I think it's a huge opportunity where the established and kind of, would say, the winners of these now, say, kind of AI race will be in a very good position. And I think there will be a lot of value generated because, again, right, we install actual infrastructure and this actual infrastructure is a value driver.

Erdem Ozturk

Yeah, very good point, Pascal. Yeah, so I think I will stay. And also, like as Pascal mentioned, one advantage is the infrastructure is getting better and better. It's an incredible scale. And in the past, we were doing simulations, optimizations. And if you are working on a complex part, it may take a day. Like, you do like, like you mentioned about CFD, CFD is like one of the most time consuming or computationally heavy solutions. We basically have semi-mechanistic methods. We don't position ourselves as a finite element solver, but in that we solve some differential equations. There's mathematics behind it. And imagine we are calculating forces and vibrations at every few millimeters. And sometimes toolpaths can be kilometers long. There are toolpaths that we have seen that takes like 400 hours just to machine the part. And imagine if you are doing simulation and optimization every few minutes, it can really take ages. But with the infrastructure increase, like with AWS, like now we had cases, like 300 operations on a part are all running in parallel, and then you get a response in like five, 10 minutes. So that's what happened. I think that will get better every year. Yeah, because the capability of competition capabilities is increasing. I think the people will be ⁓ having quicker results. And I remember a discussion with, ⁓ maybe I shouldn't mention the name, but it was a reputable chem company. said the principle was basically, why don't you get it now? You want to press a button and then we want to have the results. I said, we are not there yet. Like imagine we are pressing a few buttons and then our software is running on the cloud or on the premise server and we give a response in five, 10 minutes. But the expectation from that particular company was why don't you do it now? It will take some time to reach there, but yeah, I think we will reach there at some point, but I don't know why.

Michael Finocchiaro

I can see quantum computing when it comes of age. It would be a massive use, right? Because if you're solving all these simultaneous equations across, that sounds like a quantum kind of problem, really. In fact, I think I'm going to get someone from ⁓ the Quantum Lab in Florida to talk to Threaded Conference in Miami. So that's kind of cool. ⁓ So let's switch gears and talk a bit about ⁓ the real world. So when you guys ⁓ go into your customers, ⁓ And I know machining is one of those areas that particularly is lagging, but I think of digital maturity on a scale of one to five. ⁓ One is that people are still using email and most of the stuff they're doing is in Excel. ⁓ Five is basically the autonomous agentic ⁓ digital twins, which nobody is at. ⁓ The customers you're going into, are they closer to one, closer to three, somewhere between one and two or two and three?

Pascal Weber | Manukai

I I'd say it depends a little bit. mean, in Switzerland, I would say most of the really good established, profitable, well-run companies, they are quite high when it comes to automation. It's to me also always a bit funny, right? Because you have all this, let's say hype around also robotics, right? Where it's always like, yeah, I can solve all these manufacturing use cases, but there the reality is it's all already automated with robots. I, it is these companies, they basically work like humans work a third of the time. And, and, and, and the rest is basically all fully autonomous. So yeah, maybe it's somewhere in between, right? Okay. They're still doing this. But between one and yeah, yeah, probably around the three, don't know if five is really fully autonomous, right? Because they do one third. I still manual work and then the other two thirds are basically already fully automated.

Michael Finocchiaro

Between what? Between one and three?

Erdem Ozturk

Yeah, in our experience, large companies have expectation that they want to preserve digital threat. So those companies are leading, but again, I don't think they're at five, maybe four, three, max. And again, with the feedback that we are getting, for example, there were solutions that solves the problem in a siloed way again, but keeping the digital threat was the challenge for the... for the particular solution that we saw. And for that reason, we really paid attention to not to break the digital threat. So everything we do keeps the digital threat ⁓ traceable. So those companies are leading. But of course, we also want to work with ⁓ smaller machine shops as well in those companies, of course. ⁓ the digital maturity is low because they don't have the resources of the big players. But what is worth mentioning is when people see the effect of the technology, I think it just starts changing their perspective as well. So they see what's possible and then they can start question. Because if you don't know something, it's unknown, unknown for you. you cannot go to... You don't know what to ask, but once you start seeing the benefits, then probably it's...

Michael Finocchiaro

You don't know it, you don't know.

Erdem Ozturk

increases the curiosity and they go into maybe two, three, four, maybe five at some point. But it will take some time, of course, for the whole industry to go to five.

Michael Finocchiaro

Well, that's a great segue into my the other question, which is ⁓ my thesis is that the customers that want to move ⁓ further to the right, that one of the best ways to do that is to use startups like menu kind of productive machines because you're using physics informed machining physics, informed software ⁓ using AI. And are you seeing that? Are you seeing that customers where, you know, they're initially kind of, you know, very manual and then they start using menu, kind of productive machines and You see the needle literally moving to the right because they're starting to realize they need to break the data silos. They need to have data governance. Teams need to work together. The engineers are not your enemy. They're actually a friend. And they could probably, if you work together, you could probably get rid of some of the problems that you're consistently finding on the shop floor. ⁓ Have you actually seen that in the real world? That kind of thing.

Pascal Weber | Manukai

Sure, sure. mean, it's part of the puzzle, right? It's, I would say another component that enables them to be ⁓ more automated and yeah, it's, it's, it's a single piece step after step, I think. Yeah. mean, manufacturing always has this kind of bad image of, of being still very manual, slow, whatever. But yeah, I think it's, it's, it's not, really true. I think in many aspects, I always like kind of putting up the thesis that the CNC machine is the first real commercial use case to computers, right? mean, back in the days, we had this thing that could do computations and it was not yet even inside that we're going to have personal computers and smart watches or whatever. what use case was, yeah, let's move tools in a very precise manner and I think this innovative kind of drive still persists. It's still a very engineering-led community. And engineers, I think, are generally very open to new, to try new things. Just, think, maybe the reason why it is kind of still there, this kind of perception, is that it's just super hard to... Bring real value in an already very optimized process setting right as I said, right the third of the work is manual and the rest is already automated right this is I would say are you a quite high degree of automation compared to let's say the Law firm five years ago where things were still printed emails are still printed and put in some folders and manually by human human labor, so ⁓ I think it's a bit, I think we are on a good path and with the new technologies we're gonna kind of do another leap forward.

Michael Finocchiaro

Well, this is sort of a veiled advertisement for you guys, right? Because you guys are a very valid alternative to the solutions out of Sandvik and DS and Siemens and PTC. mean, but how do you position yourself so that you're showing that difference? And because they're going a lot slower than you on AI, how do you guys use that as a leverage, as a wedge to say, hey, know, obviously the advantage is using our solution because... You can go faster. We're going to learn more, the stuff you can't do today with CreoCam, NXCam, Delmia, et cetera.

Pascal Weber | Manukai

I mean, I would very much argue against this like competitive positioning you're doing. think again, it's all part of the same story trying to get manufacturing forward. In the end, the only thing that counts is the real value generated at the end customer. And whether this is done by an emerging startup that might be a bit more innovative or faster or by an established player in the end doesn't matter, right? And I think the same perspective is what is reflected in the very open APIs that you have for most cam systems that allow us in the end to integrate. And I think they opened the door knowing exactly that in the end it only matters what comes out on the customer side. And then you're always going to figure out a way to either coexist and share technology or be bought up or I don't know, there is a different way stand to go about it. But I wouldn't see it as as like kind of a fight between different players in the market because in the end, all try to satisfy the same goals very often are complementary to each other and therefore can very well co-exist in this huge space that you're anyways operating in.

Michael Finocchiaro

⁓ Interesting.

Erdem Ozturk

Yeah, I think the key thing is the value add, like what value you are adding. Each company must be adding value. If that's the case, of course, that maybe overlaps, but as long as there's unique value add for each company, every company can coexist. And also, as we mentioned, there's a lot of steps, like we talked from CAD to the measurements. And the big players, of course, want to cover all of it, but they also are aware that they cannot realistically do it just themselves. Like, I remember having a conversation with one of these multi-billion companies, like ⁓ senior person, senior manager. And I was saying, we are only small, we are 15 people. We cannot do all of it. We need to do it step by step. And then he said, OK, we are thousands of people, but we are in the same boat. They also have limited resources. Nobody has all the resources they want. Everyone has their limitations. And even if you have all the resources, you may not have the experts. One nice comment was, guys, we have seven PhDs in your company on the machining dynamics area. I don't know if there are lot of companies who have. like 50 % ratio of their PhDs in the company. yeah, we basically are a deep tech company solving a hard problem called chatter vibrations, which is a hundred years old problem. And I don't think there is a lot of companies that can solve it. And that's our unique value at. I'm sure Monaco also have their own unique value at. So every company who has that will be successful in the market because there's a lot of things to do in the sector.

Michael Finocchiaro

You

Pascal Weber | Manukai

I mean, there's an easy saying around it, right? You're not really competing against the corporate during competing against the product manager with a team inside that corporate, right? That's, then I think that I think summarizes it quite well.

Michael Finocchiaro

Well, I'm just trying to think about how it must be still challenging because a lot of companies are going to feel like, well, if I go with a startup, it's risky, they could fail, and then I'm going to be left in the dark and have to go back to Siemens or PTC or DS at the end of the day. Anyway, so I was just trying to understand the dynamics there.

Erdem Ozturk

Maybe one thing that's worth mentioning, I think in manufacturing, at least in my experience, competitors are not the software companies. Competitors are the established methodologies or the methods of the machine shop. For example, if there's a problem in the machine, I see a lot of people doing this. I don't know what you mean. You understand what I mean. There are knobs, control buttons on the machines that control feed rate or spindle speed. If there's a problem, they reduce the feed rate or spindle speed.

Michael Finocchiaro

The people.

Erdem Ozturk

There is a range, like they can reduce by 50%, increase it by 25%. Usually they take it down. And then if they are lucky, they solve it. They solve the problem, but usually they are not lucky. They cannot solve it. And when they solve it, they lose productivity. But what we have been demonstrating is we usually go faster. And then when we go faster, and when I say faster, sometimes it's 110 % faster. So there is no way if you did it all similar speed, all right, we'll take it 110 % faster. At max, it will give maximum 20%. So yeah, I think our competition is basically this process.

Michael Finocchiaro

Okay, interesting. There's more, yeah. Entropy is the enemy rather than anything else. ⁓ Before we wrap up, I'd like to ask about, the, so I did a post today about some of the answers that I've gotten to this question. I just wanted to contextualize it in thinking that there's a lot of, ⁓

Erdem Ozturk

There is no traceability, it's called chaotic. ⁓

Michael Finocchiaro

The demographics on people listening to my podcast and reading my posts, about 21 % are entry level because I write a lot of educational things. And I think that there's, they probably have a lot of anxiety because of AI, because of all the rumors that AI is taking all the jobs away. But every time I talk to you guys, I get the impression that that's not the case at all, that there's always going to be far more work for real engineers. So I just wanted to get your comment on that. For instance, let's put it this way. Two or three guys come out of a university or maybe three girls and a guy, and they're looking for jobs. Why should they go work for productive machines at Manukai instead of just lining up like everybody else to work at Metta, Siemens, Dassault, AWS? mean, what are the interesting challenges that are so awesome that people are dying to get to work because they just want to work on it rather than... a boring corporate job and what kinds of things do you think it's even like doing that? Is that even a better ⁓ insurance policy against being AI'd out of a job? Because the challenges are always so much more interesting. I don't know. It's just spitballing here. Go ahead. ⁓

Pascal Weber | Manukai

I mean, from my perspective, there's two kinds of people. is people that enjoy uncertainty, that enjoy creating something out of nothing, that enjoy being maybe not guided so much, be like, I don't know, fans so much, not being told so much what to do, like be in the end their own kind of boss. And then they like really figuring out and living with this uncertainty. And there is other people that like to be

Michael Finocchiaro

Told what to do.

Pascal Weber | Manukai

They like to have a number up on the wall telling them, okay, this is your goal for today. Make the number go up and down. And yeah, I would say these are the people that like to go in to work with the corporate. And I think the former ones are the ones that like to go to work with a startup where they can really create something new.

Michael Finocchiaro

You And you think that for Manu Kai, there will always be humans doing 95 % of the work, right? Because you've even said yourself, you think AI has a limited role overall, that you still need all the engineering skills.

Pascal Weber | Manukai

I mean, I have a very philosophical take on that, right? I believe us humans, are always meant or we always have to create something. So even if I automate everything for you, I'm pretty sure tomorrow you're going to find some new thing that is not yet automated that we can do with it. So that this is just human nature. It is us as humans that are always kind of in this status of, in this state of trying to figure out stuff and trying to change stuff. And this nature, I think, will kind of not go away. And therefore, there will always be new things, new challenges to solve, new problems appearing. And we're never going to end up in the state where we are just, I don't know, in a matrix, kind of in this capsule, just generating energy to the machines that actually operate the world.

Michael Finocchiaro

Ha ha ha.

Pascal Weber | Manukai

I hope so at least.

Michael Finocchiaro

So how do you want to tackle that one Adam?

Erdem Ozturk

Yeah, for us it's more about impact, like what is your contribution to the world? What is your impact? What is the impact of your work to the wider economy, wider society? So we are working on a problem, as I mentioned, it's like hundred years old problem. There is a book called Art of Cutting Metals from 1907 and it's written by the scientific management group of the time called Frederick Taylor. So Taylor's two life equations, for example, or... like he's well known in the management community as well. And he basically said that chatter vibrations cannot be solved. There is no formula or method that can help a machinist to solve chatter vibrations problem. Hence, the operator has to use trial and error. So that's a statement. And he was right for maybe six or 70 years, because it took academic community to formulate the problem and generate some models. And then, The first model started in 60s, 70s. There's a lot of models, but there is no company that can solve the problem in production like we do. So we are basically enabling companies to solve the problems that they cannot solve themselves in a very short time frame. And that's really powerful. They see the implications. So people working in the company see the implications of it. We have a very recent example with a large aerospace company in the UK. They didn't come to us for. solving productivity challenge or reducing the cost. They came to us because there was a lot of noise in the workshop environment. It's really disturbing. People don't like to work in a noisy environment where the patient is screaming and they have to wear headset ear protectors because if it's more than 80 decibel, it's not safe. So they have to wear ear protectors. Basically, the requirement was, can you make these machines more quiet?

Michael Finocchiaro

Ha ha.

Erdem Ozturk

Please don't sacrifice your productivity because we want to keep the Our guys did a visit, they measured the machines, we came back, we ran the optimizations. I think depending on the machine availability, think we might have gone next week or the following week. And then our guys stood on the next machine. We have also software that measures the noise coming from the machines and it can tell us whether the process is chattering or not. And then we ran the default. Original process is chattering, it's very noisy. And then we run the optimized process. It's stable, very quiet, and 20 25 % faster. So when an engineer sees that, of course, it gives a lot of satisfaction because, yeah, something like you can see your work results very quickly. So this, think, motivates. Also, the carry growth opportunities. So we had recruited when we had the seed round two years ago. not only three years ago, we recruited like 10 people straight away. And then those people work on many different fields. Like if they had come to a corporate, they wouldn't have the skill set that they achieved in two, three years time with productive machines. So they are becoming stronger individuals in terms of the capability, skills, skills.

Michael Finocchiaro

Well, thanks, that was a great answer, I love that. ⁓ We're about to wrap up. I just wanted to ⁓ ask you guys, ⁓ other than perhaps threaded in Warwick on the 25th of March, where can we meet you guys and have a beer ⁓ into the winter, beginning of the spring? Are you guys gonna be at any other trade shows?

Erdem Ozturk

Okay, I will be in CRP winter meeting next week, when is Friday, Paris. we will be in Paris. That Thursday, we can meet there. We are a very central location, close to Notre Dame Cathedral. And then afterwards, we have the big trade show in the UK called the MAK exhibition. So I said, it's the biggest mission to fair in the UK.

Michael Finocchiaro

Nice. Where's that? In Paris. I'll be in a Proven, unless it's Thursday, then we can have a beer. Okay, nice.

Erdem Ozturk

Of course, it's much smaller than MO Hannaur, but it's good. And we will have a ⁓ machine in our stand. So we are called machine tool company. There will be a small machine in our stands doing some real time demonstrations. So I'm sure.

Michael Finocchiaro

Right. ⁓ ⁓ should bring the machine to threaded. That could be fun. Yeah. ⁓

Erdem Ozturk

Timing will be difficult. The mission is to us on 21st of March, so we will do some preparations and then Mark is in 20th of April. So, timing wise, we want to make some time for the traffic.

Michael Finocchiaro

Okay. ⁓ That's okay. Well, I hope I see that. How about you, Pesco?

Erdem Ozturk

But we'll be in MoRiG, definitely. So we'll bring the videos of the demonstrations that we have already done to MoRiG.

Michael Finocchiaro

That'll work it. Have a good best, girl.

Pascal Weber | Manukai

So we are currently mostly focusing here in the Dock region, smaller fairs, more local events, but definitely going to meet us at the Hannover Fair this year. yeah, looking forward. Otherwise, just reach out. Our office is always open. Beer is cold. Just drop by.

Michael Finocchiaro

We were thinking of doing a... All right, I'm going to I'll definitely be going to Zurich to do that for sure. ⁓ And we're probably going to do a threaded Germany later in the year, probably in the fall. So probably in Munich, so not very far away from from Zurich. Once again, thank you so much to Erdem and to Pascal. Had a great discussion today. I I certainly appreciate a lot more about machine that I didn't know before, and I appreciate it. Thanks for taking your time and joining me. ⁓

Pascal Weber | Manukai

they're Michael Finocchiaro (1:00:04) Thanks to the audience, we were up to about 12, 15 also. So just like the last three, so we're on a good curve. And of course, this is a long tail kind of thing. be lots and lots of impressions. please, if you have any other questions, put them in the comments. Erdem and Pascal will come in and answer your questions. And of course, feel free to reach out to them if you want any more information. And of course, if you're a student and you want a job, they're always hiring, right? I mean, that's what startups do. Pascal Weber | Manukai (1:00:31) Exactly, yes. Michael Finocchiaro (1:00:35) So have a fantastic weekend. Thanks everybody. And we'll see you on the, well, I'm sorry. Well, one thing I'm going to be, I'll have my own, ⁓ I'll have a boot, a podcasting booth at prove it on Tuesday, next Tuesday from ⁓ about 1pm Texas time to 4pm Texas time. So basically European late afternoon and evening. And I'm to have a really cool manufacturing startups coming through their fuse and high bite and Pascal Weber | Manukai (1:00:38) Enjoy. Michael Finocchiaro (1:01:01) Avalara and stuff. stay tuned. I'll put a link up. There'll be a live broadcast from there. And in any case, have a great weekend and we'll talk to you soon. Erdem Ozturk (1:01:11) Thank you, Michael. Bye-bye. Pascal Weber | Manukai (1:01:12) See you around, bye bye, thanks.