Michael Bogomolny, CEO of InfinitFORM

We sit down with Michael Bogomolny, Ph.D. of InfinitFORM, which has blasted through the hype of topology optimization with a deterministic, GPU-accelerated engine that creates parts that are both optimized and manufacturable. The result: shorter design cycles because prismatic parts are ready for machining. 

In this podcast, Michael talks about:

• Funding update and market momentum
• Why mesh-based generative design fails machining
• Manufacturable, parametric outputs with feature trees
• AI as assistant for setup, critique and reports
• GPU solvers for fast, deterministic results
• Cloud and on‑prem options for regulated teams
• SolidWorks add‑in and CAD‑native export
• Roadmap for injection molding and die casting
• Beta learnings and January release timing
• Trust, IP strategy and integration into workflows

Transcript

Welcome And Guest Introduction

Roopinder 0:20

Hello, and welcome to FoDES, the Future of Design and Engineering Software Podcast. My name is Roopinder Tara. On the show, we will have guests that will discuss tools and technology that engineers will find interesting and useful. I am pleased today to have on our podcast Dr. Michael Bogomolny, founder and CEO of InfiniteFORM, Los Angeles area-based startup that may well produce optimized and manufacturable shapes. With a PhD from the Technion, Israel Institute of Technology and a rich background in computational design and topology optimization, Michael is a pioneer at the intersection of engineering software, generative design, and manufacturing automation. Michael was co-founder of Paramatters and as CTO led research into multimaterial generative design frameworks and additive manufacturing workflows. He went on to found InfiniteFarm, which earlier this year raised $12.7 million in funding. Michael, welcome to the show. Do you mind if I call you Michael or do you prefer to go back? Michael's okay? I could call you Doc.

Michael 1:26

Thank you for the great coverage. When we spoke a year ago, I just introduced the concept in general, but we've gone pretty long way into improving the product and actually putting things together so it's working.

Roopinder 1:38

It'll be good to catch up. I think when we talked last time, you had not even released the product, correct?

Michael 1:43

We plan to have official release in January next year, but we have already customers who have been using the product and purchase subscription and using it on daily basis. It's a very delicate situation where you feel like it's a time to release because we talk to engineers, and engineers are difficult audience, they the demanding audience. When is the right time to release the product? From one point of view, I believe like many startups run too fast. And at the same time, things move so quick nowadays, you have to you cannot deliver developer ever at the same time. You know, one product is not exactly things are just move.

Funding, Momentum, And Industry Context

Roopinder 2:19

I find this to be true too from where I sit. I see so many advances being made, so many in our field, so many in our industry. There's no better time to strike than now. Did all this excitement about AI help with your fundraising?

Michael 2:32

Yes, in April this year, we raised 12 million dollars from our partners, Yamaha Motors counterpart and schematic ventures, also our first investor. The story of what we are doing resonates very well. We have tremendously positive feedback from beta customers who really believe that what we're doing is super valuable. And still a lot of work in front of us, but I feel like we are on the right track. And the product we develop, it's exactly closes the gap in the existing workflows. And I'm happy to talk more about what you're doing.

Roopinder 3:04

I did notice that you had a big name in the CAD business, also become an investor in Infiniform, correct?

Michael 3:10

Carl Bass, he joined the board of directors and he also invested in the company. And his asset which brings so many smarts and insights. He's one of these guys who understand technology and business, and also a pleasure to work with Carl.

Roopinder 3:26

Are you still located in Southern California?

Michael 3:28

Southern California, yes. We have a distributed team, but we we all soon California.

The Workflow Gap In Design To Manufacturing

Roopinder 3:33

Are you permanently moved to Southern California? Yeah, I've been here for 11-12 years. We are now at ENGtechnica, and I want to introduce the InfinitFORM to this audience. So you can tell why we should be paying attention to InfinitFORM.

Michael 3:47

So the current workflows built around design of structural components are very silo, it's labor-intensive. When we have to create parts, structural components for defense automotive aerospace industries, these should be parts which withstand different loading conditions. Same time, these parts should be manufacturable, and we have to create geometry in native CAD format, whatever CAD system you you're using. And actually creating parametric shape at the same time which will stand for physical loading conditions and also manufacturable, these are three different disciplines. This is how workflow is built today. And there is a complete disconnect between the these disciplines in the existing workflow, and as a result, there are a lot of iterations and human labor in the existing workflow. How to get generate those designs. And infinite form, we believe, and we strongly believe is closing this gap.

Why Traditional Generative Design Fails Machining

Roopinder 4:47

These the generative design and these computational design systems are based on a simulation, so they do a fairly good job of handling the loads. But I think their weakness is and always has been that they can't be manufactured except with a 3D printer normally. They just are weird organic shapes, clumpy, lumpy, uh twiggy. They're just not things that can be made straight or uniform. But Infinite Form does take care of that problem, correct? It makes it machinable. Correct.

Parametric, Manufacturable Outputs Explained

Michael 5:17

Okay. Not only machinable, also the geometry represented in traditional CAD representation, parametric CAD representation. The reason lays that many of generative design solutions based on topology optimization made them. Now, topology optimization is the method, by the way, I had the privilege to do postdoc in the group of pioneers in this field, as well as I spent a few years as a developer and alter also in this field as well. I know this domain very well. The reason is that classical topology optimization, the way algorithms work is that you discretize geometry, you optimize, and from discretized solution, the only way to extract geometry is using these computational geometry algorithms that usually provide organic shapes. And the representation of these organic shapes are usually meshes, which later on people put nerves on top of them. And so they really not really edit and also not manufacture because the way manufacturability of the machining of these parts works, you have to take tool and you have to start machining, removing material.

Roopinder 6:21

People try to take what's generated by generative design and then create on top of it, almost like you would trace a design on paper, but take 3D and overlay it over parts, essentially making the whole part over again, but just using that as an inspiration. So much rework. I found that to be ridiculous.

Michael 6:40

Yeah, that's exactly what we hear from many customers. They say we run optimization, it takes us little time until it converges, and then we have to take the inspiration of it and start redrawing. And the fact that redrawing can take a lot of time, and many customers complained it will take us less time if we'll just start from scratch using the usual workflow. That's why many of these generative designs fell short in the traditional manufacturing. And that was the reason we found at Infinite Form to close this gap. We developed completely new groundbreaking technology which optimizes results, which are manufacturable and also represented in parametric cat.

Roopinder 7:17

I'll be in line to buy this as soon as it comes out because this has been a long frustration for me. Generative design was being pushed as a do-all-end-all for engineers. Oftentimes, like going over the top, like saying this is a better design than you could have designed yourself just because a computer did it or we're smarter than you. But engineers may not have the optimum design, but we are able to converge the optimum with the manufacturable. So that's what if you can do that, I think it's gonna be a gold mine.

Michael 7:47

If you look into the workflow, first design task is going to CAD engineer. CAD engineer creates shapes based on his intuition. He doesn't really know how physics would like material to be distributed. He's creating shape based on his intuition or previous experience without really getting into reality.

Roopinder 8:05

I would add to that that he is also used a parts library, standard shapes, for example.

AI As Assistant, Deterministic Engine As Core

Michael 8:09

Yes. And then whatever he creates is going to a different department of simulation engineer. We have to check if created shape viable to structurally sound. It can be over-engineered, can be not structurally sound, so you have to redesign the part to make it stronger or change material. And usually there's a lot of back and forth between design engineer to simulation engineer, simply because design engineer creating shape without physics in mind. And you also hope that this design engineer was experienced enough to apply manufacturing constraints. Because very often design and simulation engineer can come out very good design, but then it's not manufacturable. And then after manufacturing fed feedback, it goes back to design when you have to change it and run through simulation again.

Roopinder 8:50

So what we developed in InfinitFORM that I'm happy to demonstrate is that what you're describing is a lot of hopes going into that one part that hopes it'll be survived, survive its environment, that'll be manufacturable, that'll be maybe even, I don't know, presentable or aesthetically pleasing or conventional rather than radical. There's a lot of hopes there.

Michael 9:11

And plus AI, which is coming into the game. And in form, we have our philosophy around using AI into our workflow. And I'm also happy to present what we are doing. From one point of view, there is an engine who is a deterministic engine who can do the right geometry and repeat it over and over again. Or you have AI agent which or you can ask it to generate shapes for you, but it's difficult to understand how it came out with this shape.

Roopinder 9:41

So am I to assume there's two different engines in here that can be used, or are you favoring AI lately?

Cloud vs On‑Prem And GPU Speedups

Michael 9:47

No, we use AI to assist users in setting up problem correctly. If something is not clear, AI agent can help engineers to realize what the best way to set up problem correctly. And we also use AI once engine generated design to help to analyze the result and recommend ways to improve the existing design and to further adjust some parameters, rerun, and get even better design fully automatically generated by our software. Plus, AI can also fully automatically generate report. And yeah, that's a way we see AI as an assistant to the engineer. The deterministic engine which is solving the problem is working on GPU, also working very fast.

Roopinder 10:31

It's working on GPU. So that makes me wonder is it going to be local, on-prem, or is it going to be all cloud-based, or do I have that choice there?

Michael 10:40

We have with currently or we have software is available in two versions: cloud version, and we have customers from defense and of space industry due to regulations, they must be local. So we have also local version of the software, which they can run locally on the hardware. The cloud version we run on AWS and it utilizes GPU heavily, especially on the solver side and on geometry side. And we see tremendously tremendous speed up of solutions. Just to give an idea, something that two five, ten years ago, probably 10x more time today running way, way faster. Just one million elements, which approximately 3 million degrees of freedom runs on our solvers on GPU less than seconds. So it's tremendous speed up.

Roopinder 11:26

It's amazing, isn't it? If your company's named InfitiFORM, you're using infinite computing on the AWS. There's so much available there. It is practically unlimited, right? And we generate infinite designs as well. Yeah, I see. Very good. Oh, you certainly set the stage. I'm dying to see in operation.

Live Demo: Setup With AI Copilot

Michael 11:41

Yeah. So I can share my screen. Do you see the dashboard? I do. Cool. The way it's it currently the process works that in order to start new projects, usually users import an assembly as a step file. And I imported a holder bracket here. Usually you import a step file and we would like to optimize it. The workflow is fairly simple. There are multiple steps. You find material properties, classify geometry, apply different loading conditions, design goals, and manufacturing constraints. As I mentioned, this is so far very similar to a FEA program. Maybe without manufacturing constraints. Yeah. But in terms of design goals, if this is FEA software which optimizes, yes, but if it's just the analysis, you defined material, you've mesh the part and define load cases.

Roopinder 12:26

Okay.

Michael 12:26

We can do it in two ways. I can show how you can set it up manually, or we can use AI companions. So this is AI copilot.

Roopinder 12:35

Oh, good. So you have a copilot, fully functioning copilot that lets you enter natural language commands.

Michael 12:41

Yes, let's try. Let's try let's try it. Let's even start with a copilot. So, for example, let's I go. Let's see what the companion will tell us. Probably he's familiar with our workflow. Okay. And he recommends to use stainless steel as a material. And in general, he can set up fully automatically the stainless steel material properties. And he asks, would you like to continue moving in the workflow? Yes, we can just say yes. It should analyze the geometry and find the the because he's got all the information about the geometry we imported. So he recognized that the largest volume is a design space and other volumes will be preserved.

Roopinder 13:22

Okay. Michael, you said it is knowledgeable of your of your process or your history. So are you using like ChatPT 5.0?

Michael 13:30

Which Oh, we use the other engine, okay, but we are expanding and training it more and more into our workflow. We don't need such a big okay.

Roopinder 13:38

So it remembers what you might have asked previously. Or does it or doesn't it remember what you asked before?

Michael 13:44

Right now we are working on training it such that it's sensitive because some users don't very sensitive of their data. We don't want users to feel like we use their data. So currently we don't use user data. We train on our workflow and specifically our applications.

Roopinder 14:01

So the workflow it remembers is workflow that's supplied to it from your standard usage.

Michael 14:06

Correct.

Roopinder 14:07

Okay, got it.

Michael 14:08

Because we are sensitive. Users don't want their understanding of problem to be used.

Roopinder 14:12

I would wish for it to remember what I'm doing so that I don't have to. So it's more functions like an assistant to me that it hey remembers, hey Repinder, you used a channel here before, used a beam, right? Or use this kind of load, or you I would like it to remember that because ChatGPT, which I mentioned, has gotten into the habit of knowing what kind what I've asked before and even suggests to me what I do next.

Load Cases, Goals, And Milling Constraints

Michael 14:35

I fully agree with you, and this is where we're heading. We just want to be very sensitive to the data. Okay, sorry. Excuse my impatience. That's great. All right, okay. And so assistant, uh, infinite form assistant is asking, would you like to proceed setting up lot cases? Let's say yes, please find three lot cases. By analyzing geometry, he has to go to the next step and fully automatically would define lot cases. Of course, we can define load cases manually, and I can show you how it's done manually, but you see, I'm not touching anything.

Roopinder 15:04

You haven't gone into the command line or whatever command editor you have yet. You're working strictly with the copilot.

Michael 15:10

Yeah, okay. And uh yeah, so it defines three load cases. We can review them. So here are the summary. Let's see if Michael, how does it know where to put your loads? That's why we trained it fixed boundary conditions here and then applied load cases. So you see it's a normal load case. We support static vibration load cases, work on thermomechanics, we support forces pressure acceleration. So it did pretty good job here and apply, let's see, acceleration load case. Okay, and let's see, it's a different direction, and then he defined what's extreme load case. Also, I see the values of forces are larger. We are working on now just to understand when users have requirements and text somewhere, they can import, read, understand, and set it up. It would be much easier right now. It's trained on general data. Okay, so we defined uh load cases, and then I next step I am harmonizing for stiffness because it had to set up the design goal. So, what's the with setting up minimum compliance meaning maximizing stiffness? And he he asked okay, he put more, I see if he put more weighting factor on extreme load case, so part will not kind of fail higher priority and put final mask target on three kilograms. Would you prefer traditional milling or additive manufacturing part? Let's say milling. So it has to move to the next step where we define the the material cases. We defined milling. Honestly, I think maybe too many directions, design might be too complex. But it's giving you the option, you can check the course.

Roopinder 16:42

I can click it like this and park. I'm loving this. This is awesome because I don't have to know any specific commands so far. I don't have to learn your language, I can use my language. I said that's a goal.

One Optimal Result And AI Review

Michael 16:53

And if you do something incorrect, we are working, so it will highlight hey, thing is really wrong here. And that's it from here. You know, from here, would you like to start optimization? I can say yes, and it will start optimization. We are using AI in multiple steps. First, setting a problem. If I go back to the cabinet, this is a problem with setup, it's currently in the queue because it depends on the availability of instances on AWS, probably take a few minutes. It opens so, but I have like part which already was designed, and this is actually can be interesting to discuss. So, this is a result of similar similar bracket. Now, what is important, you see that all the geometry, it's a fully parametric geometry construct of how engineers would design.

Roopinder 17:34

When you say parametric, is it parametric to like a CAD program? I will I will explain in in a second.

Michael 17:41

I will explain, but you see, it's it's a machinable part if you would machine in this direction. Yeah, I see that. I see that you would be you would be able to machine first. Second, it's it's prismatic CAD model, which which was generated by as engineers warp, they create sketches, the extrusion operation.

Roopinder 17:57

This is totally yes, and it accommodates the diameter, it doesn't go any smaller than the end mill diameter.

Michael 18:02

Users will define the exact end mill diameter, and of course, the it's not just a shape, it's it's of course backed by by the by the finite element analysis. We see the stresses and deformation of this part, and we have the full availability, a full understanding how part is going to perform under different loading conditions. So, what is also interesting, let's say we we obtained, let me turn it off. So we obtained the design, and you can ask AI companion, just analyze the result. And uh, it was probably a little bit different example, maybe applied to high loads.

Roopinder 18:33

I you didn't show this, but is this what it considers to be an optimum design? Is it picked out of several possibilities all on its own? At the moment, it's generating only one result. So it makes its own assumptions and comes up with a next attempt at optimization rather than optimization, because it's not going through the usual thousand possibilities.

Automatic Reports And Stress Feedback

Michael 18:52

We believe that once you define problem in a certain way, there is one solution to it. If you would like to explore multiple options, you may be able to do multiple options.

Roopinder 19:02

But once we are fairly confident I couldn't do any better than this, even if I let it run. Yes, I'm very confident.

Michael 19:08

So I ran this optimization and it says, hey, critical issue, maximum stress is 100, 463. It exceeds material yield stress, safety factors, so part will fail. Who would you explore addressing?

Roopinder 19:20

You know, what would be nice to see is if it's a side-by-side comparison,

Michael 19:26

So you see, I asked what you would recommend. It recommends to increase to add more material because stresses are too high. And yeah, because currently it was set up to have 2.5 kilograms, 15% volume fraction. He says increase material by 10 more percent. Okay. We get choose approach, yeah.

Roopinder 19:43

We can increase. And all the output is natural language too. It's like I can understand it, it's not a table of numbers.

Michael 19:50

Increase material fraction, it should update the more material and rerun, and we can get the new design. So I see it added, it was lower material, so 25. percent volume fraction. So and we can we can further run an optimize result. But what is even nice if I would go back to our previous design when he analyzes we can ask to generate report. If engineer wants to go to his manager quickly and not to spend time on reports, we can generate report and I will I'll show how report looks like. Here we go. This is our report. This is a result it's very well structured and have summary you see the design exceeded we need to re-optimize because under certain cases the maximum stress deformation, key findings, recommendations.

Exporting Native CAD With Feature Trees

Roopinder 20:35

No saves the pay happy to do that paperwork. It's the last thing I wanted to do as an engineer to generate the report. It's already there. Now that being said though, I know it's a PDF, which normally people don't edit. There's not an option to put it out in a word format is there?

Michael 20:50

I think it should be possible if you want to edit it. Yeah. Maybe one last thing you ask me is it a parametric CAD? So the the answer is inside our engine we generate design and design based on new class of optimization we developed which allows to create sketches and extrusions. However if we want to export the result let's say for in SolidWorks native format at one of the bottlenecks in CAD industry everyone have the proprietary format. Of course we currently I just want to show quickly we developing an add-on for SolidWorks and we plan to do it for all the other CAD lenders. So by the way we were we were selected by the SOS system as a partner so we were able to develop a plug in for SolidWorks. So for instance if we design this part with the infinite form it's clear it was a sketch and extrusion etc so we we are able now to export sketches from infinite form and this plugin will read our sketches and generate them in native SolidWorks format.

Roopinder 21:50

Very nice now does it that's great that it has that interface to SolidWorks and I see that addressing some of the other CAD systems as well SolidWorks of course is the most popular one.

Michael 22:00

Does it also preserve the history or the design intent so yes if you look on the left hand side here so you will see all the sketches that we create on our side and export so that tree comes across left hand side and sketch and extrusion. So it does exist it does exist.

Roopinder 22:19

Okay very good this is so this is not that usual dumb geometry step by step file exchange or it has quite a bit of its original intelligence.

Fitting Into Existing CAD Workflows

Shorter Cycles And Practical Adoption

Michael 22:27

So we know if we zoom out a little bit in terms of the workflows inside companies all the work is done in CAD and it doesn't matter if it's an X or it's solid work or PTC, Creo or Autodesk Inventor Fusion, all work is done in CAD. We want in the design engineers in early design stage to communicate to ask all the requests for the future design from Infinite form. Infinite form export back the geometry with with the features tree back into the CAD system with native format. So now users can easily modify some sketches and edit this geometry and it have all the features tree. So to minimize the any any friction point. Finally we export manufacturable geometry which is optimal extending different loading conditions. It can be dynamic static lot cases we also work on mechanics and export native CAD geometry. So the idea is to exist in parallel with the CAD systems not to replace them for geometry creation right exactly we are not going to replace we're not compete with any CAD system not with simulation system we really identify that there is we are a layer of intelligence sitting inside the existing workflows which fully automatically generate optimal design. So imagine and you show you so in the software that I showed the time the design times run between a simulation runs one minute 30 minutes 10 minutes we have designs which run five minutes depends how many load cases you can obtain designs in in in really short period of time in native formats. You can imagine how much it shortens the design cycles within companies that within 10 15 20 minutes you can get fully parametric cat back which optimized per specific manufacturing method and per performance.

Roopinder 24:09

You've taken something they weren't particularly good at and offloaded it and then found a way to merge back into their operation for everything else they have to do which they're probably good at no sense in you reinventing the wheel for drafting or something let them do that part and you've got the optimized shape.

Michael 24:25

Very nice it worked really well no glitches no blue screen of death could you see how incident we are in our product that it's working nicely I see.

Roopinder 24:35

What's holding it up now why do I have to wait for the new year before this is available what are you using trying it out with different shapes are you more robust?

Beta Feedback And January Launch Plan

Michael 24:43

Can I stop sharing I will see your face of course okay great so again as I mentioned we talk to engineers engineers have a lot of requests and demands and it's not an easy audience. So we we have been running beta program and pilot with many big names I want to mention the names but many leading defense aerospace automotive industry and the best product you can develop it with customers feedback. This is from my experience and we have very carefully documented all the feedback and requests from beta customers the more we take that feedback into account the better product becomes and we feel like you know devil in details always in general it looks great but some small things even inks some small things or any because you see it's I I did it with text but when if I would go into GUI and click different buttons some small glitches overall we want users to have the best experience we decided to postpone the release to January because we want to get the best product we can in customers' hands and also in terms of pricing I think are very affordable and for shortening design cycles which take weeks in the news into we believe the return of investment for customers is significant. And so we decided that to offer this product to customers we would need to fine-tune it and remove all these small inefficiencies to make it available we decided to release in January so stay tuned.

Roopinder 26:11

Oh yeah definitely will it's a wise approach to wait and get it right because I think what you're up against is almost a backlash against AI. Most engineers feel like oh I've used AI and what they've used is chat GPT or something like that and got the wrong answer. So engineering AI or industrial like what your company's doing has to overcome that belief AI can be useful. You have your impressions based on LLMs but this is actually working out well. But on the other hand Michael do you sense any pressure from these big giant CAD companies coming up with this on their own?

AI Skepticism And Determinism

Michael 26:47

It's a great question. First of all I fully agree with all the previous sentences you described I fully agree with you because also many engineers say we don't know AI. We ask the same question from AI it gives two different answers. So how contrast it because our engine if you set up problem same manner you would get exactly the same result. So it's fully deterministic should be there to help you if something is not clear or to guide you. Now about your question I don't think we really compete directly with Ken Sin's companies we really complement the existing workflow.

Roopinder 27:20

Now you ask you have quite a bit of head start into a lot of intellectual property that maybe they haven't caught on to yet they haven't developed yet or they're not inclined to do this yet.

Team, IP Strategy, And Head Start

Michael 27:30

I'm lucky I have a very strong development team around me and I'm I'm I'm really fortunate to have a great team in startup with 10 12 people you can do more than big companies with hundreds if you know what you're doing and you're very focused and you have the it's a lot the quantity is more quality of people and I do believe that we have a very strong team with expertise in in the specific field of topology optimization, computational geometry and manufacturing and when these disciplines merge together it because this product is multidisciplinary it's not CAD it's not simulation it's not optimization it's not manufacturing it's kind of combining them all together. We need to have expertise in all fields what we developed doesn't exist even in the papers. In our team we have six guys with experts with PhD from a bleeding research group in Denmark by the way technical university of Denmark they have done tremendous job in in in developing this field of topology optimization led by Oli Zygmunt and and Martin Benson deeply grateful so they do tremendous job and what I wanted to mention is that we feel like to achieve what we have achieved so far we had to develop fundamentally different procedure. We of course we're not going to publish it because it's fundamentally different to allow to do it.

Roadmap: Injection Molding And Beyond

Roopinder 28:48

I have to ask is it better to publish it you chose not to publish it but it wouldn't it be better to get a patent software and algorithms world patent is a little bit tricky because then you have to disclose the algorithm itself it basically shows the world what exactly you're doing we want to keep it more like a know how know how than disclosing yeah to keep your advantage very good you're definitely ahead of the pack the pack that I see and I mean I'm seeing a lot is it's really I feel heating up the space a even a year ago when we spoke it was like why isn't there more AI for engineers nowadays I just heard somebody else tell me not a week goes by where it doesn't have another app that's using it. So I think it's an exciting time to be in this business definitely for me.

Michael 29:35

And I can mention that we I'm showing solutions for machining we have we rapidly developing solutions for injection molding which is a huge market opportunity. Maybe next time we can have another one I'll show you how injection molding works it takes all the manufacturing requirements into account like drift angles and rip generation plus we also work on die casting and the the more we do the more requests come. Then of course requests for sheet metal and other technologies come but we have to do step by step.

Roopinder 30:03

All the manufacturing processes one at a time start with machining that's the main one you got to get that down before you get to the harder ones that's a good approach and definitely aligns with what engineers want which is a far cry I find from software companies assuming what we want. They assume what we want and this is one of the problems engineers have with AI. Have you ever heard this expression it doesn't apply to AI but I'll use it for AI. The AI I'm getting is not the AI I want and the AI I want I'm not getting like it doesn't apply to you thankfully I mean I'm really appreciative of what you're doing. I think it is what engineers want. So keep up the good work. Thanks for bringing us up to speed on where you're heading I will take it up on your offer to catch up with you at a later date. I think a year is like dog years now, but not longer than a year. We should have you back on you want to show me what you're improved on.

Closing Thoughts And Next Steps

Michael 30:56

I think in terms of injection molding give us three months I can show three months we have it's working already it's just GUI and some parameters around maybe even less.

Roopinder 31:05

Great to see you again great to talk to you and great to hear about the work you're doing. Best of luck honestly this is a welcome thing for all engineers if this is something they're gonna like thank you very much we're also doing tremendous work in covering all this space oh I try kind of you to say all right Michael I'll let you go on to your work great seeing you and I hope to see you soon. Thanks a lot by the way