Zach, fantastic to see you here in Newberg, and congratulations. So excited to have the Edge Impulse team and Yan and yourself. It's just an incredible milestone for the company.
Well, we wouldn't be more excited to join the Qualcomm team. So tell me, when we met for the very first time, when I guess we spoke on the phone in October, what was going through your mind when you thought about Qualcomm as a potential prospect in the future of EDM plus? Yeah, I think...
Qualcomm's always been an amazing enabler in the whole industry, right? Creating whole industries from scratch. But it's kind of held back from serving the developer, right? High volume, top tier customers, best in the world. So my first thought was, oh boy, how are we going to go fix the developer experience? Right? And then, are they serious about developers? So I remember the one thing that you told me when we spoke on the phone that time, that is, I'm so excited that you guys are on board. You know, what comes to my mind is this is do-it-yourself AI. It's such a complicated space in the sense that people have been doing things in a non-automated fashion, and now the ability for machine learning and AI to just get democratized through the Edge Impulse tools, I think that's just phenomenal. We touch so many different ecosystems, we have so many different end markets that we are a part of, and I feel like we are at a point in time where AI has the ability to get deployed at the edge and automate so much of what has been manual. What are your thoughts on that? You've been at this for so long.
The reality in IoT is that we have so many different market segments, we have a complicated value chain, right? Component vendors, ODMs, OEMs, SIs, everybody has a role to play to get products to market. And the reality is that the data that's interesting to solve the real problems, it's sitting there with the end user. It's not with the SI or the ODM. End users know the problems they want to solve in manufacturing, for example. So Yan and I have always been passionate about democratizing the data that those end users in these industries have access to. How can we use that data to go train very practical machine learning algorithms, solve that problem, and deploy that to their edge hardware? So if you combine all that with the size of the market, you're talking about millions of developers having to hit billions of devices on hundreds of different platforms. It's a huge democratization problem. You know, when we started to look at industrial IoT in earnest in the last 12 months or so, the realization was that with AI at the edge, pretty much every single industry is going to go through a form of transformation because everything is connected for the most part, different forms of connectivity, but everything is connected. But it's not the same footprint of compute at the edge. And it was pretty clear that yes, you have computer vision, yes, you have gen AI, but you have so much data in the edge that has to be extracted and converted into something where gen AI can literally bring it to life, make it human consumable. When I see Yan showing what he's showing in terms of, hey, I'm going to send a simple query through an LLM to understand what the context is, I feel like any problem statement that I look at can just be rethought and implemented so quickly. When you started down this journey of building the Edge Impulse Studio, did you have that in mind, or is this something that has evolved over time? How did you go about it? What was the start for your vision for the company?
Absolutely evolved, right? You never know where the technology and the developer and customer requirements are going to take you. So the way that I build startups, our philosophy is, what's our North Star? Right? And so our North Star has always been how do we democratize for a million developers and enable them to build the machine learning models that they need. So we didn't know exactly, was it going to be sensor fusion, audio, was computer vision going to be the big thing, would it run on high-end CPUs with acceleration or MCUs? We didn't know. So what we started out doing was say, hey, we'll build a framework and a platform that serves the developer, and we'll be open to any type of compute size, right? Any type of algorithm class, any size of machine learning or data set. And I think that's the thing that prepared us for where we are now, where computer vision got big, discrete computer vision, we started running on MPUs and CPUs, and now with gen AI coming in, we have the ability to do very flexible prompt engineering to solve problems. That was the cascading architecture that you just mentioned, the IQ9. Yeah, this is so powerful. And so now we can extend the platform to do prompts, fine-tune, deploy down to device, and very quickly solve a problem. And so no, we didn't think about gen AI at all. In fact, just like Yan said years ago, we never would have believed that you could use gen AI for what we're doing today: synthetic data generation, AI labeling, VMs on the edge. No way, it's unthinkable.
Readiness. I think in two months alone we've seen a huge improvement, right? I mean, RB3 and this type of specialized developer platforms, they typically take weeks if not a month to get up, you know, update Yocto, install all the tools that you need, right? Get up your own developer environment. So the experience to get started with was slow. But what we saw was the Qualcomm teams taking a developer-first attitude, where no, we're going to simplify the experience, right? We're going to put more advanced ready-to-go Linux distros on the device, we're going to put IDE tools in place. And so just in a matter of months, that process has gone from weeks to a matter of minutes. So now we can take an RB3, we can directly deploy from that pulse the entire package end to end to run an entire AI workload and live stream from the device to the cloud. So very cool, right?
