Alex, founder of Fainite, discusses how physics-constrained AI models can accelerate hardware engineering by reducing design and manufacturing cycles from six to eight months to weeks. The company builds a reusable world model for physical product development that combines AI with physics constraints, enabling faster simulation, design, and manufacturing across teams and industries.
▸ Read the transcript
I think it took me a bit to gain the confidence, for example, to start a company. If I was younger, I would have told myself, start traveling earlier, like look for opportunities abroad if you have the chance and back then I was analyzing the failure of plastic materials for outer space applications that's how I fell in love with the topic So there are other players pursuing similar technologies and based on this experience I realized that there was a bottleneck in the way in which the design or even manufacturing of these products was done. We are focusing on simplifying the building of physical products the uniqueness of our solution is that we are building something that is both fast and reusable there are always new challenges and you can experiment new ways to solve the challenges. He message would be don't constrain yourself try to dream big and follow on that like go pursue what you want Hello guys Here we are with Alex from Fainite Nice to meet you, Alex Pleasure So, Alex, you are building right now a Fainite AI, right? Yeah. and it is related to physical AI, which is going to be on trend on point will you tell us first what does Fainite AI do? Yeah. So we are focusing on simplifying the building of physical products and physical products and the aim of the company is to remove the biggest bottleneck in building products, the latency between coming up with a design idea or manufacturing idea so deciding how you're going to design or manufacture some parts or components, and then waiting to see the results of the testing campaigns and or the simulation campaigns and understanding whether or not that decision that you have made in a preliminary step makes sense, right? And this is a step that normally could take up to six to eight months, right? So it's very lengthy and you need to reiterate around these processes. So basically, in practice, what we are doing is building, if you want, a world model for hardware engineering. It's a model that understands how you can do design, engineering and manufacturing of physical products and has been trained on different physics, from structural mechanics to thermal simulation, electromagnetism and yeah, it's a much faster way to do design, to do manufacturing and the layer that we are building, it's built in such a way that all the players that are taking part in the product, hardware product development, can use the same platform, right? So it's also a way to reduce the back and forth across different things, from the design team to the manufacturing team. You told me previously that you were in Caltech doing some research, but it's not related to what you're doing? Yeah, so let's say that one of the processes that we are accelerating, which are physics-based simulations, where the focus of my PhD or research studies when I was at Caltech, then I ended up working in the industry for two companies one is one of the largest engineering procurement and construction companies for the oil and gas field, and I was doing design of parts for extracting resources, let's say, and simulations and then I ended up relocating to Zurich, which is where I founded Fainite. And the first job that I got in Zurich was related to doing design and simulations of optical actuators for machine vision applications so it's like rotating mirrors that can be used for, for example, in an Amazon warehouse to track the QR codes of all the packages or many other applications. Do you have any competitors in the space right now? The uniqueness of our solution is that we are building something that is both fast and reusable and reusable is the key word here many of the other players are rebuilding these, let's say, surrogate models or machine learning models for specific physics from one client to the other, it's always rebuilding of a new model, training of a model from scratches and these models tend to be limited in the sense that they cannot generalize to new input parameters. If you're changing significantly the design of your part, the geometrical part, loading conditions, then the same model is not valid anymore. On the other end, we are bounding our model with physical constraints and that allows us to generalize to changing conditions, changing parameters. So I would say, yes, there are some players that are pursuing something very similar, but the core of the solution and the value that provides to the end client is very different. So essentially, physical AI will be a combination of AI training models for fine tuning or training, and also restrictions of the physics world? Yeah, you can see it as we are, we have a data engine that generates these data of different physics and the model gets trained, being constrained with the partial differential equation of different physics, as you have mentioned and also, we have another process that verifies all the results against the physical equation itself. What's the hardest part of going from academia to building a business? Now go on. I think, at least based on my experience, when you're like, in academia, when you're studying, you know what's the end goal It's very well defined you know how to exceed expectations you need to get good grades or finish projects on time and so on and so forth when you're moving to the real world in general, other than building a company, you need to set your own goals and you can be extremely ambitious or overambitious or even undershooting right So understanding what's the next goal that you're pursuing and where you want to go next, I think it's the main high level problem, the thing that you need to figure out first I would say doing research, it's similar to building a company to some extent, I feel like, although I'm not the one developing the technical product at the moment, there are always new challenges and you can experiment new ways to solve the challenges and in that regard, it's exactly like being a researcher, which is also the part that makes me the more enthusiastic about building a company. I would say probably the most complicated thing is to detach your, specifically for me, was to detach your technical self from, for example, if you're talking to a customer or an investor or anyone else that is not the guy building the technical product, what they want to hear is more like the outcome that they're going to get from every modification and as an engineer, you're not used to presenting things that way. right and this creates a lot of miscommunications and yeah, and also tension So probably this is the most difficult part in the sense that you need to rewire your brain and focus on the outcome and from the outcome, do the reverse engineering on how you want to explain how you're going to get there. How industries could be affected if your product was fully development, fully established? Yeah. Yeah. So an example would be manufacturing components within the ceramic industry, right? At the moment, figuring out how to manufacture them takes you six to eight months with the product at the regime. It's going to take you a couple of weeks, maximum a month, right? So you can like, it makes you more competitive as a company It allows you also to discover much faster new solutions to the same problems and ideally, it's going to lead to like an engineering world where you don't have to ask yourself whether or not it makes sense to verify something with simulations that are going to last four hours or days, but you just do it because it's so easy to do it and you can overdo it, right, to that extent. So you will maximize the performance or any of these physical products that you work on because you can run unlimited simulations. Last questions, a challenge that you have probably have. Pretty short. Yeah. So I would say that at the moment, we are very focused as we develop the technology, we are also deploying in production with several customers and understanding the best way in which you can deploy this model at scale it's the part of the, like, yeah, it's the part probably that's going to affect the most, the future success of the company and it's going to be very smart into making something that is as reusable as possible and easy to be just plugged into different workflows or different companies and different industries. Amazing. That was it. Thank you.
