Conversations on Applied AI

Shawn Hymel - Technical Content and Community Outreach in AI, ML and IoT

October 25, 2022 Justin Grammens Season 2 Episode 27
Shawn Hymel - Technical Content and Community Outreach in AI, ML and IoT
Conversations on Applied AI
More Info
Conversations on Applied AI
Shawn Hymel - Technical Content and Community Outreach in AI, ML and IoT
Oct 25, 2022 Season 2 Episode 27
Justin Grammens

The conversation this week is with Shawn Hymel. Sean is a machine learning DevRel Engineer, Instructor and University Program Manager at Edge Impulse. He creates compelling technical videos, courses and workshops around edge machine learning that inspire engineers of all skill levels. Shawn is an advocate for enriching education through stem and believes that the best marketing comes from teaching, you can be found giving talks, running workshops and swing dancing in his free time.

If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events!

Resources and Topics Mentioned in this Episode


Your host,
Justin Grammens

Show Notes Transcript

The conversation this week is with Shawn Hymel. Sean is a machine learning DevRel Engineer, Instructor and University Program Manager at Edge Impulse. He creates compelling technical videos, courses and workshops around edge machine learning that inspire engineers of all skill levels. Shawn is an advocate for enriching education through stem and believes that the best marketing comes from teaching, you can be found giving talks, running workshops and swing dancing in his free time.

If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events!

Resources and Topics Mentioned in this Episode


Your host,
Justin Grammens

Shawn Hymel  0:00  

Jump into a community with a company or community that you like, and start doing things like, you know, making tools for them or, you know, making demos generally advocating for this stuff. You have to believe in it. You know, you don't want to you don't want to really fake it, start helping them out in ways that are are free or consult with them or contract with them. That's a great way to see if it's a good fit for you in the first place. Now, not every company does this. But I've seen this work for a number of companies where you know, some of their biggest biggest advocates and people who are engaged in the community and add value, you know, they offer them jobs. That's absolutely true.

AI Announcer  0:33  

Welcome to the conversations on applied AI podcast where Justin grumman's and the team at emerging technologies North talk with experts in the fields of artificial intelligence and deep learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real world problems today. We hope that you find this episode educational and applicable to your industry and connect with us to learn more about our organization at applied Enjoy.

Justin Grammens  1:04  

Welcome everyone to the conversations on applied AI Podcast. Today we're talking with Shawn Hymel. Sean is a machine learning Dev Rel engineer, instructor and university program manager at edge impulse. He creates compelling technical videos, courses and workshops around edge machine learning that inspire engineers of all skill levels. Shawn is an advocate for enriching education through stem and believes that the best marketing comes from teaching, you can be found giving talks, running workshops and swing dancing in his free time. So I'm excited to learn a lot about so thanks, John, for being on the program today.

Shawn Hymel  1:34  

Yeah, thanks for having me, Justin. 

Justin Grammens  1:36  

Yeah, so awesome. I've been tracking edge impulse for probably a number of years now. I mean, you guys have been really sort of making your mark in this space around really easy ways to deploy machine learning. And, you know, getting people like you mentioned, like I mentioned during the intro of people of all skill levels sort of involved. And maybe you want to tell our listeners a little bit more about what edge impulse is?

Shawn Hymel  1:55  

Sure. And I'll start off with a personal story around the idea of machine learning and why edge and pulse can help make it easier. So I want to say it was two, three years ago, I started really getting into machine learning. Where I was learning TensorFlow, I took Andrew Ng's class on Coursera, which is fantastic. By the way, if you if you're want to jump in, it's a great course he goes through all the math. So I'm doing this and I'm like, Okay, I've been doing embedded systems for a while, and I'd love to get it working on an Arduino or an STM 32, or some embedded system of your choice. And TensorFlow had this thing called TensorFlow light, which was it would translate machine learning models to a more compact form to run on something like a smartphone. And then at the time, this these two gentlemen named Pete Warden, and Daniel sitzen, yaki, were working on TensorFlow light for microcontrollers. And that took that TensorFlow light and made it even a smaller subset of operations. And the idea was to have machine learning models, specifically deep learning models, deep neural networks running on microcontrollers. So as you can imagine, these are very small, very efficient, doing things like, Hey, we're not going to work in floating point anymore, we're going to work in just integer math to make it even faster and more efficient, you lose some accuracy, but sometimes that's okay. Right. And so I started tinkering around with this, and the framework for doing that it was put together and really kind of rough to us, I, you know, I'm not gonna say hackery because these people who put this together were very good programmers knew what they were doing experts in their field, but every version would come out and it would break everything you had done before. So just controlling those versions was tough. And then the actual conversion process to get it to run on a microcontroller was quite difficult. And around that time, I discovered there was this company called Edge impulse that was trying to make that process easier. And not only did they make it easier, where they would control the versions for everything, and then spit out a C++ library or an Arduino library for us, I just because of that, I started using edge impulse for my projects, unless there was something really wonky I wanted to do, they also have a great interface that would maintain your data. So rather than going through like kolab, or Python or something to train my model, and then I would take my model and do this conversion process to get it on a microcontroller, I could just send all of my data to edge impulse, they would do the training process for you with some DSP work upfront, usually for one of your use cases, but you can customize that to your will. And then they would go through that process of creating that machine learning model that would run on a microcontroller for you. And the process was just so easy. I just started doing it myself. And I would I was active on their forums, I'd advocate for them Be like, Hey, this, this tool is actually really good. And then I knew somebody who worked at the company, they hired me to make a course or two courses for them. And then from there, they had me they were like, Hey, do you want a job? And so that's how I came to work at edge impulse. And so hopefully that gives you an idea both based on a personal story of what edge impulses they're trying to make this idea of running embedded and edge machine learning much easier for people and you don't have to be a embedded systems. expert in order to use their stuff, which is really nice.

Justin Grammens  5:03  

Yeah. And the other thing that I really love is you don't even need to have hardware like you can just use your old cell phone, right. And I guess when I say hardware, of course, cell phones are hardware, but you know, you don't need to have anything fancy or an Arduino or, or any, any sort of specialized processor, they've got some really cool tools that allow you to just, hey, let's just use the gyro on your phone to do some machine learning out of the box, right?

Shawn Hymel  5:22  

Yeah, absolutely. And, you know, a lot of the demos we show are kind of canned in the sense that they come out of our studio, which is we call our web interface, our back end, or our front end to the back end that does all the training for us. But it's all web based, right? All the uploading of data, all the training, that's all web based. And right now, you can push some demos to your phone, but you can also get a web assembly library. So you can use that in your own applications. If you're doing mobile development.

Justin Grammens  5:48  

Yeah, really awesome. So just really sort of lowering the barrier for people to jump in and start start playing around with this stuff. I liked that story too. Because it's like, here's a company that's doing something cool. I'm going to show some value, I'm going to jump into the forums, sort of, you know, engage with the community engage with the people that are there, and kind of prove my worth in some ways, I guess, right, they reached out to you and you started building these courses. And I think it's a nice little story to around, I guess anybody can do this. I feel like in new technology, you know, if you find something that you believe in, just kind of dive headfirst into it, right?

Shawn Hymel  6:20  

Yeah. And that's a number of especially Developer Advocate, experts and whatnot, people who are looking to get jobs, that's a, that's a good lesson, right? Jump into a community with a company or community that you like, and start doing things like, you know, making tools for them, or, you know, making demos generally advocating for this stuff, you have to believe in it, you know, you don't want to you don't want to really fake it, start helping them out in ways that are are free or consult with them or contract with them. That's a great way to see if it's a good fit for you in the first place. Now, not every company does this. But I've seen this work for a number of companies where you know, some of their biggest biggest advocates and people who are engaged in the community and add value, you know, they offer them jobs. That's absolutely true.

Justin Grammens  7:02  

Yeah, for sure. And I in some ways, I don't think you can fake it for a long time now, right? You probably put in hundreds and hundreds of hours and late nights and stuff like that doing this and and you know, because you were just passionate about getting into this space and working with this company. So as a dev rel engineers, that's that's your official title. I mean, what does that encompass?

Shawn Hymel  7:23  

So Dev Rel is developer relations. That is my official title. And it means I do a little bit of development. But really, it's, I'm relating to our developer customers, people who want to use our tool to make stuff and both advocating for our tool. But then the other side of that is also on the flip where I learn what the customers are using any problems they're having, and bringing those back to our development team. Right. There are some things I can fix myself, but there's plenty of things I don't know. So like a lot of back end code is complete Greek to me. I'm a novice at things like JavaScript and HTML, I know just enough to like, make a breakable website. So I take that back to our team saying, Hey, this is what our customers are complaining about. But it's also teaching part of advocating is teaching people how to both use our tool and a lot of the concepts involved, like machine learning, what does a good dataset look like? How do you test that? What is a confusion matrix? Those kinds of things?

Justin Grammens  8:15  

Gotcha. And so yeah, you're primarily the people on the other end of the line, I guess, here are developers exclusively or sometimes do you engage with some sort of other sales opportunities as well, people, other people within organizations?

Shawn Hymel  8:29  

Are you talking about like a partner or a customer who would come in looking to buy, sometimes we actually have a whole group of solutions engineers, and their entire CRI are, I suppose their purpose is to work with these customers, these partners who come in, right, we're gonna pay you money. And we're trying to do XYZ with your tools. And so they actually work directly side by side with these customers to help create their solution, right? It's almost like consulting. But it's definitely not right, because they pay a subscription. That's the business model, they pay a subscription to use our tool, and part of that includes this help for whatever they're trying to create. If that makes sense. I know that like Microsoft does this with their Azure team, they have a whole group of solutions engineers to help people use Azure to create IoT solution. So I see this in some companies now, where it's less, you need to have the expert as the customer. And it's we have the experts, and we will work with you to help create your solutions on our platform. This is kind of the model that I'm seeing. So we actually have a whole group that works with them, and I interact with them. There is another DevRel person who works more more closely with solutions in helping develop some of these are working with the partners, but I don't do a whole lot of that.

Justin Grammens  9:42  

Yeah. So you're you're really out at conferences, presenting the tools probably showing just like, like what's possible to anybody, and it's kind of through, as I mentioned during the intro, sort of education and marketing. How are they related in onto each other? How have you seen as you've sort of built up your your video library which I know we'll talk a little bit more about in the future or later on in the podcast. But yeah, how do you see that sort of teaching and education related to how a company markets themselves?

Shawn Hymel  10:08  

Normally we think of marketing as like Billboards, radio ads, TV ads, magazine ads, right? A lot of that is what's what's been dubbed outbound marketing, where I try to push right? Are you going to be looking at this, this content driving down the road? Can I get in front of your eyeballs? That's what we consider outbound marketing. A lot of these tech companies, especially these b2b we want inbound in that means that you as a potential customer, or potential potential lead, might be searching for how do I do something you may not know about us? So one thing we can do is create articles. Or we can go to events and run workshops that go, oh, how do we do this, and I teach you, me or another devil person, I teach you how to fix your problem, right? solve the issue that you're having. And hopefully, it's with the tool that I'm trying to sell, right? Like, I can prove along the way that this tool is the best tool. Sometimes it's not. Sometimes it's just here's a concept. And by the way, we also make a tool that's related to this, like I might show you, I might create a piece of content that says, here's how you actually convert a machine learning model from TensorFlow, to TensorFlow light for microcontrollers, even though that's kind of what edge impulse is doing. And then go, by the way, if you want to make this much easier, check out this cool tool called Edge impulse. Because this process can be made so much simpler for you if you use it. So there's pieces to that. But to me, that's all encompassed in this idea of I'm teaching you how to do something, right? You're on the internet, you're looking up, how do I solve this issue, and I can put this content or a workshop or a video in front of you that says, hey, let me help you with this. And I represent a company doing it. And both that is marketing, because you now know who this company is, and kind of what we stand for, as well as building trust with the customers. Because I think building trust is super important. And that's something that like a billboard can't do.

Justin Grammens  11:55  

Yeah, for sure. And a lot of cases, sort of the proofs in the pudding, right. And if I hire an engineer on at my company, the first thing I want to do is take a look at their GitHub repository. show me show me some examples of stuff that you've done. And yeah, experience is so much more valuable than I think, a certificate or, or a piece of paper or grades in some ways. So through this sort of inbound marketing approach, I think people can it's kind of, you know, the cloak is open, everybody can sort of see the amount of work that you put in and the value that you bring to it for sure.

Shawn Hymel  12:23  

Yeah, there's a book I just read, called developer marketing does not exist. That same book, it's a very short read, I read it like on a plane coming back for from an event by Adam DuVander, if I'm going to hopefully not butcher his last name. And that goes over this idea of Dev Rel, and what we just talked about here. And a lot of that points to another book called utility y o u t i l i t y, and that is by Jay Baer. That's one of the first one of these educational marketing books I read, that one's really good. And I highly recommend it to anybody who's curious about what this looks like. And the idea behind that is just like, be helpful. Use that as marketing, just be helpful to people. And that creates trust, there's no other good way to build trust, and just helping other people, you know, and then being open and transparent. There's plenty of other ways to do it. But you know, outbound marketing won't do it. So those two, those two books highly recommend, highly, highly recommend if you are interested in this idea of educational marketing.

Justin Grammens  13:19  

Yeah, I love it. I love it. That's, that's awesome. So yeah, we'll have liner notes and links and stuff like that in the podcast description, all this stuff. So I'll be sure to include that as well. You know, that podcast is conversations on applied artificial intelligence. And so I'm sure as you're working on this, you know, you've got some tutorials and stuff that you've built. But maybe you could describe a fun project that you've made in this sort of HTML space.

Shawn Hymel  13:41  

Sure. One of the, I guess, visually flashy ones that I've made in the last year, if you're familiar with the Super Nintendo and the Nintendo did the Super Nintendo classic, is the little box that comes preloaded with a bunch of old Super Nintendo retro games on it, that was my system growing up. So I played the EverLiving heck out of that. And so I got one of these, and I grabbed one of those cheap controllers that you can get off of Amazon. And I hacked it because the idea was of of these old fighting games like Street Fighters, the game I actually made this for is, is you would input the move with your buttons. And the the character would shout the name, how do you get and they do the move on screen, right? And so I'm like, how ridiculous it is in just just in general in these games. And these videos are shows you watch where they shout the name of whatever, but like, can you imagine in a real fight somebody just shouting punch or kick? Right? Like, utterly ridiculous. So like, Wouldn't it be funny if somebody had to like save the name of the move in a game to get the character to do it? So I made a basic keyword spotting system where you could say hi Dukan and I have the controller to input that combo for you. And then so I've got it where this is super Nintendo controller with this electronics like, basically bolted to the front of it, it makes the most obnoxious controller because it's all a prototype, and then wires running into a hole in the back of the controller that that jumped to the pads between the buttons that you need to control it. And then so I booted it up. And there's a there's a Adafruit show Intel, I can try to find that one for you for the show notes, where I actually demonstrate this working and I've got it on the screen. And you could show how to do it in the character like a second later, because right, it takes a minute to process and input all the controllers, the character would actually do the move. It's the worst way to control a video game. But I think there's something there for right like the new form of video games. It's not just buttons anymore, right in tendo has the Wii back in the weekdays, you could do the Wii bowling and the Wii Sports by actually moving your hands. So I think there's something to be said for controlling video games through our voice. But voice is the new frontier, we've got smart speakers, and there are games on echo devices. And I'm sure like Google Home and Siri has others. But this idea of like, you know, being able to interact on another level with games, I love that.

Justin Grammens  15:47  

That's awesome. I'm pretty sure I've seen that video actually on YouTube on your, your channel, and you go through the entire all the steps. Like I said, it's completely old with regards to exactly how you built the model, how you hack the controller and everything. And I thought that was fabulous. That was really, really awesome. And it really sort of speaks to this edge ml idea. I mean, people think about, you know, Edge machine learning seems like this sort of brand new thing that I haven't really even seen devices do. But you mentioned we I mean, that's that's a prime example of that, right? There's no internet connectivity needed really at all, there's just there's some intelligence that's put on a device. Same thing with your device as well, right? It's how do you have audio interact with the physical world, independent of the internet being even present? Right?

Shawn Hymel  16:27  

Yeah, and I think there's interesting use cases here. Because, you know, we're used to bigger and bigger models being more and more accurate. And, you know, that's definitely necessary for research for, you know, object detection and things like that, you know, self driving cars, we need these things to be powerful. And a lot of times it requires large computers or back end servers. So if you think about how an echo device works, it's not streaming data all the time to Amazon's computers, Amazon's servers, it's not, that would just require way too much bandwidth, you hose your network with two of these devices in your house. So they actually sit dormant and they're doing something called keyword spotting on device. And this is an application of embedded AI or embedded machine learning, it's listening for a very particular word a lax A, and I'm not gonna say it because I've got one right up here, and it will ding at me. So once it hears that word, and that's all like you said, that's all being done on device, and it saves a ton of power, because you're not transmitting anything, you're not streaming anything. And once it hears that, it wakes up. And then it starts streaming audio to Amazon servers where you know, more powerful algorithms, natural language processing can be done to do intent analysis to figure out what you're trying to ask of the whole system. So there's two parts, there's the title, front end embedded machine learning to do wakeboard detection, and then there's the backend stuff that does the heavy lifting for natural language processing. And then from there, you know, it can send audio back, you know, there's also text to speech where it sends audio back to you to talk in a lax A's voice. So there's a bunch of stuff going on. But in addition to, you know, that's a classic keyword spotting example, there's other uses for keyword spotting, right, this idea of, can I interact with other other electronics with only a couple of words? So like the idea of doing few shot, rather than having to go through the internet? Could I, you know, make a light bulb, that's, you know, hey, so and so room turn all and it only has to know a very particular phrase or a couple of key words, and it doesn't need to be connected to the internet, or have to be triggered through some type of smart speaker device or as a safety mechanism, right? Can I think of a giant machine? Can I shout, stop? If I can't get to that safety button? Can I shout, stop and have it actually stop? Right? So I think there's a lot of uses for just audio, just audio embedded machine learning. But there's tons of other stuff like predictive maintenance, doing things like object detection, or image recognition through really basic cameras. So I could have a doorbell that dings whenever somebody is on my porch. And I don't need internet connection to make that happen. I can do person detection. But you know, it needs to be it doesn't have to be high resolution, it can be you know, a very small model that has to identify one thing and a whole frame. And that's it. It's very specific, focused idea of what we're using this for. I'm not trying to identify every object out there, just one I just need identify one. And that's then you know, from there, I can trigger something. And I don't need internet connection. I don't need a big computer to make this happen.

Justin Grammens  19:13  

Yeah, for sure. I remember as you were talking about predictive maintenance, and some of these other use cases where they're completely independent. I had a guy on the program, this goes back a number of months, but he was kind of against even having to put machine learning at the edge, even if he even if even need to, I mean, sometimes these motors and for example, you know, there's just there's certain ranges, once they fall out of a range, you can pretty much say this thing's going to this thing's going to die pretty soon. So, you know, he was kind of pushing back on maybe we don't need neural nets everywhere. You know, all the time. I don't know if you like what your feeling is on on that. But maybe a simpler approach could even be done in some cases. Is that Is that true? 

Shawn Hymel  19:48  

Oh, absolutely. I consider neural networks to be the sledgehammer of machine learning, or not even just machine learning of, you know, this idea of data driven engineering where I just feed it a bunch of data just like go figure it out. And it comes up with an algorithm and it probably not the most efficient, especially if you if you know the problem you're looking for, like, if there's just a certain frequency that starts vibrating more than others, and I know this and a motor fails, then all I have to do is just look for that frequency, right? That's fast Fourier transform and look for a spike in something and like, Oh, it's just a threshold. Right? There's no machine learning there. But I think I think that there are cases where we don't have something better. So I think it's, you know, it's about choosing the right tool for the job. I think right now we're seeing an AI explosion AI summer as opposed the AI winter. I don't know what else to call it. We call it the AI winter, what's the opposite AI summer AI spring where everything's growing, where we're throwing machine learning at everything. And I think it's useful. I think machine learning opens up a lot of possibilities that we didn't have before. So like this idea of keyword spotting, it's tough to do that without, you know, there, I'm sure there's ways to do it without neural networks, but it becomes tougher and tougher, especially if we can train something that's like 99.99%. Accurate. And it's like, you know what, it's a little less efficient. But Darn, is it easy to train to make do this, we have tools to do it, rather than trying to figure out exactly what a particular spectrogram looks like, right? We just like just have a machine figure that out. So I think ML is useful in a lot of these cases, or like this idea of generalized predictive maintenance. Right. So if you know that a motor fails in has particular telltale signs, then that's great. I can make a predictive maintenance machine for that. What if I don't know what if it's just anomalous vibration data? And I don't know. So it's like, okay, maybe I need a little more processing power. So I can sell a device that I just slapped to a machine push a train button, it now says this is what normal looks like. And then from there, it looks for non normal data. And a neural network may not be the best approach there. It might be like, oh, let's just take the FFT do a bunch of averages together. And I don't know if something like K means right? It doesn't need to be a neural network. But K means is considered machine learning these days. Same with most regression models, right? Like your like root mean square for Oh, my goodness, I'm forgetting here. Least Squares. Yeah, least squares, real basic regression. Like, this was like the first I just did an Andrew Ng's class, but forgetting the name of it, then yours? Yes. No, I know. Yeah, it's like super basic. But this is it's like what you learn in statistics class, if you take like statistics, one to one, you learn about basic regression, and this is probably what you do. And it's now considered machine learning. So, you know, the idea of machine learning for predictive maintenance, I think is great. Is it going to be a neural network out?

Justin Grammens  22:28  

Right. And I think that's probably where the crossover comes in to this concept of deep learning, right, which is, which is really, which is really where the neural nets come in. And the idea of the all these weights and neurons and, and all that sort of stuff. So I believe that, you know, obviously, there's been this massive explosion of GPUs, really, really good data. Obviously, we're getting a lot of data now, these days, faster processors, and also ones that that don't require as much energy usage, I guess, at the edge. And where do you see the future of sort of edge machine learning going? You know, within with an edge in polls, for example, you guys have got, I think, a really solid sort of base and some really good examples. But like, I don't know, have you guys been thinking or have you personally been thinking about, like, what does three to five years look out? Or even a decade look like? And the idea in the world of Edge ML? 

Shawn Hymel  23:11  

Yeah, Ithink it's something we all think about, because it's going to affect the future of the company, right? It's, you know, are we going to have jobs in three to five years, and it's gonna depend a lot on this. So I think it's something we all think about our founders, Zack and Yan, they think about this all the time, right? Because they they base their sticking the entire company on this, so and the developers sales, everybody else, you know, are we gonna have jobs? So that's a big question. And I think, to me, it follows a similar trajectory of what DSP did back in the 80s. It kind of feels like that. Not that I was around there. But that, that had a lot of promise along, especially like silicon vendors, trying to be the DSP leaders. And how does that affect, you know, what does that look like for the future? And it's honestly mostly invisible, but it affects our everyday live our everyday lives, right? Everything, everything that we have, it's like, you know, my noise cancelling headphones, right? Oh, that's so cool. They're noise can't let's use a ton of DSP. So I think we're gonna see stuff like that. But especially I look at something like IOT. And remember what IoT promised five years ago, and that was like, oh, complete home automation. And it's complete bunk. We have light bulbs that can turn on via the internet, but they're like $20 a pop, and they're mostly novelty. I mean, other other than, like, what a few people pay for just to like, have them as a novelty or, you know, accessibility is also good, right? If reaching or light switch you can't do than being able to talk to your light bulb. Good, right? I'm happy for more accessibility stuff. But you know, the refrigerator that automatically ordered stuff for me never manifested itself, but at least not without something like machine learning to assist with knowing what's in my fridge. So I think to me, Edge machine learning is going to play very nicely with IoT, whatever we have in IoT now. And IoT has been mostly invisible. It's been mostly stuff in the industrial cases where it's measuring lab equipment measure, not lab equipment, measuring and dust Drill equipment, output at factories, you know, we can get dashboards and metrics, all of those are fantastic. And that's where IoT has been really seeing a lot of growth, not so much the home automation we were provided or were promised. So I think machine learning is going to tack on to that really nicely to be able to say, not just, Oh, what is the output of this machine? But what is the output of this machine? And does it need maintenance, or here's the defects coming off the lines, where without machine learning or edge machine learning, we couldn't identify defects, and now we can. So I think IoT and especially edge ml are going to play very nicely together, and really enable IoT to bloom not into what it was promised, necessarily, but into something way more useful for factories, automation, you know, network analysis, things like that.

Justin Grammens  25:49  

Yeah, for sure. No, I I'm with you. I feel very jaded by IoT. In fact, I hung my hat on IoT for probably the past decade, even before it was called IoT, you know, machine to machine. And, um, you know, that was sort of like the word that we used in the early 2000s, until maybe 2020 13, or 14 or so. And when everyone kind of glommed around the IoT, but, but yeah, or remote data acquisition, it's just that the whole concepts were there. But people promised a whole bunch of things. And I went through the Gartner Hype Cycle. And, you know, I started a business around this idea of helping companies with IoT, but at the end of the day, I do think it's kind of languished, for sure. And all these all these things, especially in the consumer space, but you know, as you look around, I do view voice assistants as an IoT device, in a lot of ways I view are connected appliances, that that do alert me when the washer is done, do alert me when I need to order more things, though, those are sort of, you know, yes, they're convenience items. But I really am super excited to see sort of the next wave, which is this overlap. And I've kind of started using the term AI IoT, which is the artificial intelligence of things. That's sort of a term that I've seen, a couple of companies start to use, but yeah, it's really the intelligence at the edge that's really going to provide this, this this game changer to people not so much the mere fact that it's connected to the internet and sending data. Yeah, when I worked at SparkFun, we just have a joke that we would, you know, use the IoT hashtag whenever we could, because that was the big thing and was 2013 2014 as it was, you know, kind of reaching that peak of hype, but I used to joke internally that it was like, oh, people building the internet of curse word Internet of garbage? Yeah. Oh, yeah. I think there is a there was a Twitter channel, it was basically, we can use that word of the internet of shit.

Shawn Hymel  27:34  

And I would do it too, right? I would create these fun projects, but completely tongue in cheek that it was like this is just garbage. No one's gonna ever use this. And like one of my one of the funny projects I made was that it was a kill switch for your Echo device. And you could be like a Alexei commit seppuku, and it would actually, like cut its own power. And like, I posted it on the SparkFun channel, and they were like, people were like, what's the purpose of this? I'm like, nothing. So yeah, it's kind of a it's kind of a statement piece that like, if you don't want it to be listening to you all the time, because you think it's listening to you all the time, which knowing how the tech works. It's not they could they could flip a switch and be like, No, it's listening to all the time, but it's ridiculous. I don't think they would not anytime soon. But it was kind of like, well, if you're real nervous, you can tell it to cut its own tower. But it's like, why don't you just walk over and unplug it? And you're like, Yeah, but that defeats the whole purpose of having a voice assistant that can kill itself.

Justin Grammens  28:25  

Yeah, I love it. I love it. Yeah, so just doing it for the sake, you can do it. And then it was just yeah, I, I haven't checked that out that Twitter channel, but I'm gonna be sure to post it in the liner notes here. People can take a look at it. Because there were some funny things is just like, let's just connect just the most ridiculous things and have this do these ridiculous things, just because we can. Right? So man good times, good times back in the early days of IoT. It's good to see that it's maturing. And you know that there is and I believe there is value. It's just a lot of times it'll take a decade for a lot of these technologies to work themselves out. And I feel AI is still in the it's still kind of in a in it in its infancy. I think I think you're right. We're in this spring, we've gone through these winters and summers and stuff like that. But I feel more and more, at least as we are here in 2022 that more and more actually applications like the rubber hitting the road. Like you can point at things and be like, no, no, no, this is actually saving us time and money. It's improving the quality of life. It's helping with natural resources. There's just there's all these things that are really, really good that are coming out using AI.

Shawn Hymel  29:21  

Yeah, from the home side. I love my smart speaker. Right. I talked to it, it gives me it gives me whether it gives me news, it'll play music. It'll be right. I love my smart speaker. So I hope to see more voice interaction with computing devices. I still think that's a big frontier. But I think there's a lot to come from voice. Yeah, I've like Right, I know that I turned on Windows one day and it was like you can talk to Cortana and like, I'm like still like No, I'm not talking to Cortana Right. So there's a balance there's a balance and like smart speakers are offering a value that people are okay having them in their homes and listening to so, you know, there's there's some security issues with it, to be sure, but I think there's a lot of that It'll be had by voice. There's a lot of growth to be had both from the consumer perspective as well as like predictive maintenance, what we talked about, I'm super excited for self driving cars, and also also the idea of like conservation efforts. So one of the big things that edge and Paul's helped with a couple years ago was identifying poachers for elephants. And so how do you do this? And so right, so the idea was to use object detection or image recognition, cameras, and you didn't have internet connection, right? There's no star link, there's no, there's no satellite. And yeah, those are power hungry. So an Africa somewhere? Exactly. So you strap these cameras to trees, and you look for individuals or sounds right? Or you maybe you have a collar around the elephant, and you look and you identify certain sounds. So you know, you you look for these kinds of things for both tracking as well as you know, identifying poachers, identifying what have you to help with conservation efforts. So I think that's a big one. And I'm hoping to see more use cases for helping with like, like climate change, and things like that, where can we be both more efficient? And I'm hoping AI can help us and I don't see AI replacing humans anytime soon, but just making us better, right, making us more efficient, better, our jobs better and our lives. So here's hoping. 

Justin Grammens  31:11  

Yeah, for sure. No, I that is one of the questions I kind of asked people on the program is, what is the future of work for humans? As these devices around us become more and more intelligent? Like how do you see us continuing to kind of bring value I guess, to the world and sounds like you're on the side of it's going to be more of a collaborative, I guess this this, this technology, we're going to be collaborating it and maybe having it do the stuff that we don't want to do rather than it being a full on takeover. We should be worried about our jobs in the future,

Shawn Hymel  31:35  

the AI revolution where you know, the the robot uprising? No, I don't, I don't think so. I think AI is still like massively in its infancy. If you look at what even deep neural networks can do, and all the technological prowess it takes to create them. They're still about as good as like a frog at identifying an object, right? It's or, you know, a toddler, is that a toddler? I think still way better. Because a toddler can account for like things like occlusion, where neural networks are like, I give up the cat behind the box, and I see a tail I give up. Or a toddler would be like, Yeah, that's a cat. I see a tail moving and neural networks like, no, no, that's a box, I got nothing. So I think I think ML is still very much in its infancy. And right now what we have, at least for the next 5-10 years, it's going to be using ML to enable us to be better at our jobs and identifying, identifying issues, looking for anomalies, just making things faster and more expedient for us, especially around that IoT realm that we were talking about when it comes to like edge stuff. But even the bigger stuff, it's creepy, but I for one love that Google now helps me type emails, right, I just hit tab for like half the email. It just types it for me. Like people were like, oh, Google knows what I'm talking about. I'm like, I got nothing to hide here. Just give me my whole email. I love that. And so things like that I think are fantastic and help us in our everyday lives. Personally, I think we have a lot of work to do before we even approach something even like human intelligence, or even something as smart as a frog. And that is we have to revisit this idea of of our hardware. I think the hardware is ultimately the limiting factor behind what AI can do because we basically created a computational machine. And then we have the computational machine think like a really powerful calculator, right? That's our CPU. And we're trying to get that to do things like pattern recognition that our brains are wired to do with milliwatts type nano Watts milliwatts, right. Nothing, no power, and we can do this immediate object recognition. And to do that on a computer, we're talking watts of power, because we're basically emulating, trying to emulate in our in our naive way of emulating what our brains do how neurons operate in a neural network. And it's really like this bad clergy effort to do it. So personally, like I'm starting to learn reinforcement learning where it learns over time, and learns optimal solutions to things or like, I'm gonna be able to play video. But let's be honest, I wanted to have play play video games. And that's where it all leads to games. Yeah, of course, I just want it to like, I'm so bad at a number of games. I'm just like, You know what, I'm going to make a computer program that does it better than I can. So I'm doing reinforcement learning. And then after that, I'd love to get into spiking neural networks and neuromorphic computing. So everyone talks about quantum computing, and I think that has potential in the future. But I would actually put my money on neuromorphic computing, for being able to actually create a better AI system. Because those rely my understanding, my limited understanding is those are built to handle from a hardware perspective, things like spiking neural networks. And they can do things like object detection, a with like, nano watts, it's incredible. But they're still very much in their infancy. And I don't know what the limiting factor is right now. I want to learn more about it. But there's like a handful of chips in the world that have been produced to do this.

Justin Grammens  34:41  

Wow. That's, that's fascinating. Yeah, I have not been sort of following that. I've heard the term before

Shawn Hymel  34:46  

and I could be way off, but this is like watching a couple of YouTube videos. Well, I

Justin Grammens  34:51  

think the crux of your argument is very sound like that. We need a new approach to how we're going to solve these problems. Because if you think back to initial computers, yeah, it was Very much a calculator. It was like, Look, we, we as humans cannot calculate this stuff fast enough. So we're basically going to take vacuum tubes and and we're going to like route signals certain ways, and essentially have a big abacus that's going to do stuff. And that's how it was built from the ground up. And that's how CPUs function. And I think what you're really saying is, is there's a new way for us to approach it at a hardware level, just at a very, very base level. Because these problems are different now, you know, 7080 90 years later than then then what we were sort of thinking about them and how we use computers in our lives.

Shawn Hymel  35:29  

Yeah, absolutely. And, and if we want it to even act like a brain, it needs to make and break physical connections, right? Think like FPGA. That's the other bonkers things, I think we think about a brain and it's on the hardware, I guess, the wetware level of you know, it strengthens neurons. And that's, you know, we give that a weight, you know, that's kind of a weight in a neural network. But it's we're really trying to emulate something here. That's, that's very complicated.

Justin Grammens  35:55  

Yeah, for sure. There was a, there was a book by a guy, he started palm, I don't know why I'm spacing, the name of the book, I actually just listened to it on Audible, like six months or so ago. But it really, really sort of like dives into what he is thinking the way that the way that our brain functions, the way that the sort of the neural pathways that are are built up. And it's, it's fascinating, because it's all chemical, right? It's basically chemicals that are happening within our brains that are doing these these these weights and measures, and the fact that it can do it so quickly. And these, essentially, these networks can be built and rebuilt and torn down and change and evolve over time, especially with memory. Just it's fascinating how humans do it. So really, really, really cool book, I'll find the name of that one and add it. But you know, one of the things I did want to talk about, we touched on a little bit at the beginning, around just using education and creating a video series. And, you know, as I've been doing this podcast over the past couple years, I've had a lot of fun having conversations with people, but I haven't actually moved this to more of a video format. Anyone that's listening to this, it's all just been strictly audio and not because I don't want to and I think video would would bring in a lot more interesting aspects to it. But when you go to teach something to somebody, I really think video has a different medium, right? That's why you go to a class and go to a school with somebody is you can be there a in person or B if not in person, there's a whole thing around actually showing and demoing something, you know, using video over over audio. Just curious, because you've been doing this for how long has your YouTube channel been up for a while.

Shawn Hymel  37:20  

So my YouTube channel, I read I don't post much to that, because it's usually I'm selling the videos. So when I was doing it initially, my first video came out in 2013, so nine years ago, but it was on Spark funds channel, the very first video I made was exploding different types of capacitors to see how they would react to undervolt or reverse voltage situations. Turns out pendulums like to explode and balls of fire, as opposed to your electrolytics, which just give you a puff of smoke, which is fun. And then your ceramics don't do a whole lot. They just kind of break and fail. That was a fun video. And I was super nervous doing that video if you watch me and like sweat. And I talk fast as it is. But I was like talking to real fast because I was so nervous on that video. And it was just something fun to do with the marketing department. And I did a couple of those like that. And eventually they were trying to grow their marketing efforts in like 2014. So they were you know, I was an engineer at the time. And they're like, Hey, you want to come over to marketing and do this full time? Sure, why not? So that's kind of how I started doing both Dev Rel and a lot of these educational videos. And I shied away from doing like the projects. As much fun as projects are. I shied away from doing because I noticed that I could consistently get views doing educational stuff. And it would continue to get views over time or as a project. Right? I think about projects, you know, they might go they might go viral every now and then you get like a mark Rober situation, or Simone your search, if I can pronounce correctly, you might get that situation where somebody just continually produces high level quality content. And more than anything, they have a great story around their project. But most of the time I found working like oh, I make the project it would be the, you know, Amazon kill switch. And I'm like, I get the joke. It's tongue in cheek and either I wouldn't present it correctly or what have you. And people are like, I don't I don't get it. So, you know, my projects like rarely go viral and like I still make them for fun. But I understand that like I can do educational content, and it will continually get views. It may not be 10 million views. But I can get 10,000 100,000 views pretty consistently from the videos. So I just kept doing those and got better at presenting this educational content. And I would always do my educational content not as like your content Khan Academy or your even like your theoretical class of like, let's work through the math. It was more of like, I just got out of school. I just got to engineering school. And I knew all the theory but I don't have any hands on experience because we had to use either we didn't do a lab, or what we use was like this TI chip. I'm not using ti anymore, right? That was just right, whatever it was. So like one of the first things I did I remember I guess a few years later I did like a whole PCB series with key Canada that's out of date at this point, but I did it I sold this to Digi key and that did really, really well for a longest time you've typed KiCad into Google, like my videos were there. was to come up. And people have done like later versions and better stuff. So but like for a few years, you're like, I'm teaching people how to layout PCBs. And so those did those did really well. So that's where I discovered this educational content works really well for marketing, because people would come to our channel, and watch this stuff. And it's like, oh, I build trust with SparkFun. Or I build trust with Digi key or Jim post, wherever it is. And that's where I, you know, like, I sell my services, making videos every now and then I put something on my own channel. And that was like, you know, the Super Nintendo controller, while I'm teaching myself, you know, TensorFlow light for microcontrollers, or like the pumpkin that farts when somebody comes near it, right, those kinds of fun projects I put on my channel. And honestly, like, the content that did best on my own channel was the educational content I made around microbead. And those were, like, poorly filmed in my apartment in Boulder. But they did super well for microbead for the longest time. So most of my subscribers came from like two videos, and those were the ones once again, educational content,

Justin Grammens  40:52  

I always am kind of harping on the new technology. And I was I was very, very early into Android development. And I created a blog post on how you essentially can create HTTP REST client to do post and, and puts and gets, and this was very early on, but boy, that was the most by I mean, bite by X 1000s percent, most viewed blog posting that I that I did on my website. And it was like, hey, you know, people are googling for this stuff. Like, they don't know how to do it. And I'm essentially the source of truth. And it's, I think it's cool, Shawn, that, you know, yes, you're helping spark fun. You're helping Jim Paul's helping wherever it is, but But you are your own brand, too, as well, to be honest, you know, people are coming to see you. And they're going to continue to follow you no matter where you go in your career, or 510 15 years from now or no, if you ain't no, if you can continue to put out good content that helps educate people, they'll continue to follow you and it'll open up a lot of doors. I think it's awesome. I think it's a great, great way to do that.

Shawn Hymel  41:45  

Yeah, it's worked really well for me. And it's it was definitely a very deliberate effort to make my own brand around it, which is why I always wear a bowtie. Whenever I'm you know, in a, I'm not gonna say in public because, you know, I'm not putting that on to go to the grocery store. But for videos where if I go to an event, you know, I'm easily recognizable. It's kind of the shtick, I'm starting to play with like some Tik Tok videos that are more in the moment of me, like, Oh, I'm tinkering with this. And I've got messy hair, and there's no bow tie. And I'm like, Oh, I'm tinkering with this. I'm trying to play with that. Just as like, Can I make a more casual kind of brand thing because it's also like, a complete pain to go dress up and look nice. And, you know, comb my hair and put on the bow tie, like it's a whole 30 minute ordeal. So if I'm gonna make a quick in the moment, tick tock like, I'm not doing that.

Justin Grammens  42:25  

No, well, plus, I feel like your videos, you have different cuts, you snap to certain things. I mean, there's definitely an editing process that somebody does behind the scenes to clean these things up at the end of the day, they, they they look fabulous, but that is time consuming.

Shawn Hymel  42:37  

Yeah, I used to do it myself. For my own videos we had we've always had editors since I was there at SparkFun. And I'm actually know hiring somebody Grace Flix, I think she has her own website. So if you want to check her out, she does great work. She helps me edit all of my videos now. So but I usually come up with like a shot list, like a script or a shot list. And I provide that to her so that she knows kind of where to edit stuff. And then I'm working on one right now that can be a little more story driven, where I kind of just blab at the camera, at least for the beginning of it. And I'm like, let's, if you can edit this, and I'm going to find some B roll. Like let's go back and see if we can add B roll and make it a little more enticing because I don't know, right? I think about like these not streamers, but people who make these videos and they're always have these cuts and things. And it's like they're casually telling the story. But then there's like this interesting things going on visually. So I'm trying to figure out how to get that going in my own stuff right now. Because right now, because right now all my educational stuff was very planned out, like I need to show this thing on the screen, or I need to show this diagram while I say these exact words. And it feels very scripted is the unfortunate thing, right? It never has that that casual, I'm telling you kind of what's going on. I've tried working on getting better at that. So I'm just trying some different things to see if I can be a little more casual, but also tell a story and remain technical at the same time.

Justin Grammens  43:51  

That's definitely definitely a balance. I'm curious to know like, is there any sort of content length that you have found that has been like kind of like a sweet spot? Like with regards to like, is a five minute 10 minute, 60 minute? Or does it really not matter?

Shawn Hymel  44:04  

It's dependent is the best answer I can give. A lot of times if it's, you know, if somebody's searching for how do i xyz? How do I you know, set up this tool chain? Right? The shorter the better, because let's be honest, somebody's just going to skip through to the part that's going to answer their question. And it saves you a lot of editing time. If you can make it short. Just be like, here's how you do it. YouTube, I think gives a little more preference to longer like 10 minute plus videos. But let's be honest, the Google algorithms are going to show what answers people's questions if you're trying to do these like how tos when it comes to entertainment, stories, entertainment, you know, even like projects I think like I go to like Mark rubbers exploding glitter bomb package that he made.

Justin Grammens  44:43  

Yes. Yeah, my kids watch that stuff.

Shawn Hymel  44:47  

Great. And the reason that did so well is not because he dove into the tech not I wanted to see the technical aspects, but it didn't get 100 million views because he went into the technical aspects. He told a good story. That's how you get The views is you have to tell a good story. And it was compelling, right? Like, oh, let's get the it's the classic revenge plot. And it's a problem people can relate to it like, oh, is my package gonna get stolen off my porch. It's a very relatable classic revenge plot, like the hero's journey, like it meets all these like classic storytelling elements, but from like a Mythbusters type of a thing. And Mythbusters was also very much classic storytelling, right? They won't even show many technical details, they're like, let's do a montage of us making it and like now we're gonna spend 20 minutes whisking commercial sprinkled in of us testing it is more about the storytelling. And I want to say it was Walt Disney who had the phrase, it was something like entertain first and educate second or something like that, I would have to look up that I'll follow find that exact quote. But it was something along those lines. And I'm like, I'm learning that I'm learning that like, if you want something to get viewed, you got to make it entertaining. But there's something to be said for a procedure of creating, how to content technical content that solves people's problems. That says that you title it correctly, to meet your SEO, your search engine optimization, people land on it, watch it, you slowly build trust with them over time. And it doesn't have to be me or somebody who's human, amazing technical presenter, you could write it in a blog post form and answer people's stuff. And so but then that's repeatable, and that's repeatable, that a company can do and that's the key, you don't need a particular personality. To do it. You just have to be pretty good at technical writing and explaining like, step by step how you do something.

Justin Grammens  46:22  

Yeah, yeah, for sure. For sure. I love it. I love it. This has been great. This, this has been great. Sean, any, I think I think you know, as you continue to build out your channel and continue to see what edge impulse does is just going to be an exciting future for sure for you and for whatever company you work for. So really, really cool. I did finally remember the book. It's called 1000. Brains by Jeff Hawkins. So if you haven't read it, I highly recommend just checking it out. It's it's really, really cool, talks a lot about sort of how the human brain works and how these neural networks are built. So both from a physical standpoint, but then also sort of how he sees it going in the future. And again, he was the founder at Palm and he was it near the end of the book, he talks about him talking about handheld computers, like back in the 90s and just completely been poo pooed you know, everyone being like, No way, there's no way you know that people are going to actually want these handheld things, which, you know, you think about it. It's basically a cell phone these days, right? Or a smartphone. But he was he was seeing it literally 30 years, I think, you know, in advance. So cool stuff. Anyways, I'll put a link to that. How do people reach out to you, Sean,

Shawn Hymel  47:23  

the best way is either on LinkedIn, so shining email on LinkedIn, or through my Twitter, at Sean hemo. I'm also on Instagram, that's like at Sean underscore email because Instagrams whole thing failed. When I tried to create an account with my other name. Without the underscore, then I'm now tinkering with tick tock to see what that's all about. So at Sean email there,

Justin Grammens  47:43  

well, awesome, Shawn, I appreciate the time again today. And, you know, this was a really great conversation. We talked a lot about artificial intelligence, machine learning, you know, talking about the edge talking about video, I mean, I really, really respect what you're doing here in the educational space with regards to just getting good content out to people and helping them you know, become better technologists. It's, it's really, I mean, I think I am at where I am in my career, thanks to all of the teachers that I've had. And the Internet has just been beautiful here to really expose everybody to these new areas. And you know, I'm still learning machine learning and artificial intelligence thanks to people like you and, and others that are doing really, really cool stuff on the internet. So appreciate you taking the time and being on the program today and, and sharing with us your story. Yeah, this

Shawn Hymel  48:24  

has been fantastic. Thank you so much, Justin.

AI Announcer  48:28  

You've listened to another episode of the conversations on applied AI podcast. We hope you are eager to learn more about applying artificial intelligence and deep learning within your organization. You can visit us at applied To keep up to date on our events and connect with our amazing community. Please don't hesitate to reach out to Justin at applied If you are interested in participating in a future episode. Thank you for listening