AI • IOT • NETWORK EDGE

Christina Cardoza: Hello and welcome to “insight.tech Talk,” where we explore the latest IoT, edge, AI, and network technology trends and innovations. I’m your host, Christina Cardoza, Editorial Director of insight.tech, and today we’re going to be talking about AI partnerships that spark developer engagement and innovations.

Who better to discuss this with than two companies embedded in the AI and developer communities. Today we’ll be speaking to Paula Ramos from Intel as well as Jason Corso from Voxel51. But as always, before we get started, let’s get to know our guests. Paula, a good friend of the show; for those of us who haven’t heard your previous conversations, what can you tell us about yourself and what you’re doing at Intel?

Paula Ramos: Yes, for sure. Thank you, Christina, for having me here. So, I’m so excited. So I, Paula Ramos, I have a PhD in computer vision and machine learning, and I’m working at Intel as AI Evangelist, working with multiple products and multiple developers around the globe.

Christina Cardoza: Great. And Jason Corso from Voxel51, first-time guest of the podcast. What can you tell us about yourself and Voxel?

Jason Corso: Likewise. Nice to meet you all. Thanks for the invitation, Christina. So, Jason Corso. Yeah, I have a PhD in computer science. I’m a Co-Founder at Voxel. At Voxel we make a software refinery to help you work with your data, your models, various needs, and kind of refine them into production visual AI.

I’m also on the faculty of robotics and EECS at the University of Michigan, where I’ve done research for the last 10 or 15 years in computer vision and machine learning, all at the boundaries between the physical world and what we can do with computational systems these days.

Christina Cardoza: Awesome. So you’ve been in this space for a long time now and have probably seen it evolve even—it feels like every day something new is happening, and it’s evolving even further. So that’s where I wanted to start off the conversation with you, Jason. If you could just talk about what you’re seeing in this space, how it has changed over the last few years, where we are today, and what are the trends shaping where we are.

Jason Corso: Yeah. Indeed, it has changed quite a bit in the last, even the last few years, also the last 20 years. Like when I was doing my PhD, we were looking at things about how you can use computer vision to understand gestures and so on to interact with the computer, and look where we are now, right? 20 years later it’s been quite a wild ride.

So last few years, let’s see. I think there probably are two major developments I would argue in the last couple years that really are driving the way we all think about AI. So the first one is probably pretty obvious, right? Like the availability of these large language models that capture huge token lengths and can embed actually natural human language into the language model that’s there to really give us a resource in which we can interact with rather naturally.

Now, I mean, there are an awful lot of questions around what their limitations are and their capabilities are, but at the same time I think it’d be easy to find lots of different applications, right? I think in the beginning of this year I wrote some quick note on LinkedIn about how I think LMs will evolve in this coming—in 2024, this year. One of those key elements that I thought was that we would see a true revolution in how we think about search—and just information search, information gathering, and all that and so on. And I think we really are beginning to see that.

I think on the other one, though, I’d probably point to an appreciation for the role that data has begun to play or has been playing in the development of various AI ML models. Everyone when you go to school, in grad school, you take your machine learning course, and you go and start training models to recognize digits and so on. You just go quickly download a data set, either it’s from some repository or your professor gives it to you, and most of the focus is on the algorithm.

And so we’ve built this culture of the model is king. But if you really think about what’s happening, even various leaders in the LLM space—to bring back to the first one—have begun to talk about the critical role that data, good data, high-quality data plays in this marriage of model, code, and data to build the AI systems that we’re using.

So I don’t know exactly where that appreciation is going to lead us. At my company, for example, we focus heavily on the role that data plays and providing developer tools for engaging with data alongside their models, rather than just expecting you to gen up some scripts to visualize your data or whatever, right?

But I think it’s good for me, because it’s a long time since when I was—like, 20 years ago my data sets were dozens of samples, hundreds of samples, right? Now we have data sets that are dozens of millions of samples or whatever. So actually managing them and understanding the failure modes and the distribution and so on is very difficult and requires, I think, new thinking.

Christina Cardoza: Yeah, absolutely. And you mentioned the search and information gathering. I’m definitely seeing on the consumer side AI being more prominent in these areas. When I search on Google or anything now, instead of just getting a list of links, an answer from Gemini comes up.

So it’s interesting to see how AI is evolving, but I’m glad you brought up LLMs, the repositories, and algorithms, and this data, and these models, because it’s really the developers that are pushing these advancements forward. A lot of times on “insight.tech,” we’re writing about advancements in manufacturing and retail and education, how businesses are using AI to transform their spaces; but what’s behind these transformations are really developers that are building these solutions that are working with LLMs.

So, Paula, I’m curious from your take, because you work with a lot of developers, you talk to a lot of developers in this space, what has their role been in keeping up with AI? And how can they even continue to compete in this space with all of the advancements and skill sets happening?

Paula Ramos: Yeah, that is a great question. I think that all of the developers are looking for their path every day because the things are changing so fast. But the main things that we need to have in mind as the developers—what kind of challenges we have—is that we need to drive innovation in a huge field that is there: artificial intelligence. So we need to be creative, we need to build intelligence applications, and we need to solve problems.

Maybe we have the same problems that we had 20 years ago, as Jason was mentioning, but we have better tools right now. We have a better way to approach those solutions, but we need to be so creative with those solutions. So, still we have a lot of tools, and we need to think about the final user of the applications.

So I think that there are some challenges right now in still we have room to improve: that is model development, data management, or how we can deploy those models in the easy way. We need to use a cloud system, or we can use an edge solution. We need to think about, independent of that, for sure, the skills that we need to find could be different, but basically having developers programming in different kind of languages, organizing or producing different kind of data sets.

Also something that is really important in this field is the open source community. Open source community is changing the cadence of the AI, because when we have these models open to everyone, they can access those models. So they can access those data sets and improve and improve those models round by round of those data sets, round by round.

So I think that the responsibility that we have as a developers is huge in this new era of AI. For sure, I think roles are in different kind of sectors. We can talk about manufacturing, retail, but more than that is what kind of problem we want to solve today. Could be complex, could be simple, but the solution always will be the simplest as possible, and this is the main challenge that developers have right now.

Christina Cardoza: Yeah, I love how you said we need to drive innovations, we need to create intelligent applications, we need to solve problems. Because developers aren’t in it alone; they don’t have to build it from scratch. They can leverage partners like Intel and Voxel and community members to make some of this happen.

For instance, I love that Intel has the Edge Reference Kits, and sometimes you guys are walking them through how to build a solution and giving them the code to do self-checkout or to build something in manufacturing, and they can just customize it after they learn a little bit more about it and how to do that.

So I’m curious, in what other ways can developers partner with companies like Intel, and how that’s going to benefit them to reach out into these different areas and to ask for help or ask questions and be a part of the Intel community or other open source communities?

Paula Ramos: That’s a great question. So, we have multiple channels right now. As you mentioned, we have the Edge Reference Kits that developers can access. In an easy way they can find a solution—complex problem with an easy solution—where we are trying to show them with tutorials, code, videos how they can navigate that specific vertical: manufacturing, retail, healthcare, LLMs as well, and working with multiple models.

Intel has a variety of solutions. Basically we have solutions—we have hardware accelerators for retraining, for fine-tuning models, but also we have solutions that work at the edge. Or also you can use your laptop—you can use your laptops to work with AI. And we are creating a specific framework that is called OpenVINO, where developers can use OpenVINO to optimize and quantize a model. That means that they can use the same infrastructure that they have, they can use the same computer, and they can run LLMs, optimize and quantize LLMs—INTEGER*4, for example—or they can use the integrated GPU that Intel also provide the processors.

I think that Intel with OpenVINO is enabling developers to easily prove and test these LLMs. And this is just one step behind the real solution, the solution that we want to put in the production systems. So they can create pilots; they can impress the bosses with the tutorials and examples that they can run in their own laptops before moving to the real or the final production system. And Intel has this possibility also. Developers can access Intel Developer Cloud to test multiple hardware before to buy that hardware. That is really cool. And also accessing accelerators and accessing also the latest, for example, AI PC.

So we are provisioning a lot of tools to developers, and also we have—I almost missed that—but we have an amazing repository where developers can test the latest AI trends. So we have OpenVINO notebook’s repository, where if something happened today, literally in two days we will see the notebook with that specific model, for sure. This is for the open source community. So you can test there, for example, Llama 3.1, YOLOv10, and the latest AI trends. And this is a great tool.

And the most important thing is we are not forcing developers to buy specific hardware to run those models as well, so developers can also run these models in the actual hardware that they have. We are also supporting ARM, and we are also supporting a variety of Intel hardware. Also integrated GPUs—that is the most usage, an integrated GPU—that we can see in the world.

Christina Cardoza: Yeah, it’s great that you are making it easy for developers to get started with the equipment or hardware that they have. And a lot of the kits and the challenges we were just talking about and repositories—these are ongoing things that are available to developers at any time. But I’m thinking about—I know recently, which probably feels like forever ago, you were at CVPR, and there was a competition and challenge going there. So that’s more of a one-off, timely challenge that is available sometimes to developers going to these events, having these things happen.

So I’m curious, Jason, because I know the company was also at that event, but there’s other events that you guys host or that you’re at that have these developer challenges. I’m curious, what would you say is the importance of developers going to these events, engaging in these communities, and participating in some of those competitions?

Jason Corso: Yeah, it’s a good point, right? I mean, even just before CVPR, Voxel had our first in-person hackathon, actually in New York City, and it’s that type of engagement where we really see excited developers engaging with new technology and then really trying to work together on new teams of solve a new problem.

That was really fun, but I think one key angle for developer events is obviously education, right? Learning new things. And I think if you take my earlier answer about how AI has evolved and think about a key trend for the future, a key trend that we’re seeing for the future is language combined with vision combined with new compute capabilities and openly available data, and these foundational models to really tackle new problems in what at Voxel we call visual AI.

I think we’re going to see increasing contributions to that effect, but how do you do it? What do you do? One has to go to developer events or other types of conferences like CVPR or whatever, truly, to really stay abreast of what’s happening there. I mean, for me it’s, in some sense, the educational side is very natural, right? I’m a faculty member; I teach. I’m not teaching right now this year, but last year I taught intro to computer vision. So three hours a week I was doing this developer event, in some sense, for 300 students to learn about computer vision.

So I think one thing we’ve learned at Voxel is this AI space is evolving so rapidly that it seems like everyone—even faculty members who’ve been in the field for ages—we’re in constant information-gathering mode. It’s impossible to stay up to date with everything from cutting-edge research papers on one hand, all the way to what are the new APIs and libraries that you have to then, that you have to learn.

And so to do this, at least at Voxel, what we’ve tried to do is maintain a weekly—at least one per week, if not more per week—sort of technical output that in some form of an event, like different formats, that really allows the community to stay engaged. So we have an events calendar at voxelv51.com that we can include in the show notes. I think we have something like two dozen events scheduled between now and the end of the year.

Just personally, for example, every Monday at noon Eastern I maintain these open office hours where anyone can sign into them—they’re on Zoom. We talk everything from—a couple weeks ago we were reviewing someone’s paper, and we went through slides and an actual technical model. All the way to like—oftentimes I get asked, “This is my first time thinking about getting into computer vision. What should I look at first?” Right? So, pretty broad. But we have some hackathons, virtual meetups, and so on. So I think that it’s like raw education just about foundational capabilities, but also these developer events really help engagement just from staying up to date with what’s happening.

Christina Cardoza: That’s great, and that’s awesome that you have those open hours that developers can just join and start to learn. I’m curious, because obviously there are virtual conferences, then there can be conferences in different parts of the world, and it can be tough for developers: they can’t go to all of them or there’s just so many out there, it’s hard to choose from. Is there anything coming up that you want to call out that developers should have on their radar? Or is there anything, any other resources available to them online, that you think that they should take advantage of?

Jason Corso: What Paula was saying earlier, being open source is like the gateway to fostering innovation, right? Like our software at Voxel51 is called FiftyOne. It’s on GitHub. We have the permissive licensing for the open source component of it, which is basically one user, one machine local data. You can fork it. You can submit PRs. We make releases—I think it’s on the order of every one to two months. Every release that we have has some content from our community, and we’ve been educated so much over the last four years since we released it about—from community needs and community contributions.

Most recently we have this new functionality called Panels, which—FiftyOne is basically a visual component as well as a software SDK for doing the work that we’re talking about here, like data and model refinement. But with Panels you can build functionality for the front end without knowing how to write React or JavaScript or anything with UX. You can write it right in Python, and all of a sudden you can still enhance the GUI functionality.

So I think those are great ways—actual events, but also just becoming a part of open source projects is another way to really to get involved in the developer ecosystem for AI.

Christina Cardoza: Yeah, absolutely. And I think it also helps, companies like yourself who have these open source models. You might not have picked up on something that somebody in the developer community picks up on, and they can really be a part of that community and make changes and point things out and contribute to companies and projects like yourself. So it’s always great to be a part of those discussions, see what’s going on, hearing what developers are talking about, as well as some of the ongoing challenges that they’re facing in these spaces.

Paula, I know OpenVINO—there’s a huge GitHub community around there, and you mentioned a little bit of the kits and some other things that Intel offers, but I’m curious, in what other ways does Intel foster that innovation and that community engagement for developers?

Paula Ramos: That is a great question, because we have been working so hard on that part as well. So we have multiple ways. We are creating multiple ways to create this innovation with developers. We have one program, that is the Innovator Program, where we have multiple developers around the globe that they can try, they can test technology. They can make their own applications, and they can share that with us. So just stay tuned, for example, in my LinkedIn or in my network as well: we are highlighting some of these innovators. This is one thing that we have. And basically they create their own repository. They fork their repository, and they create new applications or improve the application with the contribution.

So another thing that we have is Google Summer of Code. We have a program with Google every year where we have multiple proposals, and we have several developers around the globe as well working with us for three months with different mentors in the OpenVINO team. And, for example, you mentioned about CVPR.

So, we worked with Anomalib. There is a library that also we have in the OpenVINO ecosystem, and we have two proposals last year about Anomalib, and one of these proposals the student that was involved in Google Summer of Code and the mentors and the professor as well, they created a paper. The paper was submitted in the workshop of anomaly detection, Visual Anomaly Inspection Workshop at CVPR, and that was accepted. So we are closing also the gap in between industry and the academia with conferences. We are also participating with the students and developers in those conferences through programs as, for example, Google Summer of Code.

But more than that, for sure we are moving so fast also in relations with universities: what kind of things we can work with universities, helping them to create some research and research proposals that Intel also can support.

At CVPR we are sponsoring as well the challenge in this workshop about anomaly detection. We try also to invite developers, and we create a marketing campaign around the challenge to invite developers to participate in that challenge. We received more than 400 participants and more than 100 submissions. That was an amazing and remarkable number around maybe one month and a half that we received, and we can see how the knowledge is moving in using anomaly detection.

For sure, talking about OpenVINO we have multiple things. As I mentioned before, OpenVINO is an open source tool, and we have a repository where we have different kinds of contributions depending on the product. So we have OpenVINO, OpenVINO notebooks. We have OpenVINO build and deploy. In that repository, OpenVINO build and deploy, you can find all the Edge Reference Kits that we have been talking about today. OpenVINO notebooks, you can find the tutorial; and in the OpenVINO repository you can find the API.

So we have a huge ecosystem where we are trying to touch not just the inference part, also the training part with anomaly detection, Anomalib, and also OpenVINO Training Extension. So we have a huge ecosystem that I really want to invite all the developers and all the people that are watching this podcast or listening this to this podcast to visit those repositories, visit the organization, “openvinotoolkit” in GitHub, and you can find all the repositories that I’m talking about.

Christina Cardoza: Absolutely. It’s exciting hearing all of these different resources, all these different ways developers can get started. I’m excited to see, moving forward, what types of solutions and innovations developers continue to build, and I hope they take you guys up on some of these events and meet you—whether that’s in person or virtually. I know sometimes it can be intimidating when you’re getting started in these areas, but having companies like Voxel and Intel support developers, that’s great to see.

And I also saw, Jason, in addition to the virtual office hours, there’s availability to do one-on-one meetings. So if developers feel intimidated somehow or don’t want to ask a question in a group setting, it’s great that you guys are making yourself available to help developers when and where they need it.

So I want to thank you both again for joining us on this podcast. Before we go, if there’s any final thoughts or key takeaways you want to leave developers with as they go on this journey, engage with each other, and engage with yourselves. Jason, I’ll start with you.

Jason Corso: Great. Yeah, thanks very much. So, I mean, first parting thought would be that I think I just want to express my thanks to the developer community that we’ve built over the last four or five years. We wouldn’t be where we are today without the community. It’s such a vibrant and rich environment.

But the second thing is that actually we’re hiring; we’re hiring developers. I mean, actually across the board we are as we grow, after we closed our Series B earlier this year. But for this conversation, machine learning engineer roles, both for core engineering work as well as developer relations work. We believe in developers so much, so we hire individuals that are fully trained and can write papers, can write code, and so on, but their role is actually building bridges with the community.

And then maybe just the last parting remark is that we, as a company we are open source driven, but we do actually have dozens of customers that use our commercial enterprise version that we call FiftyOne Teams. It kind of relaxes that individual user local data work and allows you to develop the same functionality together in teams—in the cloud or on-prem. And we’d love to engage in conversations around FiftyOne Teams as well with your community. We have customers, many of which are in the Fortune 500, but across manufacturing, security, automotive—a pretty broad-base customer base. So, thanks.

Christina Cardoza: Yeah, absolutely love to hear about job openings. It shows this space is growing, this space is becoming important, and some of the innovations and transformations that we talk about on “insight.tech” wouldn’t be possible without developers. So, exciting opportunity for anybody listening to go join the Voxel51 team.

Paula, always love having you on the podcast. Thank you, again. I feel like every conversation there’s something new to talk about, something new happening in the AI space. So, curious what our next conversation will be about. But before we go, are there any final thoughts or key takeaways you want to leave with us?

Paula Ramos: Yes, for sure. So, first of all, thank you. Thank you, Christina, for creating this space to talk about what we have. And thank you also to Voxel51. We have been creating a great relation with Voxel51—different conferences, we try to share some space together.

And this also talks pretty well about that we have the real intention to work in the open source community. So we are open to work with all of you: try to find the best path to developers, because here the most important thing is developers. So, the company for sure is really important. We have a lot of things to learn from the company: what kind of products we can provide, what kind of tools we can provide to developers. And always we are thinking that we need to enable you to use this hardware in software that we can provide and you can accelerate; you can improve your pipelines and your workloads. That is the main intention.

We have right now a lot of things to share with you. So we talk about OpenVINO, Edge Reference Kits, but more things are coming in the future. For example, we have the new AI PC that you can try. We have a new engine in the microprocessor—that is the NPU, Neural Processing Unit—that we can also expedite and accelerate part of the conventional and generative AI, conventional AI, generative AI, process with that small device. This is one of the things that we can talk about in the future, Christina, for sure. Thank you again, and I’m looking forward to connecting with all of you.

Christina Cardoza: Absolutely, and you talked about earlier how some of these innovations or these tools you have available are making it easy for developers to start working no matter what hardware they’re using, and the AI PC just makes it that much easier for the AI development, deployment, performance of your solutions, all that great stuff. So I know Intel has a lot of resources around AI PCs that we’ll make sure to provide to developers as well.

But thank you both again for joining us today. Thank you to Intel and Voxel51 for these great resources and communities you’ve created for developers and spaces for them to get started and get that support. Until next time, this has been “insight.tech Talk.”

The preceding transcript is provided to ensure accessibility and is intended to accurately capture an informal conversation. The transcript may contain improper uses of trademarked terms and as such should not be used for any other purposes. For more information, please see the Intel^® trademark information.

This transcript was edited by Erin Noble, copy editor.