403: Engineers Are a Difficult People
Transcript from 403: Engineers Are a Difficult People with Shawn Hymel, Elecia White, and Christopher White.
EW (00:00:06):
Welcome to Embedded. I'm Elecia White, alongside Christopher White. Our guest this week is Shawn Hymel, and we're going to talk about teaching, learning, and maybe some marketing.
CW (00:00:17):
Marketing? Hi, Shawn. Welcome to the show.
SH (00:00:21):
Hey, thanks for having me. Yeah. I'm excited to talk about all those things.
EW (00:00:25):
Could you tell us about yourself as if we met at, I don't know, Hackaday Supercon?
SH (00:00:32):
Yeah, absolutely. And you picked my favorite conference, because that always feels like a gathering a family to me. You probably recognize me wearing a bow tie. That's kind of my signature move. So if you've seen any Google videos or YouTube videos with me, I'm generally in that bow tie.
SH (00:00:48):
Right now I'm not wearing it, because I don't need to be seen on a podcast. So I will make that admission. Part of who I am is creating technical content for people to teach them. I want to make a electronics accessible to ideally everyone. And I recognize that a lot of the topics are fairly complicated.
SH (00:01:09):
So I do my best to break those down and teach them in whatever manner I can. And that includes video. Maybe it's written. Maybe it's demonstration, whether it's a project or demo code that can be shared via GitHub.
SH (00:01:23):
I just want to teach people and get them excited about electronics, about machine learning, about embedded systems, and so on. I did some work at SparkFun Electronics as an engineer. I moved over to marketing to do more of that teaching and marketing side. Part of that included learning more about marketing.
SH (00:01:42):
And then I worked for myself as a freelancer for a little while, generating content, a lot of blog writing, a lot of videos. And much of that appears on Digi-Key's YouTube channel, where I did a number of series around KiCad, FPGAs, real-time operating systems.
SH (00:01:58):
And I recently took a full-time position at Edge Impulse to do more or less the same. I'm a DevRel, or developer relations engineer, where I'm making content, giving webinars, going to events, and generally teaching people about embedded machine learning.
EW (00:02:13):
Alright, we want to do lightning round, where we ask you short questions, and we want short answers. And if we are behaving ourselves, we won't say, "Are you sure," or, "Why," or any of those things. Are you ready?
SH (00:02:24):
Yes, let's do it.
CW (00:02:25):
Polka or swing?
SH (00:02:27):
Oh my goodness. Swing all the way.
EW (00:02:30):
Favorite dev board.
SH (00:02:32):
[Ooh] man. You know what? I continually go back to...the Arduino Pro Mini, the one that SparkFun put out years ago. I use that in almost everything and that 328P is just tried and true. It works almost every time.
CW (00:02:50):
Introvert or extrovert?
SH (00:02:52):
Definitely introvert, despite what a lot of people may think upon first meeting me.
EW (00:02:57):
Engineer or educator?
SH (00:02:59):
Can I pick yes?
EW (00:03:01):
Sure.
CW (00:03:03):
If you could teach a college course, what would you want to teach?
SH (00:03:06):
So I actually have a couple of courses that I think are in line with kind of college, but they're pretty compressed around embedded machine learning. They're up on Coursera for Introduction to Embedded Machine Learning and Computer Vision with Embedded Machine Learning.
SH (00:03:21):
I would say they're kind of at a college level, but I compress and abstract away a lot of things. That being said, if I could actually choose what I wanted to teach, it would be something called Life Skillz, with a "z" obviously, or just How to Be a Good Person.
SH (00:03:35):
Kind of like Home Ec where you learn basic, how to stay alive, manage a budget, cook yourself food, but also how to socialize, deal with rejection, not be a jerk in the world. I'd love to teach a class like that.
EW (00:03:49):
Where do you source your bow ties?
SH (00:03:53):
Anywhere I can get them. As old white guy as this sounds, but Brooks Brothers make some of my favorites. There's a local place out of Boulder that I am forgetting the name of right now. Carrot & Gibbs is the name.
SH (00:04:08):
And there were a couple of others that I like. But it's usually Brooks Brothers or any type of local shop that I can find. When I'm wandering around cities, I will try to pick some up.
CW (00:04:19):
Do you have a tip everyone one should know?
SH (00:04:22):
Let's see. A tip that everyone should know. Let me think about this one. So, general tip I like to think about is, I'm going to go to that life lesson here as far as when it comes to engineering, and you want to sell your thing, or you want people to use it.
SH (00:04:41):
The idea of "If you build it, they will come" 99% of the time doesn't work, which is where it helps to know a little bit of biz dev, a little bit of sales, and a little bit of marketing if you care to have other people use the thing that you're making.
EW (00:04:56):
Well, that's actually going to make a good transition. So let's just go right there. So we hired a social media person a little while ago, and we haven't really gotten a lot of new listeners, although I think we've strengthened the channels, and our newsletter looks nice now, and all that.
EW (00:05:15):
But the podcast market as a whole, since the pandemic started, has kind of been going down. And we'd like to reach more listeners because I really think we do a good job. Chris does a fantastic job on making us have good sound.
EW (00:05:32):
And I think we help people stay in engineering or get into engineering and enjoy technology in a way that many other people don't...I mean, Shoeless Joe Jackson, that was the guy. I've always had a, "If you build it, they will come" mentality.
EW (00:05:55):
And my first job was at HP, where it was said that the marketing would advertise sushi as cold dead fish because they were very, "Here's what it is." And yet we've stagnated, and I want to do better. How do I do better?
SH (00:06:11):
That is a super broad question, but...that's definitely the type of question that I think most marketers ask themselves about them, their company, the product, what they're trying to market.
SH (00:06:26):
There's a saying in the marketing world that something like 50% of all marketing doesn't work and 50% works. And the trick is figuring out which 50% works. And it's a constant game of figuring out what works in your particular market, what resonates with your audience, and there are scientific approaches to doing that. And there's a good book around it.
SH (00:06:49):
And the name of the book is escaping me right now. So if it's something I can throw into the show notes later, I'll get back to you on that.
EW (00:06:56):
Yes, please.
SH (00:06:57):
And the author is a legitimate researcher, a PhD in marketing, and she breaks down how to approach marketing like a scientist. It's a pretty technical book. It's super dry. They talk about areas where you can expand in the sense of, "How do we create product? How do we create social capital?"
SH (00:07:16):
All of these things, and you can map them out. You can find ways to measure, and you iterate on that. And it's not as simple as just, "Oh, I think that the color red is going to appeal to this audience. Let's put it on our site." You measure it. You do red versus blue, and you learn. And it takes time.
SH (00:07:35):
And I think that's the big thing a lot of people don't understand about good marketing is, it takes time. We always hear these great stories of, "Oh, I imagine that the sound of the toothbrush is what draws people in. And they do it. They make the toothbrush sound a certain way. And it generates a hundred times revenue."
SH (00:07:55):
There are a few stories that do that from a couple of marketing geniuses throughout the years, but reality is that marketing takes time, effort, and a lot of science, a lot of hypotheses that you have to test, and come back, and iterate on, and build. And that's really hard.
SH (00:08:10):
And especially when people just want to say, "Oh, I'll pay you for one video, and that's going to generate a million views." And you're like, "Not really." That's not quite how that works if you're trying to generate a market for your product. And in this case, I guess it's a podcast.
SH (00:08:23):
Just the one video on YouTube doesn't generate an audience. It might get some views. If we're lucky,...it goes viral. There's no viral button. It's just, you hit on something that the audience resonates with, and it causes them to share. And there are some ways you can tailor that, but there's no guaranteed, "This video will be viral."
SH (00:08:44):
So all of those things in mind, marketing is a tough thing to do. And you talk about the podcast. I find it interesting that podcast listenership has gone down since we ended the Hello Blink Show.
SH (00:08:58):
I haven't paid attention, because I thought there was an uptick initially during the pandemic as people were stuck inside, hopefully listening to more podcasts. So that's interesting for me to hear that in general, it's gone down. And it's a question of who is your audience.
SH (00:09:13):
That's always where I start with. Who do you want to attract? And I know that for Embedded.fm, it's a lot of very technical people, and you've got great guests. You do a great job on the show. You have a good product. There's nothing wrong.
SH (00:09:24):
I looked at the iTunes stats and you were blowing Hello Blink Show out of the water. You've got five star reviews and lots of listenership. So I think you have a very good base, and it's how do you reach beyond that? And there are a number of ways to do that.
SH (00:09:40):
You can grow a community, say host a Discord channel, host a forum that draw people in, but that's not your top level funnel, right? That's just keep people engaged. When you start talking top level, you start thinking SEO, organic.
SH (00:09:54):
How do people search when they search for something on Google, right? What are they looking for? And so you need to tailor your content around that...This is a blind spot of mine. I don't know how good Google is at saying, "Let's find what's being said in a podcast and make it searchable on the internet."
SH (00:10:14):
So really that comes down to having great show notes and making sure you hit all of your good SEO things. It needs to be more than 500, 800 words, whatever it is at the given time, maybe slap in some pictures, link to things so that people can click through them, and make it a good page.
SH (00:10:35):
And the big thing is to make sure your title, when you're picking a title, that people can find.
EW (00:10:38):
No, no.
CW (00:10:41):
No, I won't do that.
EW (00:10:42):
We refuse.
CW (00:10:42):
Never.
SH (00:10:44):
I'm sure you all know all these things. This is marketing 101 type of thing.
CW (00:10:47):
Yeah.
SH (00:10:47):
How do you draw people in? You can also run targeted ads. And this is where people think, "Oh, I just spend a hundred bucks, and I get a thousand dollars in revenue," and target ads have gone way down in usefulness in the last decade.
SH (00:11:02):
And so really what targeted ads can do is help you learn about your audience. And I would look at things like Facebook, where you can do things like, "Oh, I want middle-aged women who have a degree in whatever. And we can run ads against them versus another segment of the population."
SH (00:11:18):
And you start to learn, "Oh, this segment engages with these ads more than this. And...you might have an idea of who's listening to your podcast. But when you run some of these targeted ads, you start figuring out, "Oh, these people actually care more than I thought," and you can start generating content more for them.
SH (00:11:35):
And then that's when you start ramping up like, "Oh, let's send them more things that they're engaged in." So there's that side of it...You could do an email newsletter. I'm actually pretty bad about doing those, but I know that those -
EW (00:11:46):
We have one.
SH (00:11:47):
You do. Okay. I'm personally bad. I should sign up for that. You're pointing out my ignorance here.
CW (00:11:53):
Well, we don't talk about it on the show. We...really haven't talked about it on the show much.
EW (00:11:57):
And we only recently got it to be more than basically an RSS feed. So now it tells you the show notes. It...gives you the transcript. It pulls out a quote. We hint what the next show will be. And if there is anything else, bonus content, like if my class is open or whatever, or we have a giveaway going on, that's all in the newsletter now.
SH (00:12:20):
Oh, perfect. So now that we've talked about it on the show, I would say it's a great time to connect those things, right? Use the newsletter to drive people to the show. Use the show to drive people to the newsletter. The more you can connect those things, the more people see how well your product is, I would say supported.
SH (00:12:38):
Or at least it gives the illusion of your product being supported, right? I know that anytime you're working on something, it always feels like you're scrambling to throw it together. At least that's how I feel as an engineer. But having content around it really helps if you want to engage with the newsletter rather than the podcast.
SH (00:12:52):
You don't want to duplicate those efforts exactly, or make the newsletter just a transcript, but you can link to the transcript, pull out quotes, like you're saying, offer a picture, keep people engaged in that readable format. That's a great way to keep people coming back and engaging with your work.
SH (00:13:09):
So...I know I throw out a ton of ideas here, and it's really trying to figure out what works for you, what feels natural. And the other thing is maintaining brand authenticity. As soon as people think that you're trying to push something on them, like a used car salesperson -
CW (00:13:23):
Yeah.
SH (00:13:23):
- they're turned off immediately, especially engineers, is what I have found. As soon as they feel that pushiness, they're gone, right? They're out.
CW (00:13:30):
I was going to ask, that was my next question is, I have this notion that engineers consider themselves at least resistant to all marketing. They'll put up ad blockers. They will pretend that none of this has any effect on them. Is that actually true, or is it just that it requires a different approach?
SH (00:13:50):
I think it requires a different approach. A lot of engineers I know subscribe to newsletters, right? And if you ask them if that's marketing, what are they going to say?
CW (00:13:59):
No, it's a newsletter.
SH (00:14:01):
Right, right. But it's absolutely marketing. It's a form of engaging with an audience, and ideally the newsletter provides something, right? Good marketing should give something, right? It's what's relevant in the industry, or, "Here's some good resources."
SH (00:14:17):
A good newsletter knows how to write things for the audience that they engage in. And if you're using HubSpot, or Mailchimp, or one of those, you'll see click-through rates, how long people read it. You'll get those stats back to see how long people are staying on that newsletter.
SH (00:14:33):
And that really helps. That helps you craft it to maintain better readership. And so if I feel like the newsletter's doing something for me, and I read it, yeah. I stay subscribed. Absolutely. But I am resistant to ads as well.
SH (00:14:47):
So I think we consider marketing to be that general, "Oh, big billboards, the fluffy thing, the woman throwing the hammer at the TV in the Apple commercial." And those all express great things. And if you're talking about a why for a brand, they help convey that.
SH (00:15:03):
But I'm with you. A lot of engineers are like, "Yeah, that's just fluff. I don't care about that." But if you know how your audience thinks, say it's mostly engineers, offering technical things or technical advice in their fields, they're going to be super engaged with that. How do you target that individually? I don't know.
SH (00:15:23):
I can't tell you that one. But news, or relevant things, or a cool interview, they're probably going to engage with that. So knowing your audience is the most important thing. Knowing your audience and offering something, right? Do that 51% of, "I give you something."
SH (00:15:38):
And then later on you can be like, "Well, hey, if you haven't heard our show, come listen," assuming your primary goal is to get listeners.
EW (00:15:45):
Yes. Our primary goal is to get listeners. And that's kind of odd, because what you're saying, give them information. Well, that's what we want to do. We just want them to listen. And so that's a little odd.
EW (00:16:01):
I know that a lot of people are using podcasts as a way to nurture a community, to create a community for later conversion to sales...It's part of the funnel.
EW (00:16:15):
But for us, the bottom of the funnel is, "Listen to the podcast. Stay an engineer. Be enlightened about diversity. And think about tools, and enjoy technology, and engineering, and science, and learning." And that's all we want.
CW (00:16:31):
I don't even enjoy technology.
EW (00:16:34):
I know. If you can't be a good example, you can at least be a horrible warning.
SH (00:16:42):
And there's your quote for the show.
EW (00:16:44):
So yeah, what you're saying makes sense, except we're not selling what we're selling.
CW (00:16:50):
Except it sounds like a lot of work, Shawn. It really -
EW (00:16:53):
It really, really does.
CW (00:16:54):
That's the other problem with being an engineer is, I don't really do want to do anything.
EW (00:16:58):
I mean, we already do a lot of work for the show, transcripts and tweets. And -
CW (00:17:03):
So is this the sort of thing I could just buy?
SH (00:17:06):
Yes, actually. There are marketing services out there. How good they are -
CW (00:17:11):
Yeah.
SH (00:17:11):
- is going be up to what you're trying to sell, how well they know the audience. Because sometimes they're just going to be like, "Well, what you need is a logo, and all these things." And you're like, "But I really don't think my audience cares about those things."
EW (00:17:22):
We have a bitchin' logo.
CW (00:17:22):
Bitchin'?
SH (00:17:23):
Oh, your logo is awesome.
CW (00:17:24):
Wow.
SH (00:17:26):
I love your logo.
CW (00:17:27):
What year is it? I had a question, well, about podcasts specifically, and maybe you have a thought about this, having had your own podcast. The way I relate to podcasts, after I've been listening to one for a while..., I don't really care what they're saying a lot of the time.
CW (00:17:46):
I'm just enjoying the conversation and the people I've had in my ears for a couple of months or whatever. And if that's the way most people relate to podcasts, that seems like a very difficult thing to market, to say, "Hey, join us, and you can pretend we're your friends."
EW (00:18:05):
That's because I wanted to target ads to podcast listeners first and engineers second.
CW (00:18:10):
Yeah.
EW (00:18:11):
And it sounded like those weren't categories people had.
SH (00:18:15):
Well, maybe you could run two different ads, buy out smaller slices, spend half on each, and see what performs better. Half of marketing is science.
CW (00:18:25):
And the other half is what?
SH (00:18:27):
Making stuff up and seeing what sticks.
EW (00:18:29):
Yeah. For podcast listenership going down, people aren't commuting anymore. That seemed to be when they listened to podcasts.
CW (00:18:38):
Yeah we had a real cliff March 2020 that was immediate and obvious. And...it crawled back, but it's definitely less than it was. And maybe we're annoying. I mean, that's just possible we're annoying.
EW (00:18:48):
I mean, it's very possible we've just become very annoying.
SH (00:18:51):
I highly doubt it. I also think that when you consider other podcasts, your big name podcasters, engineering is somewhat of a small slice.
CW (00:19:04):
What?
SH (00:19:07):
What? So it's kind of one of those setting the expectations of like, "Okay, who do we want to join this?" And if it's really for engineers, understand -
CW (00:19:14):
Yeah.
SH (00:19:14):
- it's not going to be Joe Rogan, right? He's a terrible example, but you know what I'm saying?
CW (00:19:20):
Thankfully not going to be. Yes.
SH (00:19:21):
Oh, yeah. Goodness. Yeah, really don't. Yeah. But as far as popularity goes, he's a name that -
CW (00:19:28):
Right.
SH (00:19:28):
He's a household name that everyone knows. So set your expectations. I always say go after the niche market, because it makes a lot of sense where you can foster that community. And also the idea of creating a community, because right now it's only us talking.
SH (00:19:44):
But I know that Chris Gammell invites people to join his forum that he runs, Contextual Electronics, which is a great way to foster kind of that, it's a lot of freelancers and hardcore engineers asking each other questions. And it's a good community.
SH (00:20:01):
So maybe a Discord channel or something where it's not just, "Oh, hey, listen to the podcast. But "Hey, listen to the podcast, and hang out with us on Discord," or whatever it might be so people can feel more engaged. Once again, this is one idea about creating community engagement. I make no promises as to how well that will work.
EW (00:20:19):
It works pretty well. Because we do have one, sorry.
SH (00:20:24):
No, you're good. Pitch it. Let's go.
EW (00:20:26):
It's for Patreon supporters. So you give us a buck once and you get the Slack link. And it's not like I ever check to see if you ever give us money in the future. And the Patreon group is pretty fantastic. People bring in technical problems and we discuss them.
CW (00:20:45):
And jobs and stuff.
EW (00:20:46):
And last week we had the podcast about scheduling and that all came out of a discussion -
CW (00:20:51):
A conversation there. Yeah.
EW (00:20:53):
- from the Patreon slack group. It's a really interesting group. I mean, I bring my problems there too.
CW (00:21:01):
I bring my complaints there.
EW (00:21:02):
Well, yes. We have a channel called "complainaratorium" and it's just a list of what people don't like today.
CW (00:21:08):
Yeah, so if you don't want to join Twitter, and you just want to complain about stuff, you can join our Slack for a dollar and then complain all you want to some other people.
EW (00:21:17):
We also have a "good stuff" channel. That one's good too.
SH (00:21:20):
So that's a more traditional marketing funnel where...you have a sales. You have a conversion. It could be a dollar.
EW (00:21:27):
Yeah.
SH (00:21:28):
But your community is actually locked behind that. And you'll have to forgive my ignorance about not knowing about that one. But your community's locked behind whatever paywall, right? It could be a very simple paywall. So it's more of a traditional funnel rather than having both side by side.
CW (00:21:44):
And...I mean, we're not trying to make a lot of money off of that. It's mostly a, "Here is a small hurdle that you must go over to join this community so that we don't have -
EW (00:21:53):
To be serious about it.
CW (00:21:53):
Yeah.
EW (00:21:54):
And because monitoring it takes time.
CW (00:21:57):
Yeah.
EW (00:21:58):
And making sure everybody stays nice, which they do. But it isn't free and open, because we do want people who are at a little serious about wanting to be in the community instead of just signing up 10,000 people and having them never show up.
CW (00:22:15):
So it sounds like we're doing mostly the right things.
SH (00:22:17):
Yeah.
CW (00:22:18):
Just do a little bit more.
SH (00:22:20):
And...you could open it up for free if you wanted to. But in my experience, if you want to kind of keep the trolls and whatever out, only have people who are serious, absolutely charge a buck. I mean, that barrier is not a lot to people who want to contribute and be serious about being part of the community.
SH (00:22:42):
And I find that it adds value to it, right? It pays a little bit for your time, but for the consumer, it feels like, "Oh, I'm now committed to be being a part of this, because I put in something that means something to me." It's a dollar, whatever,
EW (00:22:55):
Yeah. And it pays for transcripts, and it pays for the social media help we get. And then that leads to the newsletter looking nice. And so we don't make any money off of it.
CW (00:23:07):
No. Because we've got to pay for hosting out of that too.
EW (00:23:09):
Yeah. And mics, which we didn't have to give Shawn, because he already -
CW (00:23:12):
To shift away from the podcast, if that's okay.
EW (00:23:14):
Yeah. Oh yeah.
CW (00:23:15):
So this all makes sense when you've got some recognition out there. How do you build from zero? If I've got a little product I want to make or some content that that's brand new, and I'm an unknown person, what is the path there?
EW (00:23:37):
You go on a podcast.
SH (00:23:38):
Yeah. Absolutely. Putting yourself out there is the first way to do it...Harris and I would tell people on the show, we've had a few people on who had the exact same problem, right? "I'm building this thing. I don't know the first thing about trying to sell it."
SH (00:23:55):
And the first thing we tell them is, "Start telling people what you're building as much as you're comfortable sharing," because we get it. Sometimes you don't want to give the big reveal. You don't want to give away too much. But if you can have parts of it be open source, start posting on a project site like Hackster or Hackaday.
SH (00:24:13):
...Create a social media account right now. I find personally that Twitter is best for the engineering types in my experience, if that's going to be your audience. If yours is a photography or an art thing, maybe Instagram or TikTok.
SH (00:24:28):
Know your audience first, find a social media, and try to snap a picture of what you're working on once a day, the parts of it that you can share. Make a tweet. Does it suck? Yes, absolutely. I hate it. I hate doing it, but it's kind of part of the job.
SH (00:24:45):
And I have some fun, great conversations with people, but social media to me is a job, right? It's just one more thing I have to do to build an audience. And it takes time. I can't stress how long it takes to build some of these.
SH (00:24:58):
Maybe you'll get viral and lucky, but the reality is 99% of us, if we're just starting kind of an account, a business, it's going to take time to grow it. You can do some tricks. You can buy followers. You can buy email lists. But to me -
CW (00:25:11):
Does the buying follower stuff actually work? It feels kind of scammy.
EW (00:25:14):
...Yeah. Gross.
SH (00:25:15):
It's very scammy.
CW (00:25:16):
Okay.
SH (00:25:16):
It's very scammy. It will erode trust, right? It's a quick fix for a bandaid if you want to look important...for a very short amount of time. Once people find out about it, especially engineers -
CW (00:25:29):
Yes.
SH (00:25:29):
- it comes across as that scamminess. So personally I would avoid it. I prefer to grow stuff naturally in the sense of, people should choose to follow me, right? I shouldn't have to buy that.
EW (00:25:41):
Opt in. It's always about opt in.
CW (00:25:42):
And really when you're buying that, when you're buying followers, you're buying the little number so that other people see that -
SH (00:25:47):
Yeah.
CW (00:25:47):
- ...because the followers you're buying are probably not -
EW (00:25:51):
Real?
CW (00:25:51):
- doing anything for you. Right?
SH (00:25:53):
Exactly. Yeah. They're not. They're just so you have that number so you look legitimate or important. And any little tweaks you can do to look legitimate or important, right? Have a fun logo with an old-timey radio. That helps a ton. And look kind of like a business. If you want to be a business, get that LLC rolling.
SH (00:26:13):
Those kinds of things can help make you look legitimate. And start talking about what you're doing. Start engaging with people, other followers you can find, the hashtags you want to go after. And just start commenting on conversations that you see, and that will help grow your audience over time, at least still on social media.
SH (00:26:30):
And make sure you have some type of call to action on whatever page. Say you have your social media profile, say like, "Oh, I'm building this thing at blah, blah, blah, blah.com," or, "Check out my LinkedIn thing," or drive them somewhere.
SH (00:26:45):
Or like, "Hey, you want to keep up with what's going on? Sign up for my newsletter." And...whether people click on it or not, it starts to show that, hey, you're serious about building this thing. And over time, and I mean years, people start to pay attention.
SH (00:26:59):
...If somebody came to me, and they're like, "We need 10,000 followers in a month," and I'm like, "You can buy them. That's about your only option," or you throw a lot of money at some famous person to do influencer marketing to talk about you on whatever podcast or whatever it might be. That's about it.
SH (00:27:17):
Otherwise you're going to events to show it off. You're trying to get on podcasts. You're asking people, you're trading content. You're on social media.
SH (00:27:25):
You're trying to build a site that looks good with a newsletter. You're trying to do all of these things. And yeah, if you're alone, you're trying to do this while building a product, which is rough. I get it.
EW (00:27:36):
We have no call to action anywhere.
CW (00:27:38):
What do you want to do?
EW (00:27:39):
I mean, we're engineers. We're like, "Oh, you -"
CW (00:27:41):
Don't email me for sure.
EW (00:27:42):
We have a show.
CW (00:27:43):
Don't email me.
EW (00:27:44):
We have a show. We think it's really cool." And that's pretty much what everything says. We don't say, "And you should listen."...That's implied.
CW (00:27:53):
No.
EW (00:27:53):
Our listeners are smart.
CW (00:27:54):
I don't want people to listen. I want them to subscribe, but I don't want people to listen to me. That would be awful. Why would I want that for anyone?
SH (00:28:01):
You just want the number.
EW (00:28:01):
But no, it's because -
CW (00:28:03):
Yeah, I know.
EW (00:28:03):
- the people are smart. Of course they should know that they should listen.
CW (00:28:08):
I also have a notion that video is the thing now. And -
EW (00:28:12):
That's the hardest thing.
CW (00:28:13):
That's the thing that people engage with most. It's the true of Instagram, and YouTube, and all these things. They're all shifting to video and people don't respond to text. They respond a little better to images, but really they respond to video. And I was kind of learning about this when I was doing my band's record.
CW (00:28:29):
And it's hard to do video. It's 10 to 20 times more difficult than building anything else. So I'm resistant to that. I'm also resistant to people seeing me. So is that true?...Do people seem to engage with video more, and is that something people should consider more?
EW (00:28:46):
And can we use puppets?
SH (00:28:48):
Yeah. You can make video however you want. I love puppets. I don't know. Because I don't have a good apples to apples comparison between the two. I think it's different media for different audiences.
CW (00:29:00):
Yeah.
SH (00:29:01):
You can't do video in a commute, right?
CW (00:29:04):
Right.
SH (00:29:04):
Unless you want to risk a crash, but once commuting comes back, I'm sure there will be an uptick in podcasts, just because you can't do video. TikTok is going outrageous right now because, yeah. Video is hotness, and it's appealing to a certain crowd.
SH (00:29:22):
It's appealing to that younger crowd who want to make videos, share, do the music thing. But I've not really found what works for me in TikTok. I've seen some people, engineers, who have done kind of educational things, but they're very brief, one-minute things. And...they either look like garbage, or they're very well produced.
SH (00:29:41):
And...I know that took them a lot of time to produce this one-minute video for TikTok, and that's it, right? It's social media. It's ephemeral, right? It's gone.
EW (00:29:51):
Yes.
SH (00:29:51):
It might exist on your profile, but it's why I don't like social media. It's not evergreen. I want people to search for my content, right? If you type FPGA into Google, right, I want people to land on my video every time, not it's gone in social media.
CW (00:30:03):
Yeah.
SH (00:30:03):
So that's why I don't want to spend a ton of time to make a video that hopefully goes viral, and a hundred thousand people see it, and then tomorrow they're on to the next viral video. I don't care about that.
CW (00:30:13):
And you've got to keep feeding that, right? It's not -
SH (00:30:15):
Constantly.
CW (00:30:16):
You've got to keep feeding that.
SH (00:30:17):
You do. But it works for some people. It absolutely works. I'm not saying it doesn't work. I just hate it.
CW (00:30:24):
Good.
EW (00:30:26):
Yes. I mean, I didn't understand how much work social media was if you weren't just using it as a way to -
CW (00:30:34):
Complain?
EW (00:30:35):
Deal with excess thoughts.
CW (00:30:37):
Sorry. That's what I used it for.
EW (00:30:39):
Well, yes. But I usually just, the things that I didn't know who to say them to, I would just spew them to Twitter, and -
CW (00:30:47):
Right.
EW (00:30:47):
- get randomness out there.
CW (00:30:49):
And hope they didn't know they were directed at them.
EW (00:30:52):
It's all subtweets. This is hard. I mean the whole, "you have to spend a lot of work on it." It's hard to hear as an engineer, because I want to spend all my work building things.
SH (00:31:05):
The cool stuff.
EW (00:31:07):
Yeah.
SH (00:31:08):
Yeah.
EW (00:31:08):
So are there people who like marketing?
CW (00:31:12):
Yes. They're called marketers.
SH (00:31:16):
Hold it. I'm dying. Oh my God. Yes. Yeah. You can get marketing degree. You can do marketing. My experience is marketers will sometimes struggle with understanding the engineering audience. It is a very different audience than what a lot of marketers may have experience with, if that makes sense.
SH (00:31:41):
Like we talked about earlier, a lot of engineers I know are the first to run ad blockers, because they don't want marketing. But you can absolutely market to engineers if you know what kind of content they consume, right? They want to listen to Embedded.fm.
SH (00:31:56):
They want to watch the how-to tutorials, or things like Hacksmith, or probably Physics Girl, or some others who are fantastic YouTubers. And they're educational too. They teach you these technical things about the world. Engineers engage with that kind of content.
SH (00:32:14):
So knowing that, you can craft content targeted at engineers. And that's what I've been doing for a while. Mine's more educational than it is entertaining. But entertaining still works. You just have to speak the language.
SH (00:32:27):
And finding somebody who knows both the marketing side, how to do all of those things in marketing, and how to talk to engineers, that's that's not usual.
EW (00:32:38):
Do you like to learn things? Do you like puzzles? Do you really enjoy blowing crap up? Please come listen to Embedded.fm.
SH (00:32:47):
I'd watch that in a heartbeat, or listen.
CW (00:32:48):
Blow stuff up...What you said...about some marketers not knowing about how to market to engineers, I kind of got a real good view of that when I was at Fitbit, because sometimes marketing would turn their view inward. And at all-hands meetings they'd produce these nice videos, and they would be directed at engineering.
CW (00:33:14):
"Okay. Here's the product you just worked on. Here's how we're selling it. And here's this sizzle reel of all this stuff." And we'd all just be sitting there going, "Whoa. No, no, no."
EW (00:33:24):
"That's awful. Please wash it off."
CW (00:33:27):
"I'm cringing. Please make this stop." But it probably worked outside the company.
SH (00:33:32):
Yeah.
CW (00:33:32):
So yeah,...engineers are a difficult people.
EW (00:33:41):
Engineers are a difficult people.
SH (00:33:43):
Oh, yes.
EW (00:33:43):
That's nice.
SH (00:33:44):
Sounds like we're all speaking from experience here.
EW (00:33:47):
You mentioned content creation, which, I mean, I remember telling somebody that I was an engineer. And then I mentioned something about the blog, and then I mentioned something about the podcast. And they were like, "Wow, you just are a great content creator."
EW (00:34:02):
And I'm like, "No, I'm an engineer. I'm not a content creator. Don't say those words." But you embrace them. And so they must mean something different to you. What do they mean? What is it that non-me people think content creator means?
SH (00:34:21):
I'd love to know what they mean to you, but I will answer the question first. To me, content creation is any sort of media generally put out on the internet. Blog posts, social media posts, YouTube videos.
SH (00:34:38):
If I create a cat meme, that's content. If I write a 4,000 word essay about how this particular microcontroller is terrible, that's content. It's just the all-encompassing word for creating material to be on the internet and ideally viewed by others.
EW (00:34:57):
I think the difference is whether or not there's a secondary purpose to it.
SH (00:35:04):
What do you mean, secondary?
EW (00:35:06):
Marketing.
SH (00:35:08):
Oh, got it. So you're saying content creation as far as, I'm creating ads that mimic something helpful.
EW (00:35:17):
That mimic something helpful. That's it...I mean, you have ones that are actually useful, and then they're sponsored by various people.
EW (00:35:27):
But there's so many times where it looks like it's going to be useful, but it's just an ad in the end. And I hate that. And that was what I always thought content creators did, is they basically got paid to promote something in a sneaky way.
SH (00:35:43):
Got it. Okay. Thank you for helping me understand that connotation. Because I think there's terrible content, and what you're describing is terrible content. I think especially engineers will pick up on that B.S. very fast, right?
SH (00:35:55):
As soon as it's like, "Oh, do you have this problem? 25% of people have these problems, and there are ways to get over this." And...you read this long article, and it ends with, "If you want to learn how to get over this problem, then pay us $25 for this ebook." And you're just like, "I'm out. God, why did I waste five minutes of my life?"
EW (00:36:12):
Yes. Yes.
SH (00:36:14):
...This is a squares and rectangles kind of thing. Is that a form of content? Sure. It's written material that exists on the internet meant to be viewed by others. It fits the strict definition of content. Is it awful and subversive? Absolutely. It's the used car salesman.
SH (00:36:31):
It's kind of like...the used car salesman coming out to you and talking to you. Are they using language? Yes. Right? Okay. It fits into that technical definition, but it feels gross, right? You want to take a shower when you're done.
SH (00:36:45):
So I try to create content that is ultimately helpful. And my, what I call top of funnel marketing is, I want to create content that's educational. It helps people. And is there a secondary motivation for selling something? Yeah. But it's so far down the line. My goal first and foremost is to help you.
SH (00:37:08):
I really want to teach you embedded machine learning, right? Am I going to happen to use this tool called Edge Impulse? Well, yeah. It makes the job easier. Am I going to use Arduino? Yeah. It helps me teach this concept...
SH (00:37:21):
Once I started doing freelance stuff, I've had people, "Hey, we'll pay you. What's your rate to make a video for this product?" And I turn down a lot of work, because I don't want to just make a video selling their product, right?
SH (00:37:37):
I want to teach with the best tools that I think are the good teaching tools, which is one of the reasons I really like working with Digi-Key, is because they have a vast array of tools. So when I went to teach FPGAs, I got to look through their catalog and say, "You know what? I think this is the best tool," right?
SH (00:37:51):
They're not trying to push a particular product. But it happens to be sponsored by Digi-Key. And they own the content, and it lives on their channel. And they can gain followers that way.
SH (00:38:00):
So is there a motivation to gain followers so that people can eventually stumble upon Digi-Key ads for stuff? Yes, of course, but the stuff I create is first and foremost, trying to teach and help. It just happens to be done through a company rather than through a university.
EW (00:38:17):
Can you put the Edge Impulse TinyML on an ATmega328?
SH (00:38:23):
So what I like to tell people is you can run machine learning on anything. Any processor will run machine learning. It's just a question of, do you have enough space? Do you have enough RAM flash? And can the processor speed meet your timing requirements?
SH (00:38:37):
A lot of what we consider machine learning now is really neural networks and deep learning. And so will a 328 run a neural network? Probably. I mean, if you're talking like two nodes, yeah. Yeah. Oh yeah. Is it useful? No.
CW (00:38:52):
That's called linear regression at that point.
SH (00:38:54):
Yeah. Yeah, exactly. It's linear. Can you run linear regression with a, oh my goodness, what is it called? A non-continuous activation function.
CW (00:39:05):
Right.
SH (00:39:06):
Right. That's kind of all it is. Can you do that on the 328? Oh, yeah. Easy. Is it going to train it? Probably not. You can run inference. Is it useful? Also probably not. And it's going to probably be kind of slow. So Edge Impulse really targets your ARM Cortex-M3s, M4s, beyond. I've run it on a Cortex-M0.
SH (00:39:26):
So it's possible. You're just missing a lot of those really nice optimizations, like your DSP optimizations, that allow you to speed up inference.
EW (00:39:35):
So how did you learn about machine learning?
SH (00:39:38):
So this is an interesting one, because I actually have my master's in it, just, I didn't call it machine learning.
EW (00:39:46):
Your background is computer engineering with a master's in EE, -
SH (00:39:50):
Correct.
EW (00:39:50):
- and that's machine learning?
SH (00:39:52):
No. So what my master's thesis was on was using something called Hidden Markov Models -
EW (00:39:57):
Oh.
SH (00:39:58):
- to classify RF signals. And so -
CW (00:40:00):
Oh, neat.
SH (00:40:01):
Yeah, it was actually pretty slick. And this was in 2010, 2011, prior to our current blossoming of deep neural networks. It was around the same time that I think Google Brain was doing their stuff with collecting all the images to classify cats. And we didn't even call it machine learning.
SH (00:40:18):
My advisor at the time just said, "Hey, we're trying to do this thing where we're classifying RF signals. And we think Hidden Markov Models will do it." They were like, "Oh, don't worry about neural networks," right? We were at the end of that winter, that machine learning winter, we just didn't know it at the time.
SH (00:40:33):
So I learned how to train a Hidden Markov Model using...either the Viterbi or the Baum-Welsh. And I can't remember which one was trained. I think Viterbi's their forward propagation. This was 10 years ago, so you'll forgive my ignorance.
SH (00:40:45):
I'm sure somebody will be like, "Well ,actually," and tell us what the terms are. But we trained it to recognize different RF signals based on I think encodings. And it could look at your frequency spectrum and say, "Oh, we think that's QAM," Q-A-M, however you pronounce that, quadrature, whatever.
SH (00:41:03):
"We think this is single-sideband, blah, blah, blah, different types of modulation and coding." And it was four or five we could do. And while that was proven at the time, I was actually writing CUDA code for NVIDIA cards that would paralyze all this and running tests to see how fast we could speed this up.
SH (00:41:22):
And taking the algorithms and paralyzing the algorithms and figuring out, "Okay, can we take O(N^2) to O(N) or O(log N)," and making the algorithms faster that way. And it turns out just doing one or two, not so great.
SH (00:41:37):
But when you start talking like, "Oh, let's run multiple models against this," like a forest or a tree, much better. I'm sure there's plenty of libraries and other frameworks that allow you to speed them up even better now. CUDA has blossomed. You can run TensorFlow on top of CUDA, and it's fantastic.
SH (00:41:55):
So that's what I was playing with at the time. And the funny thing is, throughout my entire thesis, throughout my entire master's, not a single mention of machine learning was used. The term was just not used, because I was in the electrical and computer engineering department.
SH (00:42:08):
"That's not machine learning. We're just matching these patterns." It was pattern recognition. It just also happened to be machine learning, because we were training an automated algorithm that would update its properties to solve some problem. It met the definition. It just wasn't a neural network.
CW (00:42:25):
Even when I was reading about neural networks in college, I don't think the term machine learning existed -
EW (00:42:30):
No.
SH (00:42:30):
Right?
CW (00:42:30):
- for that. Yeah.
EW (00:42:31):
And I mean, I took an AI class in college, but it was AI winter...I mean, we talked about it, but we talked about it like history and maybe someday, but not -
CW (00:42:46):
Remember in college, the Apple II had just gone off for sale -
EW (00:42:51):
That's not true. No.
CW (00:42:51):
- by about three or four years.
EW (00:42:52):
No.
CW (00:42:52):
So there wasn't much going on in AI.
EW (00:42:56):
To think I liked you for your 386.
CW (00:42:59):
It was a 486.
SH (00:43:01):
[Ooh].
CW (00:43:03):
DX250, I'll have you know.
SH (00:43:06):
Did it have a turbo button?
CW (00:43:07):
Of course it did. They all had the turbo button. Yeah.
SH (00:43:10):
Yeah. Okay. Okay.
CW (00:43:12):
Got to have the turbo button.
EW (00:43:14):
How did you learn about FPGAs?
SH (00:43:17):
I knew them since school. I learned about them when we were doing computer architecture. It was sophomore year. And...we designed a 4-bit processor as a group project. It was, I think, based on MIPS, if I remember what we had to do, and this was 2003 or something at the time.
SH (00:43:36):
And we started learning about FPGAs like, "Oh, we could write it up in Verilog." And that was how we turned it in. And we could run it on a simulator, but our group actually was like, "Oh, can we borrow an FPGA board?"
SH (00:43:48):
Because we wanted the extra credit, because that's the kind of students we were. And we got it to work. We got our processor to work on an FPGA. But honestly after that we never really touched them in school. And I know that there are a lot of great engineers who specialize in them. They create hardware language for them.
SH (00:44:06):
That's what they do. And FPGAs have a certain niche, and they're very, very popular. And it wasn't until many, many, many years later that I'm sitting at SparkFun, and I've had this question a few times. People come up and they're like, "Oh, I've heard about these things called FPGAs. What can you do with them?"
SH (00:44:22):
And my super snarky answer that I never tell people, but for some reason I'm going to say on this podcast is, what would go through my head is, "If you have to ask, you probably don't need an FPGA."
CW (00:44:33):
Yes, that's what I say to neural networks most of the time.
SH (00:44:39):
Well, maybe I can change your mind about neural networks, but I still stand by that for FPGAs. Yeah. So what I realized is that people really want to know what they are. They want to tinker with them. And there's some good content out there, videos by people who know FPGAs a lot better than I do.
SH (00:44:58):
But I found that they were generally fairly disjointed in taking you from like, "I don't know anything," and they would cover the super basics of what a cell is, what a lookup table is, and they'd show you...not even really blinking a light, but let's do a simple adder or a simple state machine.
SH (00:45:16):
And then they would jump to just something massively advanced. And I'm like, "I feel like I still don't understand Verilog or VHDL from watching your video series."
SH (00:45:23):
I'm like, "Okay, I'll start kind of with what they had, and I'm going to build the building block so that you can go watch, or engage, or read books with the more advanced stuff."
SH (00:45:32):
And...that's where I decided to put together this FPGA series for Digi-Key. And yes, I had to reteach myself Verilog, because we probably touched VHDL, or I don't even remember what we did in school. I mean, I'm talking 20 years ago at this point.
SH (00:45:49):
So I had to reteach myself that. And that was the goal..., "Okay, I want make a series that actually helps people, or people who haven't seen them in a while, learn them."
SH (00:45:58):
Because at least for me, schools didn't do a good job of teaching FPGA. They taught computer architecture and the theories around that, but not, "How do you implement this well in hardware?"
EW (00:46:08):
Did you start making the series, start thinking about the series, and then have Digi-Key sponsor you? Or did Digi-Key say, "We want some videos, and we'll give you a pretty broad range of topics to choose?"
SH (00:46:22):
It was a collaboration effort. My time spent working with Digi-Key..., we would maintain this brainstorming list, and we'd meet every other week. And we'd kind of go through and be like, "Okay, what's on the horizon." We'd try to figure out what's coming up in the next couple of months.
SH (00:46:36):
And something that was on there for a long time, for more than a year was, "We should tackle FPGAs." And they're like, "Yes, yes, yes." And it was finally like, "Okay guys, I'm going to do this. I'm going to...reteach myself Verilog, grab an FPGA kit."
SH (00:46:49):
And I think that the tooling, the other big thing is the tooling was finally there to make it accessible in an open source, easy format -
EW (00:46:55):
Yeah.
SH (00:46:55):
- so you don't have to download this massive 8 gigabyte IDE with buttons everywhere. And you're like, "Oh man, this is proprietary. And I don't know where to start with this." And could I teach that? Yeah, absolutely. But...then I'm teaching Xilinx, not really general FPGAs.
SH (00:47:11):
So I found an open source tooling set that allowed me to teach, "Let's focus on just the Verilog so you can get to your hands around that. And once you...understand that, yeah. Jump to the more advanced tools if you want." But it was really also the tooling that was finally there. And it's still kind of beta-ish, but it worked.
EW (00:47:28):
This might be too delicate of a question, but did they pay you as an engineer or as a content creator, which isn't usually as much?
SH (00:47:38):
Oh, you're asking about levels. I won't talk about specific numbers.
EW (00:47:42):
Sure.
SH (00:47:43):
Because first of all, I was not an employee of Digi-Key. I was a contractor. And so part of my negotiated contract rates with them included things like, "I'm going to pay an editor to help me with this," right? So I had some built-in costs, so how do I translate that to saying -
CW (00:48:02):
Right.
SH (00:48:02):
- "What's my salary," right. And I could talk about my salary, but then it's like, "Well, my taxes look different and an LLC looks different than this. And I've got pre-tax purchases based on my deductible." It's really hard to do an apples to apples comparison.
SH (00:48:15):
The numbers that I would try to break it down to was I generally tried to maintain something of an engineering salary, because I considered myself an engineer first in some respects. Even though I like to say I'm both an educator and an engineer, I try not to lose that technical side.
SH (00:48:33):
Because I feel like a lot of people, once you kind of learn a technical side and you just create content around there, you stay within that realm rather than going, "Okay. I need to learn how to do an RTOS, because I haven't touched an RTOS since grad school, and that was 10 years ago."
SH (00:48:49):
"But the technical knowledge is still there. I reteach myself this," and then it's like, "Okay, I'm going to teach others about this." So could you have just an educator do that, who is a great teacher, but doesn't know the technical side?
SH (00:49:01):
That's going to be tough, and vice versa, right? Somebody who just knows the technical side, but can't teach very well, also difficult. So because of that specialty, I will usually try to negotiate something closer to engineering pay ranges.
EW (00:49:15):
How did you learn to teach?
SH (00:49:18):
Let me see. This probably started back when, second or third grade, when my teachers approached my parents and said, "Shawn is...a good student, but he struggles with public speaking." And sure enough, anxiety, sweats, I blanked.
SH (00:49:39):
I remember one talk that I had to give in fourth grade, third grade, or fourth grade. And they brought out the camera, which was the worst thing in the world. And I just couldn't remember what I was supposed to say. And I just kind of mumbled blankly into the camera.
SH (00:49:51):
And they played it in the library at my elementary school. It was like, "This is the most embarrassing thing I have ever seen in my life." And so since then my parents worked with me, and I'm super privileged and super lucky to have such great parents.
SH (00:50:07):
They worked with me to help me, right? Like, "Oh, I have to give a speech tomorrow for school." And this was a book report, right? Really basic stuff in fifth grade. So they would help me like, "Oh, let's do slides," because in 1995 slides weren't a thing. So my dad rented a slide projector.
SH (00:50:22):
He had one for his work, and we made slides, which was wild at the time. So he taught me the tech behind it. He also made me practice in the mirror and then practice in front of them.
SH (00:50:32):
And I would spend three nights doing this until I could recite it. And that helped build a confidence of it at an early age. And giving a talk in front of an audience, I still get nervous. I still get the butterflies in the stomach. And once I kind of find groove, I'm good. But the anxiety is still there around public speaking.
SH (00:50:52):
And since then, since early elementary school, I've always enjoyed doing the speaking. I've enjoyed practicing it. I would always be the first in class to get it out of the way, so I don't have to worry about it, but I'm like, "Yeah, I'm energized. Let's do this. Let's just nail it, get it done," so I'm not the one waiting until the end.
SH (00:51:09):
Because then the anxiety's worse when you have to wait to give a talk. So...it started really, really early because I was first terrible at public speaking and then got better, got better. I did Toastmasters when I was in Boulder. That just helped nail a few speaking things.
SH (00:51:24):
I still stumble with resets, is the thing I still have trouble with. So I usually edit those out of my own videos, and I'm guessing Christopher will probably edit those out of some of these too, but no big deal. Um, sometimes "ums" like I just did there when I'm trying to think, but I work on not doing as many of those.
SH (00:51:41):
In addition to public speaking, I also started teaching swing dancing when I got involved in it, so 2010 I started swing dancing. I want to say it was probably when I moved to Boulder. So 2013 I started teaching it, even beginners, and I fell in love with the routine of it.
SH (00:52:00):
Every Friday I'd show up and I'd teach with somebody basics of swing dancing. I loved teaching beginners. It would take them from nothing. A lot of these people did not have any dance experience. And my goal was within an hour, make them feel sort of comfortable doing something on the dance floor.
SH (00:52:17):
And the key there was teaching the same lesson over and over again. And we had some plants in the audience, people who didn't want to take the advanced lesson or wanted to learn how to teach. We put them in the audience and after the lesson we would take them aside and go, what did we do well? What can we improve?
SH (00:52:35):
And we would take that. And after doing this a dozen times, we had our beginner lesson down. And we got to the point where we had a hundred people in these beginner lessons, and it would take up half of a basketball court, because we were doing it in the CU Boulder gym. And we would just nail these lessons over and over again.
SH (00:52:54):
And...it was a lot of fun teaching. And I got involved in doing videos at the same time with SparkFun, doing my first video and working on getting more comfortable with camera work. Because if you watch those early videos, I am so nervous. You can probably see me sweating.
SH (00:53:08):
And I am talking a lot faster than I am now. And it was awful. I still go back and watch them and just [oof], nine years ago. I've improved on camera work, and it's just, yeah. Cameras were frightening to me.
EW (00:53:22):
I like how you had a problem, and it became a career. I mean, it's kind of amazing.
SH (00:53:33):
Yeah. What I find is a lot of people, that's a story I hear over and over again, right? I have a speech impediment, or I have a learning disability, or something like that.
SH (00:53:43):
And they work so hard on it that they become really good at whatever it is, or teaching it, or something, that it somehow becomes part of their career. It's an identity for them. And the same for me.
EW (00:53:55):
So back to machine learning, what is the best intro, or hello world, or blinking LED projects to introduce someone to machine learning on devices?
SH (00:54:08):
So when you say on devices, microcontrollers, embedded systems?
EW (00:54:12):
Yes.
SH (00:54:12):
I'm assuming it applies. Yeah.
EW (00:54:13):
Yes.
SH (00:54:13):
Embedded.fm. It's got to be, right? So the basic one, especially for Edge Impulse, is gesture recognition. And it's a good one...It's something physical you get to do, right? The idea of the magic wand. I wave a board, and make a gesture, and it goes, "Oh, you made this gesture."
SH (00:54:32):
You waved, or you did back and forth, or it was just sitting there, right? A little bit trivial in the sense of, "What am I going to use this for other than making a Harry Potter style magic wand," which is fun. I've seen some fun projects with that, but what problems does it solve outside of cosplay or art?
SH (00:54:51):
Not super sure, but it does demonstrate kind of your blinky. The one that I personally like to use is keyword spotting. How do you get your A-L-E-X-A device to recognize you when you say something? And I won't say it ,because I have one sitting right here. But that's a really fun demo.
SH (00:55:08):
I think that brings together a lot of embedded concepts, that keyword spotting kind of seems simple on the surface, right? I capture some audio, maybe do a little bit of DSP to pre-process that, to get some kind of features.
SH (00:55:23):
I send that off to a trained neural network, and it tells me if it heard the word that I trained it to listen for. And it's a cool project, because it does have applications beyond just waking up a smart speaker. I could yell at a machine to stop doing whatever it's doing as a extra safety mechanism.
SH (00:55:46):
I can tell my phone to do things. I can, let's see, give directions to a robot. There are lots of things. I think keyword spotting has a little more use beyond it, but it is more complicated, right?
SH (00:55:59):
You start having to talk about, "I need to meet timing requirements." And that becomes a fairly intermediate or advanced microcontroller or embedded system concept. Because now you're saying, "Okay, I need to fill a buffer. I'm sampling a microphone at 16 kilohertz or eight kilohertz."
SH (00:56:15):
That's probably as low as you want to go for voice. "And I need to fill a buffer." And when you think about it, that's a lot of data to fill a buffer with. And I have to be able to fill that buffer at the same time I'm performing DSP and inference.
SH (00:56:28):
So now you're talking, "Okay, now I need to run a real-time operating system, or I need to bring in things like hardware interrupts with direct memory access so that I can fill a buffer while inference is being performed on the main process," or maybe it's a dual core, like your ESP32.
SH (00:56:43):
I have one core worrying about filling the buffer, and it sends stuff off to another core that does the processing for you. That's why I really enjoy keyword spotting, but I wouldn't say it's blinky.
SH (00:56:54):
I would say it's non-trivial. And doing keyword spotting really helped me grasp the embedded implications and how difficult it is to do on something like a microcontroller.
EW (00:57:06):
That's interesting. I mean, the throughput problem is always one that kind of gets people a little mixed up on, because the idea of continuous input, and output, and yet windowing, and overlapping windows, it is not as easy a problem as it looks.
SH (00:57:25):
Absolutely. Yeah. And in this case,...you generally have to use a sliding window to capture that data.
EW (00:57:31):
I have some listener questions for you.
SH (00:57:33):
Alright.
EW (00:57:35):
Well, actually the first one is not a question, but a comment that your Hello Blink Show, which was your podcast, was fantastic. Why did you stop doing it?
SH (00:57:48):
I really appreciate it. I love to hear that people liked the show and that they found it useful and/or interesting. It was kind of a career decision for both myself as well as Harris. We did it for about a year, and we had a lot of fun. We met some really cool people doing it, and we're really happy that people are still enjoying it to this day, right?
SH (00:58:08):
We've got some downloads that are still happening on the site. We have it up, so feel free to check it out.
SH (00:58:13):
And it really was a focus on trying to help engineers, or people wanting to make a product, or software people, understand a little bit about the marketing and sales side, especially people who want to branch out on their own, quit their jobs, and create a product, or a service, or whatever it is, and market themselves, or market their product, and just to give them an idea of what that looks like.
SH (00:58:35):
And we chatted with a lot of cool people who had done it or were struggling with it, and they could share their stories. And it was even more niche. We talked about how niche engineering is. That was even more niche in the engineering world. We didn't see a lot of growth.
SH (00:58:48):
We have a few loyal fans and people who really liked it, which we were so happy to have. And we met great people on the show, but really it came down to me wanting to move away from just being kind of this marketing strategy advisor.
SH (00:59:03):
That's kind of where I thought my career was going, what was that, a year and a half ago, two years ago...I wanted to be kept on as a retainer to do kind of marketing training for how companies can create content and market to engineers.
SH (00:59:22):
It would probably pay very well, and I'm sure there's a need for it. I just really didn't want to get away from the technical side of things...Anytime it comes up in my own career, it's like, "Hey, do you want to go into management?" I usually turn that down, because it removes me from that technical side.
SH (00:59:38):
So I just figure out how to navigate my career to grow in my career and avoid being in management. That's been my MO since I started working in 2006 or something. But it was kind of a pivot for me. And it was also a pivot for Harris when we chatted, and we decided to end the show.
SH (00:59:55):
He wanted to move away from trying to just gather these one-off clients and work more closely in a capacity that was helping people with the sales cycle for CRM...Did the podcast have a dual purpose? Did it help both of our careers? Absolutely, right? We get back to that conversation of creating content to help us.
SH (01:00:22):
And for both it was kind of a funnel to meet people, to have content out there that would introduce us as the subject matter experts in sales and marketing, and they would hire us to do these things. And what we saw them hiring us for kind of pivoted.
SH (01:00:38):
And we didn't see the show fitting into that picture anymore. And it allowed us to pull back to focus on more of the things we wanted from those that pivot in our careers, if that makes sense.
EW (01:00:50):
Okay. A question from John Schuch. "The breadth of your tutorials is kind of amazing. How do you develop that scope of expertise?"
SH (01:00:59):
Oh, I really appreciate it. I have to say that I don't have a lot of in-depth expertise. I'm pretty wide in what I know as far as embedded systems go, except for maybe that DSP machine learning stuff. I've been diving more into that recently.
SH (01:01:14):
I'm just about done with that hands-on machine learning book, which has been really fascinating, but it is drinking from a fire hose. It's doing a lot of stuff in Python. But other than that, when it comes to FPGAs, other than what you saw in the YouTube, I haven't tinkered with FPGA since college.
SH (01:01:31):
It was really just learn enough to get to a few basic demos. And then I have kind of that knowledge now. I've created content. I can always go back and reference that content. But that's it, right? If somebody were to ask me questions about FPGAs outside of what I showed in those videos, I would be very ignorant of it.
SH (01:01:50):
It's really, I kind of have a beginner maybe touching into intermediate levels of knowledge around embedded concepts, across a variety of things. I have a friend who's a FPGA developer, and I asked him about, "How do I doing the double buffering to avoid metastability?" And he helped me understand that.
SH (01:02:10):
So I lean on some people to help me understand the more advanced stuff that's out there and kind of same for RTOS, except I did tinker with RTOSs outside of that class.
SH (01:02:20):
Because you kind of need to, now with Arduino, if you're using some of the advanced boards. And they abstract it a little away, but I'm like, "Oh, I can run multiple threads. Let's figure out how to do that." But some of the advanced stuff in RTOSs, I don't know anything beyond what's taught in the course, if that makes sense.
SH (01:02:35):
So it's kind of broad for the sake of teaching it and creating content, but not super deep. I have not a lot of knowledge around things like software engineering practices. So I kind of stumble my way through continuous integration.
SH (01:02:50):
If I have to do a pull request for something on GitHub, I'm like, "How does Travis CI work? I don't know. I need those little check boxes to show up. Hopefully my code's good enough." And that's kind of all I know. How to set it up? No clue.
EW (01:03:05):
And a question from Twitter, from aray. Do you have any predictions for the next eight years for tinyML or embedded machine learning?
SH (01:03:15):
I wish I could say I had these amazing predictions about what the world's going to look like, where the trends I see going. Voice is huge. And that goes a little bit beyond tinyML. Right now we're relying on keyword spotting to do some of that for us.
SH (01:03:32):
But the ability to talk to our computer I think is really, really undertapped at the moment. And I know that people are working on it. We have the smart speakers. We have Siri. We know what those things look like, but I think having more natural conversations with computing devices is going to be massive in the future.
SH (01:03:49):
And a lot of that will require larger machine learning beyond just your embedded side. So on the embedded side, there are some areas of growth that I see. Vision is a big one. Not just doing self-driving cars, whether you consider that embedded or not, but things like being able to detect if somebody's in frame.
SH (01:04:12):
So we have a camera. Your doorbell cameras, a lot of them can do person detection now. A lot of that's probably being done on the device itself. I don't think mine connects to the internet. And can it see a person? Yeah. It knows when there's a person at my door. And a lot of that's tinyML.
SH (01:04:28):
You can also do things like smart buildings where you can identify people. I'm sure y'all have experience with those really awful PIR sensors that they try to replace all the light switches with. Yeah. You're laughing. You know exactly what I'm talking about.
SH (01:04:40):
And...you're trying to work late at night in the office at 6:00 PM and the lights keep turning out. You have to get out of your chair, and wave, and jump around, because those PIR sensors are so bad.
SH (01:04:50):
So something like a camera, while it might require more power, perhaps there's ways to just take a picture a second or a picture every minute, and say, "Hey, is there a person in this frame?" And it's been trained and does some tinyML to identify people.
SH (01:05:03):
And it says, "Oh yeah, leave the lights on," or "Make sure the building or this room is at a particular temperature." So smart buildings is a big one. The other big thing is predictive maintenance. That's big in industry. That's kind of getting into your IIoT, your industrial IoT applications, where it's pretty hidden from us as consumers, right?
SH (01:05:24):
We always think about IoT, the smart home. And we really haven't seen much of that outside of the smart speaker, right? That was the B.S. promise to us eight years ago, right? It's really industry that's seeing a benefit from sensors being on devices or these think machinery or robots.
SH (01:05:41):
And we want to know if something's going wrong so that...we don't have to have somebody there watching the machine the whole time. A few sensors, like a vibration sensor, can tell us if a bearing is going out. And yeah, you have to spend a few dollars more to put a sensor on it. And maybe you have to train it.
SH (01:05:57):
But the idea of saying, "Oh, this robot arm is about to break. You should probably go maintain that and replace. This bearing is about to go," or whatever it might be can save you thousands or millions of dollars and prevent big accidents because a bearing broke or a motor gave out.
SH (01:06:14):
And you can know before it actually happens. And that's a big area of study right now to figure that out. And...NASA has put out data sets to try to do this, because satellites, right? You put something up in space and you're like, "...Is it going to fail on us? We don't know," right? "Do we need to go repair it?"
SH (01:06:29):
"How much life do we have left on this giant camera or research station?" And it's being researched. And that's a perfect application for embedded machine learning for identifying when things are going to break, whether that's sound, whether that's motion, cameras, whatever it is. That's another big one. And that's a ways away.
SH (01:06:52):
And five years I suspect we will see, hopefully in two or three, we'll see more applications of that. But I like to think about what's going to happen in eight years, that five to ten range. Better interactions with humans and computers, that's going to get a lot slicker.
SH (01:07:07):
And things are going to start telling us when they're going to break before they actually do, maybe not your blender, but your car.
EW (01:07:14):
Yeah. I mean, that makes a lot of sense. The car has many of those sensors. It's time for those sensors to go into something smart instead of waiting to talk to the mechanic after it's broken.
SH (01:07:24):
Yeah. And...rather than just one light and an error code, it would be great to be like, "Oh, hey, there's a weird vibration on this thing," or, "There's a knocking," and it can tell you more about it rather than just, "Go to your mechanic."
EW (01:07:39):
And I can see the computer interface, because I have been thinking what I want is to be able to change to a different program with voice command, Word, Google Doc, something...that I don't have to move my mouse for that I could talk to.
CW (01:07:55):
I'm sure you can make that happen.
EW (01:07:56):
I probably could, although I don't know how to go to the app.
CW (01:08:00):
Yeah.
EW (01:08:00):
But I can see how computer interfaces are going to change, even in small ways like I want, but there'll be bigger ways too.
SH (01:08:09):
Yeah. In ways we can't imagine. That's the thing that I can't impress enough. If I could predict what it looks like in eight years, man, I'd be a lot wealthier than I am now. But I'm not really. I'm making engineering money. That's about it. But yeah, it's going to be ways we can't fathom, and that's where I think machine learning is so cool.
SH (01:08:27):
To me, it feels like those early days...You read about early days of computing and people hacking on these big mainframes to invent things that they weren't intended for.
SH (01:08:37):
And that's so cool with machine learning right now is people using them, the idea of like GANs and generating content or generating deep fakes. Dangerous, but really slick. That's a cool application.
EW (01:08:52):
And if people are interested in learning more about machine learning, you have some courses. You mentioned them earlier?
SH (01:08:59):
Yeah. So I've got two on Coursera. One is Introduction to Embedded Machine Learning, and the other is Computer Vision with Embedded Machine Learning, where we build on that first course. And we use the OpenMV Cam or the Raspberry Pi to train things like image classifiers or object detection systems.
EW (01:09:18):
Are they expensive?
SH (01:09:20):
Which things, the boards or the courses?
EW (01:09:22):
The courses.
SH (01:09:23):
No, they're actually free. You can optionally pay to have a certificate if you want to show it off on LinkedIn, but they are absolutely free for anyone to take.
EW (01:09:32):
There was a button here that said "financial aid available," and I was wondering.
SH (01:09:37):
Yeah, that's Coursera's thing. They put that out there. That's what they do. If there's any money involved..., whether it's a certificate or you have to pay for the course, they slap that on. I don't have much control of that.
EW (01:09:49):
And Shawn, it's been really great to talk to you. Do you have any thoughts you'd like to leave us with?
SH (01:09:54):
So the one career advice I can possibly give to people if they're thinking about things is the idea of finding "Blue Ocean." It's a good book, but you could probably get the idea of it. It's a book written by W. Chan Kim and Renée Mauborgne...It's really a business book. It's a business and marketing book.
SH (01:10:15):
And it's finding areas where there aren't a lot of people, there isn't a lot of activity, right? If somebody opens a McDonald's on your street, is your first thought like, "Oh, I should open a Burger King, because that McDonald's is doing so well."
SH (01:10:28):
And it's like, "Well, there's already a fast food restaurant there." That's now red ocean, in the sense that there's sharks in there. They're already getting at whatever's to be fed on. It's a terrible analogy, but yay marketers.
SH (01:10:39):
But...as long as you're still interested, and it's motivating to you, try to find areas where there's not a lot of movement right now. What has, they call it blue ocean, right? You're sailing. There's no sharks in the water and go out and find areas that are potential growth that don't have a lot of activity.
SH (01:11:00):
Don't open the gas station next to a gas station and expect to do as well as that first gas station, right? Because now there's two. It's a little bit of scarcity mentality, but the idea is finding areas that aren't. So I hope that helps, thinking about career advice. That's how I try to approach some things.
SH (01:11:14):
I happened to be a decent fit for machine learning stuff, because I had studied it, and I had a decent grasp of it. And I took Andrew Ng's course on Coursera, which was amazing.
EW (01:11:25):
It is amazing.
SH (01:11:26):
Oh, you took it. Oh, it's so good. I tell people to take that all the time as a first step into ML. And I fell in love with it. I really enjoy it, and doing it on embedded systems, at least at the time and a little bit right now, it's still fairly blue ocean, right?
SH (01:11:41):
There's not a lot of companies tackling this concept of throwing deep learning algorithms on microcontrollers, and it's weird and bizarre. But there's market for it. There's potential things to come from this.
EW (01:11:55):
You have to be careful. Well, no...It's all about the features, but that is an entirely different show, and we should not keep you any longer. Thank you so much.
EW (01:12:07):
Our guest has been Shawn Hymel, Senior Developer Relations Engineer at Edge Impulse and Technology Marketing Strategist, Advisor, and Content Creator. Links will be in the show notes, or you can find Shawn on YouTube, Twitter, and Instagram, as well as his website, shawnhymel.com.
CW (01:12:26):
Thanks, Shawn.
SH (01:12:28):
I really appreciate, and I'm honored you all invited me to be on the show. Thank you so much.
EW (01:12:34):
Thank you to Christopher for producing and co-hosting. Thank you to John Schuch for suggesting Shawn should be on the show, and to our Patreon listener Slack group for questions.
EW (01:12:45):
And of course thank you for listening. You can always contact us at show@embedded.fm or hit the contact link on embedded.fm, where you can also sign up for newsletters.
EW (01:12:56):
And now a quote to leave you with, from Bill Nye. "There's nothing I believe in more strongly than getting young people interested in science and engineering for a better tomorrow, for all humankind."