Roadrunner Innovator Spotlight: Sarah Boisvert

What is fab.ai and what about it excites you??

I’ve worked in advanced manufacturing for a long time. I’m passionate about using emerging technologies like AI and 3D printing to develop new industries and help workers, so much so that I’ve even written a few books about it. Broadly speaking, people have been talking about the potential for 3D printing to change the world for two decades. The fact is, it’s just still too hard —?for experts and everyday people alike. Fab.ai is the first large language model (LLM) developed specifically for advanced manufacturing. I’m excited about this because it tackles perhaps the biggest challenge holding back the world of 3D printing: how to train people to actually make use of the technology.??

3D printing isn't like Computer Numerical Control (CNC) machining, which is the most prominent manufacturing process in which software controls the movement of factory tools and machinery. With CNC, there’s usually someone in the company with all the insider knowledge. You know, that person you can always rely on – let's call him Joe – who knows the answer to pretty much any question, not just from books but from hands-on experience. But when it comes to newer technologies like 3D printing, we often don't have a Joe in the back room. We lack that tribal knowledge in our companies. That's where fab.ai steps in. It's like having a mentor right there with you, ready to answer any question you have, without having to spend hours searching online or flipping through manuals. It's a game-changer for navigating these advanced manufacturing tools efficiently and effectively.

So, if I understand correctly, it seems like fab.ai could help us cut down on the time it takes for workers and beginners to get up to speed. That's essentially what fab.ai is all about, right? Making the learning curve a bit less steep for everyone involved.

Absolutely, enhancing the efficiency of workers is a key goal for us. Fab.ai isn't just for beginners; it's also for experienced professionals who might encounter new challenges. Ultimately, it's about maximizing yield. In additive manufacturing, yields tend to be lower compared to subtractive methods. By empowering our manufacturing teams to tackle tasks efficiently without spending days figuring things out, we can increase our overall output. Recently, we completed a demo of our proprietary LLM, in which we asked the LLM to design gears. While some might see it as just designing a single component gear, the complexity involved is significant, and it showcases the capabilities of our technology.?

Designing a gear is no small feat in engineering. There are numerous factors to consider, especially when scaling the size. Anything that involves movement adds an extra layer of complexity. For a beginner, it could take around three months of training on the software, followed by a couple of weeks of trial and error to work out the design kinks. So, all in all, it might take three to four months to design a gear from scratch. However, for our seasoned designers, it's a different story. Depending on the specifics, it might only take them one to two hours to whip up a gear design. But here's where it gets mind-blowing: our LLM can churn out a gear design in just a couple of minutes. It's unbelievable.

The leap from hours to minutes, or from months to minutes for beginners, is incredibly impactful in terms of productivity. When dealing with complex systems, assemblies, and sub-assemblies, the potential is even greater. We're not looking to replace designers; rather, we aim to enhance their skills. Imagine a scenario where a designer, even with experience, designs 10 gears for a sub-assembly. With fab.ai's assistance, this process becomes significantly faster and more efficient. Then, you can focus your creativity and experience on the broader aspects of the design, bringing everything together into a cohesive whole. It's all about leveraging technology to empower designers to do what they do best.

I sometimes think about it like Microsoft Word. It's amazing how technology has transformed tasks like typing letters from something tedious and messy into a quick and easy process. Fab.ai operates on a similar principle. We're not aiming to replace designers; rather, we see it as a powerful tool to enhance their capabilities and productivity by doing the rote tasks .

In the realm of 3D printing, especially in areas like powder-based advanced printing with metals and plastics, there's a lot of nuance and artistry involved. By ensuring that our designers and operators have access to the most efficient methods, we can significantly increase yields. Looking ahead, we're exploring simulations and predictive tools to anticipate and mitigate failures before they occur, saving time and resources. This increased productivity translates to tangible returns on investment for companies. It also broadens the market for the industry by making the technology more accessible and less intimidating. I remember back in 2016 or 2017, at the RAPID +TCT conference, people were eager to adopt new 3D printing technology like HP's latest printer, but ultimately people had no idea how to use it. With fab.ai, even technicians coming from other areas can be supported and mentored, facilitating the integration of new technology onto the factory floor. It's all about streamlining processes, increasing efficiency, and ultimately driving innovation forward.

It’s interesting to think about LLMs being applied to manufacturing. When we hear LLM, most of us think about ChatGPT, which produces text for us in response to a prompt on our computer screen. But it’s a whole different thing when an LLM can produce a physical product that we can touch and feel. What does that translation look like??

It’s really not trivial. Fab.ai’s CEO and Founder Michael Howard would explain AI to me from 30,000 feet, but I did not grasp the complexity of what they were able to achieve until we really got into it. What I’ve seen with ChatGPT is that it’s very shallow and broad, and it’s not very accurate. We did an experiment where we asked ChatGPT all the questions one would ask if it wanted to 3D print something. What we found was that ChatGPT was about 80% accurate. It knew a lot. But for the things that were off, ChatGPT was way off. With those results, we then decided to train the algorithm. The way you do it is by asking AI questions in such a way that leads it to the right answer. The only way I was able to do that was because I knew the topic area so well. So, it was clear the current system is really bad. Michael and Rohit, our lead technologist, explained that what they were going to create was going to answer all of my concerns. Then I started to see that translation. The difference is that our LLM is a large data set that is curated and deep. It covers things that are not on the Internet, like obscure research papers. And so, I feel that it is very trustworthy. When you make this transition you’re talking about —?from the 2D to 3D —?the LLM has to be correct.?

It's evident from your writing that you’re interested in understanding how new technologies are going to impact people who have traditionally worked in blue collar industries. How do you think something like fab.ai will affect them??

Digital technology has integrated into many blue-collar professions. Consider your UPS delivery person—they're now equipped with a computer. Gone are the days of signing for packages; instead, they snap a photo of your delivery on your doorstep. Our entire world has embraced digitalization. This shift is particularly natural for young digital natives who are accustomed to sourcing information online. I believe that fab.ai will only increase our ability to do things, which by the way, employers love because they boost productivity and profitability. The fact that fab.ai can go from prompt to product is truly remarkable. I've yet to encounter anything as intuitive and user-friendly as conversational interfaces. You can simply express your needs in plain language. Granted, you need to provide specific details for the AI to function optimally, but it comprehends real English, marking a huge breakthrough.

I'm curious if you could share a bit more about what it's like to interact with the printer. Is it akin to using an app on your computer, or is it more like navigating a website??

We can tweak the user interface however we want. I think for regular users like researchers or subscribers, it'll look like an app on your phone or computer. In factories, people often use tablets, so we might design it to work well on those too. On one side, you ask questions, and it types out the answers, like Copilot from Microsoft. On the other side, you see what you asked for, like 3D printing files or instructions for the machine. It's smart enough to show stress points if you're designing something like a table. The system adjusts based on what you need, like making models showing where something might break under stress. Big companies who get their own version of our system can probably tweak it even more.?

I can't wait to get a close-up look at this! My final question for you is an aspirational one. Your book really dives into how additive manufacturing could spark the birth of new companies and whole new industries. So, picture this: it's 10 years from now, and that fab.ai has become a huge success. What changes do you hope to see in the world because of it?

Well, my main drive behind all this is to revive the middle class and close the wealth gap. It's heartbreaking to see folks who followed the traditional route of going to college, racking up debt, and then struggling to find decent-paying jobs in fields they're passionate about. I mean, the dream is to provide for your family, right? Not everyone wants to be the next Bezos, though that'd be cool. But realistically, most folks just want a stable life for their loved ones.

Companies need skilled technicians and operators on the floor, not just engineers. Many businesses, like General Motors, are already shifting their hiring practices, which is a step in the right direction. We've seen this trend of "degree creep," where unnecessary qualifications are demanded for jobs that don't really need them. But with the right tools, we can empower workers who never thought they'd have a shot in the tech world.

Now, about those robots everyone's fretting over—I don't see them stealing our jobs just yet. In fact, they'll need humans to program, monitor, and fix them. And let's be real, machines break down. So, what we're witnessing is the rise of a new skill set, blending blue-collar know-how with digital prowess. Many people find joy in working with their hands, creating something tangible they can hold at the end of the day.

It's all about providing a path to a better life, and that's something everyone can get behind, regardless of political leanings or where you're from. Reviving the middle class, ensuring families can thrive—it's a universal goal that resonates with folks across America.