SDS 545: Scaling Data-Intensive Real-Time Applications — with Matthew Russell

This is episode number 545 with Matthew Russell, CEO of Strongest. 

Jon Krohn: 00:00:10

Welcome to the SuperDataScience podcast, the most listened-to podcast in the data science industry. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. I’m your host, Jon Krohn. Thanks for joining me today, and now let’s make the complex, simple. 

Jon Krohn: 00:00:41

Welcome back to the SuperDataScience Podcast. Today’s guest is the widely read, widely talented, and broadly experienced entrepreneur, Matthew Russell. Matthew is the founder and CEO of Strongest, the leading technology platform for global fitness events, which is growing into an application that uses machine learning models to make you stronger, faster, and fitter than ever before. Matthew is the author of four books published by O’Reilly, including the classic, Mining The Social Web, which is now in its Third Edition. 

Jon Krohn: 00:01:12

Prior to founding Strongest, Matthew served as CTO at several software companies. He holds a Bachelor of Science degree in Computer Science from the United States Air Force Academy, as well as a Master’s degree in Computer Science and Machine Learning from the US Air Force Institute of Technology. In this episode, Matthew details the tech stack he uses to make it possible to provide data from fitness competitions to millions of spectators all over the world in real-time, how he rapidly tests machine learning models for deployment into portable devices like the iPhone and the Apple Watch. He talks about multi-objective machine learning functions and why they’re so widely useful in real-world applications. The three critical traits he looks for in anyone he hires, the values instilled in him by pursuing and military education, and the key skills he wishes he had learned earlier in his career. Parts of today’s episode, particularly in the first half, we do get fairly technical as we dig into the open-source software stack that enables the scalable deployment of data-intensive real-time applications. That said, much of the episode will appeal to anyone who’s excited about physical fitness or commercializing AI. All right, you ready for another awesome episode? Let’s go. 

Jon Krohn: 00:02:30

Matthew, welcome to the SuperDataScience Podcast. It’s awesome to have you here. I’m so excited for this episode. Where in the world are you calling in from? 

Matthew Russell: 00:02:39

I’m calling in from Franklin, Tennessee. We’re just a couple of suburbs south of Nashville. 

Jon Krohn: 00:02:46

Nice. It must be wonderful this time of year. It’s midwinter for us in in the Northern Hemisphere, but I suspect that’s the kind of spot you get to enjoy a nice weather year round. 

Matthew Russell: 00:02:59

Indeed, yeah, I was out running shirtless last week. It’s unseasonably warm and it is a nice time of year here. 

Jon Krohn: 00:03:07

Super jealous. So, I know you through Austin Ogilvie, whose episode came out recently, it was episode 535. And Austin, when he found out that I didn’t know you was surprised, because apparently we have a lot of common interests. And now that we’ve kind of planned our conversation for this episode, I 100% agree, so I’m really glad that Austin made that connection. How do you know him? Have you known him for a while? 

Matthew Russell: 00:03:35

Yes, I’ve known Austin for many years now. So, Austin previously started a company called Yhat. I was an early investor in Yhat. And common interest there, just many things ranging from data science, which was essentially the premise of the company into philosophy and technology and many other things. So, we really hit it off, stayed in touch, and he’s now an investor in Strongest. So, we’ve linked together in some common business interests over that time. 

Jon Krohn: 00:04:09

Cool. Yeah, you’re both fascinating characters. That episode was absolutely fascinating for me to record. And yeah, I’m looking forward to this one you mentioned Strongest there. So, you’re the CEO and founder of Strongest, which you founded in 2018. And you’re currently capping off your pre-seed investments with well-known CrossFit brands and Austin Ogilvie, evidently, the serial data science entrepreneur, not a bad person to have as one of your investors. But yeah, so I know that the ink hasn’t dried at the time of recording on some of these CrossFit brands that are investing in your company, so we’re not going to mention them by name. But they are, if people are into the sport of fitness or into weightlifting, these are the most recognizable names that are backing your company, so that is super cool. As of today, you are the leading platform for fitness events. So, over 250,000 competitive athletes have used the Strongest platform for events, that for me and people I know, in the CrossFit community are household names. 

Jon Krohn: 00:05:16

So, things like the Rogue Invitational, which blended CrossFit and Strongman events. I know you mentioned, there was some technical complexity to that for having a platform like yours run for an event like that. You were involved in the CrossFit Open announcements, which involves hundreds of thousands of athletes around the world, paying at 20 bucks. And then they are in this giant pyramid that allows them to compete at the CrossFit games, ultimately, which happens once a year and the top few dozen male and female athletes compete in that giant competition. But yeah, anybody can go in at the open and your platform was what facilitated the open announcements. And yeah, I mean, other big events like Water Palooza is just, it’s a who’s who for you of the biggest competitive fitness events on the planet. So, 250,000 competitive athletes have already used your platform across 1500 fitness competitions. And recently, you’ve had 300% year over year of revenue growth. It’s absolutely fascinating. And I’m sure you’re going to want to talk about that, this fitness platform itself. But this live event platform is just the thin end of the which for something much bigger, so you’ll probably want to talk about that as well. So yeah, the floor is yours. 

Matthew Russell: 00:06:38

Yeah, so as you said, we currently are operating an amazing platform for running functional fitness events, so it’s pretty versatile in terms of providing a point of sale solution for competition organizers to collect payments, to score the events, to disseminate leaderboards, to manage schedules, which really gets complex. But yeah, that’s really a marketing platform in the grand scheme, because we are customer for the events platform, it’s really the competition organizer. Now, the interesting thing about a competition organizer is that they have an audience, and they bring that audience with them to the events. So for every organizer, you have orders of magnitude of more people that the organizer brings along. And yeah, we’ve recently launched our iOS and Watch iOS app, in the app store. You can search for it. It’s called Strongest and it delivers a hyper personalized experience for the athlete to help the athlete achieve more goals by understanding their goals, their data, the stimulus response relationships of the training program that they’re enrolled in. And, I’d love to tell you as much about that as you’d like to hear. 

Jon Krohn: 00:08:00

Yeah, super cool. I know that there are data science, data model AI involvements in that consumer facing app that is still growing right now. But let’s quickly first talk about the Live Events platform that you’ve already built and that is powering these thousands of different competitions out there. So, it’s a web app. I imagine that there’s huge amounts of data flows. You have all these different competitors. And people probably expect to see their results on the leaderboard and basically, real time, for example. So, how have you fleshed out that tech stack to be able to handle so much data in such a performant reliable way? 

Matthew Russell: 00:08:45

Yeah, it’s a great question. So, it is a problem that at a small scale doesn’t seem like a particularly challenging problem. Most problems in isolation that are not at scale, usually, aren’t always that challenging. But when you think about a situation in which you have like a group of really high profile athletes or a very large event with many athletes, and you start to think about the numbers of people who may want to keep up with the scores, so every person in the world may care about an event, now- 

Jon Krohn: 00:09:24

Right, right. It isn’t just the athletes. There’s, with an event like Water Palooza, the Rogue Invitational, there are millions of people who want to be seeing in real-time how the standards have changed as a result of the event that just finished. 

Matthew Russell: 00:09:39

That’s right. Yeah. You may have lots of people following particularly well-known athletes, even on the larger regional events with athletes most people may not have heard of by name. They all have partners and friends and people that are keeping up with them, so go figure if you have one organizer bringing hundreds of participants. Well, those hundreds of participants bring multiples along with them as well, so you do end up with a problem of just scaling a web app in a performant way. 

Jon Krohn: 00:10:15

Yeah, we had- 

Matthew Russell: 00:10:16

So, our stack is- 

Jon Krohn: 00:10:17

In Episode 533, we had Brett Tully. He was talking about having massive scale machine vision algorithms and he said for his team, their mantra is, “Everything breaks at scale.” So yeah, so I’m not surprised. Just to kind of echoing the point that you’re making here about things that work when you’re just doing them locally, on your own. They seldom scale up how you might think they should. So anyway, yeah, would love to hear, yeah, more about your stack and how you’re overcoming these challenges. 

Matthew Russell: 00:10:48

Well to that point, there are sort of, I think, a natural order of operations, right? You make it possible first. After you’ve made it possible then you can do things like make it performant, make it scale, make it beautiful, so there’s just a natural evolution. So, we started this company, really, as just a side project. There was a local fitness competition that needed a scoring system, just did this for fun. That particular event was very successful. People started to ask, “Hey, can we use your scoring system?” And I’m thinking, “Well, my scoring system? Oh, yeah, my scoring system, right? Yeah, that.” So, started to just sort of experience this natural local organic growth and of course, people start to need things. “Well, hey, can I capture payment with this? Can you support some other scoring policies?” 

Matthew Russell: 00:11:43

Well, over time, you end up with a scale problem sort of in two dimensions. You may run one large event that could produce a traffic spike on your platform or you may be running many smaller events simultaneously because the nature of the web traffic is such that you generally don’t have a fitness competition on a Wednesday morning. Most of your competitions are going to be Saturday mornings, Sunday mornings, it tends to be clustered. And when the traffic is high, the traffic is high. And when the traffic is low, the traffic is low. So in terms of just scaling that, the Cloud has been indispensable, just in the sense of building on a Cloud platform. If you’ve built your web app in a way that you’re managing state, so everything in software, on some level, is really just state management. Software is just like a web app. It’s just a state machine, if you want to get really abstract about it, like back to Computer Science 101, it’s just a state machine. And if you manage your state in your web app, according to best principles, you can generally scale horizontally, which is where Cloud computing comes in, right? You can respond to spikes in traffic. You can use some technology to help you cash more effectively.

Matthew Russell: 00:13:11

So for us, we’ve built on Heroku, as a platform. It’s been really great for us. Lots of good add-ons. We’ve automated all of that infrastructure management with Terraform, which is solving a really important problem, right? If every time you needed to spin up a new environment, you had to log into a UI and point and click and remember all of the boxes that were checked or unchecked, like that is my worst nightmare, right? Because who’s ever going to remember that? You’re always going to mess it up. Well, Terraform is interesting in that it provides infrastructure as code, which is just a fancy way of saying, “We’re going to declare our requirements in a nice readable format and we’re going to let the software robots take care of provisioning that and deploying that and making that completely cookie cutter reproducible for us.” 

Jon Krohn: 00:14:14

Nice, that makes a lot of sense. And so, by using a tool like Terraform to manage your infrastructure, it allows you to automatically scale your systems horizontally, which means adding more compute servers into the picture. And yeah, so I can see how that makes a huge amount of sense. You’re on a Saturday morning and all of a sudden, you’ve got a whole bunch of fitness competitions going on all over the world. And so, while on Friday, you only needed a couple of servers running, all of a sudden Saturday morning, Terraform somehow in interaction with Heroku can get a sense that there’s more demand than usual. And maybe, it can automatically scale up and add more servers on that are running the app separately and allow all of those competitions to have highly performant real time results. I probably butchered some of my, trying to recap back to what I said there, but you can let the audience know what I got wrong and yeah. 

Matthew Russell: 00:15:17

Well, yeah, that’s a good summary. I mean, I would think of it like this. The job of provisioning the full stack is what you delegate to Terraform, so that ultimately, you are literally running a script and you’re provisioning everything to get a complete replica of your production system, which can be handy for a variety of reasons. And then in general, whether it’s AWS or Google Cloud or Heroku, there’s some knobs you can turn and of course, you can preconfigure those with Terraform to give you some of that nice horizontal scaling in response to response times, in response to traffic loads, things of that nature. 

Jon Krohn: 00:16:03

Cool. Many thanks to the master of Data Science Program at the University of California Irvine for sponsoring today’s episode. The UC Irvine Master of Data Science Program blends statistics and computer science principles with partners from industry to empower students to innovate in the field of data science. Located on the tech coast of Southern California, students of the program will enjoy a powerhouse ecosystem where over a third of Fortune 500 companies are located. Take a giant leap in your data science career through the UC Irvine Master of Data Science Program. To learn more, head to www.superdatascience.com/UCI. That’s at www.superdatascience.com/UCI. Check it out.

Jon Krohn: 00:16:53

Much more eloquent and knowledgeable than my ability to summarize it, but it all made perfect sense to me. So, that’s super cool. So, I imagine the challenges with building and scaling a web app, like your live events platform is quite different from the kind of tech stack that you need to scale your iOS app. So, we mentioned earlier how this live events platform is kind of the thin and the marketing wedge to grow the strongest platform, which will be this tool that allows people, large amounts of people all over the world to be using an iOS app or their Apple watch. An integration right there with the Apple Health kit to allow them to progress towards their goals, to get guidance on how they can be achieving those goals. So I guess, it would be useful, for me and for the audience, to have a bit more detail on how that app works today. What your vision is for the future? And then yeah, how you make that happen? What the tech stack is like? 

Matthew Russell: 00:18:03

Yeah, so at the core, Strongest is a connected AI platform in the strength and conditioning space and as a maximum, I would say that its job is to fundamentally understand fitness and fitness data as a domain. It’s a bit like your own personal exercise physiologist at the core, that’s really the mission of the technology within the company. And so, when you start to think about that, the data model is paramount. It all starts with the data model. And we’ve approached building an app, I think, in a different way than most fitness apps have approached it. So, when I look at the app store, I see literally a wasteland of fitness apps, right? There are more fitness apps than you can even count, and so- 

Jon Krohn: 00:18:58

There are. 

Matthew Russell: 00:18:58

So, it’s a very crowded space, in that regard. Then you say, “Well, what do all of these fitness apps actually do or most of them anyway?” Well, most of them are really content delivery platforms of some kind, right? They’re giving you a description of some exercise or workouts or training. They’re saying, “Go do this.” In a level beyond that, you may get some videos of movements or movement standards and a level beyond that, you may get a full video of an entire workout that you would follow along with. So, it’s very much a content delivery platform and many successful ones. What we’ve done is sort of, we’ve sort of played to my strengths as a technologist and data scientist and we’ve approached this from really the other spectrum, which is let’s start with a deep understanding of the data, right? So, one of my principles for data science is like “Know thy data and know thyself.” And so, if you’re going to build software that has to really know the data, you’ve really got to know the data and you got to know yourself because you’ve got to know your own biases in this as the creator of the technology. And you’ve just got to at least be cognizant of that and where some of the sharp corners can be, as you scale this thing. 

Matthew Russell: 00:20:29

So, starting with the data model, it’s like, “Well, how do you even wrap your head around that?” Right? This took some time to think through. Well, in one of my previous roles, I spent about a decade at a company, really amazing experience. We were building natural language processing technology in the earliest days, I mean, we were literally taking research papers out of Stanford, prototyping them in my lab, and then productionizing this for government organizations, banks, healthcare companies. And so, I developed a particular set of skills in natural language processing. And when you start to think about the content in a fitness domain, there’s a lot of language, right? We have a lot of humans communicating and it’s a lot like a recipe, right? It’s semi structured, but it’s not. There’s no canonical representation, per se. It’s like, “Hey, I want you to go run a mile and then I need you to do 100 pull-ups and 200 push-ups and 300 squats, and then run another mile. Oh, and wear a 20-pound vest if you have on and we’re going to call this Murph.” Right? So, like in CrossFit community, a very famous workout. Great.

Matthew Russell: 00:21:49

Now, how do you take that language and have a machine represent it in a canonical way? First of all, how do you even represent that in a canonical way? Which is a separate problem from, well, how do you take that human language and get it into that canonical representation, so that the machine can actually do something with it? Great, well, what is the machine actually going to do with it? Well, it depends on you, your goals, your skills, what’s safe, what’s not safe, how you’re feeling. There are a lot of variables and we would call this a multi objective optimization problem, if we were to just stick a label on it. 

Jon Krohn: 00:22:29

Right. I think we’re going to have more on multi-objective learning problems coming up later. So, I can see how, yeah, so having all these different kinds of objectives that we are simultaneously trying to optimize. So, keeping people safe, keeping people fit, obviously, is one of the big things that people probably think about first. And probably, a lot of apps think of that as the only objective to optimize for. But there are lots of other things to be taking into consideration, as you mentioned. So, that makes a lot of sense. So, these all sound like really tricky problems. I mean, without divulging anything, proprietary, how are you tackling these? How are you structuring the information, so that a machine can be guiding people in their individual workouts, I suppose? 

Matthew Russell: 00:23:20

Right. So, if we go back, so I’ve taken very much a bottom up approach to building the company. We built a very powerful base that is a technology platform, and we’ve worked our way up the stack and all the way through to launching version 1 of the app and they go to market. But if we think of that initial base and that data model, you have to, I think in all aspects of life, you really have to focus on the fundamentals, right? There’s the saying that the pros are just better at the fundamentals. To me, that’s an important little bit of wisdom here. But what are the fundamentals of fitness and what are the fundamentals of data representation? Well, at the end of the day, I think the fundamentals of fitness or at least one of the most important ones here is a movement. So, what is a push-up? What is a pull-up? What is a squat? Right? What is a deadlift? That these movements, and then data representation, well, how do you represent something in a machine readable way? Well, generally you express a set of features. We call that a feature vector. 

Matthew Russell: 00:24:34

So then you say, “Okay, well, what is the feature vector for a push-up? What is the feature vector for a pull-up? How are those things unambiguous and different from one another? What are the set of features you have to represent in order to ensure that there’s no ambiguity?” And in the grand scheme, it’s a fairly finite universe, although, it is sort of an open universe in the sense that you could invent new movements, new movement standards, new pieces of equipment. But in general, it’s fairly finite. And so, it is a tractable problem that you can approach. And it does take a lot of thinking to get that right, but once you can represent the movements, you can move up a level in the stack and say, “Well, what is a workout?” Well, a workout is a particular procedure you are running, where the atomic units are parameterized movements, right? Go do 100 pushups within a time limit. Go do 200 air squats within a time limit. Sequence them in a particular way after one another. And while those are not the exact quantities of the Murph workout we mentioned, did that on purpose because what you really need in a training program is a way to achieve a goal. 

Matthew Russell: 00:26:03

And so, as you move up the stack one more level to a program, you’re starting to look at cause and effect, stimulus and response. Trying to understand what would the adaptive response be if you put on your exercise scientist hat of doing certain quantities of the movement with a certain work to rest ratio? And you do that in some type of iterative fashion, taking checkpoints along the way, figuring out, “Well, are we actually trending toward the goal or not?” Right? That’s the definition of learning and a learning problem. 

Jon Krohn: 00:26:42

Cool. Yeah, I think I’m starting to piece together how this all works and it sounds cool. I’ve sampled a lot of different fitness apps over the years and it does sound like you’ve got a unique angle here that could create benefits for users across all of the objectives that they have, like fitness and staying healthy, avoiding injury. And I guess things like their time constraints as well have to come in. So yeah, go ahead. 

Matthew Russell: 00:27:13

Yeah, I was just going to say that the time domain is very important, right? So as a human, you’re working on a timescale. You have goals, you have objectives, you have reasonable expectations. And when you start to think about the efficacy of any exercise program, the time domain is critical, both within a workout, right? The stimulus and response of a workout. Work to rest ratio, right? Go do three sets of 10 on the bench press and rest two minutes in between the sets, right? That’s a very different stimulus or response than saying, “Go do 100 push-ups as fast as humanly possible.” Right? That there’re different stimulus response relationships there, but also within the program. Well, are you going to do three workouts, one per day for three days, take a rest day? Do two more workouts, one per day for two days, take a rest day? Are you going to work out five days in a row? These are all variables that at the end of the day the consumer wants to achieve something. And so, as cool and interesting as the technology may be and as easy as it is to get lost in the data science of it, the exercise program has to work. It has to have an efficacy, that’s acceptable. 

Jon Krohn: 00:28:44

So yeah, so we’ve talked about a bit of how you’re taking exercise movements, turning them into feature vectors, and then we can combine these exercises movements together into a series of movements to program an individual day. And then now, you’re talking about that even higher level of, okay, how do we structure somebody’s week or their month based on how many times they want to work out and how long those workouts are? Is there already a machine learning model that you’ve developed that is working to optimize some of these optimizations that we’re looking for, like time and fitness level and avoiding injury? Yeah, tell us about that if you can. 

Matthew Russell: 00:29:28

Yeah. We’re very early in the process, to be completely honest here, right? Where it took a great deal of time to build the foundation, just to even get the representation to a point where we believe, well, these are the right feature vectors. And they are composable and there is an algebra we can perform on them in a particular way, in terms of particular machine learning technology. I’m a pragmatist when it comes to this and I generally like to start with the simpler, easier to use technology and work my way up the stack again. So as a data scientist, the number one KPI for me and what I’ve always thought as like the most important KPI for data science, it’s maximizing the number of experiments, you can run per unit time, right? 

Matthew Russell: 00:30:26

So, it might take me a week to set up a harness to run an experiment, but then I should be able to run the next experiment a lot faster. And as I learned things, keep moving a lot faster, right? So, I want to not just have a velocity, but I want to see that velocity increase over time. So, a lot of our time so far has been really just shipping Version 1 of the platform, nailing the data representation, getting a good experience in place because those users are paramount, right? That’s the whole goal here is once someone downloads the app to make a great impression to keep them around and to learn everything we can about their wants and needs, as well as learn what we can to just improve the core app and its personalization mechanisms along the way. 

Jon Krohn: 00:31:24

Cool. 

Jon Krohn: 00:31:26

Struggling with broken pipelines, still dashboards, missing data, you’re not alone. Look no further than Monte Carlo, the leading end-to-end data observability platform. In the same way that New Relic and Data Dog ensure reliable software and keep application downtime at bay, Monte Carlo solves the costly problem of data downtime as detailed in Episode No. 499 with the firm’s brilliant CEO, Barr Moses, Monte Carlo monitors and alerts for data issues across your data warehouses, lakes, ETL, and business intelligence reducing data incidents by 90% or more. Start trusting your data with Monte Carlo today. Visit www.montecarlodata.com to learn more. 

Jon Krohn: 00:32:17

Yeah, that was a great explanation and I love that idea of spending some time configuring a test harness in a way, so setting up your experimental infrastructure. You have the software that you’re going to need, maybe the hardware that you’re going to need to be able to run experiments quickly. But then once you have that harness developed, being able to iterate rapidly over different kinds of models, different kinds of features, and get that testing time down, that’s a really good tip for our listeners. So, a lot of your experience with software development, as well as with data science, has been shaped by your past decade, holding Senior Technology Leadership roles, including the Chief Technology Officer, the CTO role at several tech firms. So, I’m not surprised that you have all these kinds of useful practical tips for listeners. One particular question I have for you is I know that you, for example, are right now hiring iOS engineers to help you work on the iOS app for Strongest. So, what do you look for in the software engineers and the data scientists that you hire? 

Matthew Russell: 00:33:33

In general, I think you have, think of it as like it’s been said as the soft skills and the hard skills. So, soft skills, for me are those are the most fundamental. I think, you generally can’t change people and it’s not really my role to try to change people. And I can’t really teach things that are beyond some level, within a timeframe and create a good economic return on that for the company, so I always start with, what, back to first principles, right? Integrity is always the core and it’s like so obvious that my tendency is to not even want to mention it, but for me if I can’t trust you, I can’t work with you. And as much as I wish that I could trust everyone and work with everyone, experience has just proven otherwise, but I start with integrity. When you make a commitment, can you keep a commitment, even if it hurts, right? So, we move into grit, right? Are you a person that’s going to persevere and work through the hard stuff? Well, along the way, are you going to manage my expectations? 

Matthew Russell: 00:34:57

Well, communication becomes key. I need to trust. I need to know that you’re going to make and keep commitments, no matter how gritty you need to be. But that’s not enough, really. I need my expectations managed along the way because I need to know what’s going on. I need to make decisions and so being a great communicator, it’s very important, especially under pressure. When you have crucial conversations, how’s that going to work out, because guess what? In any problem we’re solving, we’re going to have crucial conversations, right? Emotions are going to be high and run strong. The stakes are going to be big and we’re going to disagree on some things. Well, that’s a crucial conversation. So, I generally look at those soft skills, regardless of the role. 

Matthew Russell: 00:35:48

And then in terms of hard skills. It’s really whatever is required. So, if you’re just getting started, you’re very early, you may need really great senior or principal engineers to help lay down architecture patterns and foundation. But there’s the saying that if your junior engineers cannot read and understand the code that your senior engineers wrote, you should fire your senior engineers. So eventually, it has to be the case, that you’ve produced a foundation that you can start to fill in with other roles organizationally. And so eventually, you would expect to specialize the labor a bit. Things could get a bit more narrow or unfocused, say going from a senior engineer to a junior engineer, in some cases, because you just sort of naturally have to evolve that way to scale something. And in the moment, yeah, iOS is very much the core foundation of our iOS and Watch iOS app, you, you would have to have practical skills and experience making your way through the many iOS frameworks to be effective as an iOS engineer. 

Matthew Russell: 00:37:12

But as the app evolves, you can start to plug into that architecture. You could be a junior iOS engineer and have a little bit of experience and a desire to acquire more experience. So, that would be a set of skills in particular. Most of the data, not surprisingly, starts within the experience of the app, either on your phone or your watch, that eventually makes it to a server. So once you’re on the server, slightly different set of skills. It’s not iOS anymore once it’s on the server. Now, you can use a different tool chain or a different stack. And you generally use the best tool for the job. So, a lot of my data science experience, and my deepest depths are very Pythonic, so we very religiously, at this point, use Python and it’s been really good to us. 

Jon Krohn: 00:38:10

Cool. Yeah, those are really great pragmatic tips and good examples that color what you look for in hard skills for specific roles. I loved the soft skills that you mentioned – integrity, grit, communication. Those are all so valuable in anyone that, with any software developer, we have to have a huge amount of trust because they spend hours, maybe even days if they’re a bit more senior, on their ow. And you have to trust that they’re tackling the hard problems in an efficient way, so yeah. So, I’d love those points. If people feel, if a listener feels like they have integrity, grit, and good communication skills to manage your expectations. And in particular, if at this time, they also have skills with the iOS atmosphere, the iOS environment, rather, how can they apply? I guess, they can go to the Strongest.AI website? 

Matthew Russell: 00:39:06

Yeah, strongest.com is- 

Jon Krohn: 00:39:07

Oh, strongest.com, I forgot that you bought that. Yeah, it’s just, it sounds like such an expensive domain that I don’t think that somebody, but yeah, you did it. 

Matthew Russell: 00:39:18

Yeah, we- 

Jon Krohn: 00:39:20

So, yeah, of course. That’s where they go. 

Matthew Russell: 00:39:22

We’ve got the dot-com If you’re going to build a consumer brand, which is very much what we’re doing in the vision of the company, it’s just memorable, right? You don’t need to remember some weird suffix. You just need to remember, “Oh, strongest.com.” So, that was a key piece of the broader strategy. 

Jon Krohn: 00:39:44

It makes so much sense, yeah. And my apologies for forgetting. And yeah, so but they can head there and obviously, you have a careers page. Cool. So, hopefully- 

Matthew Russell: 00:39:58

Yeah. LinkedIn is another great way to connect, right? Because ultimately, we’re going to have to talk about professional experience and just generally get to know each other on that level, so that’s always another good way to find me as well. 

Jon Krohn: 00:40:15

Cool. There you go. And at the end of the show, we’ll talk about all the various ways to get in touch with you. And of course, we’ll be sure to include things like the hiring page for Strongest, as well as your LinkedIn page in the show notes. In addition to all this experience that you have as a CTO as a senior leader in technology over the past decade, you also have done a lot in terms of written and oral communication with the public about technology. So, it’s amazing that we’ve already come this far into the episode and we haven’t even had the opportunity to talk about all the books that you’ve written. So often, I have book authors on the show and that’s where we focus almost the entire episode. But the things that you’re doing, as a practitioner, as a technologist, I thought were even more interesting than all the books that you’ve written. Not to say that they aren’t interesting, but it’s just, you have so many things going on. So, you’ve written four books for O’Reilly, many of which have several editions, so 21 Recipes For Mining Twitter, PayPal APIs: Up and Running, Dojo: The Definitive Guide, and now in its Third Edition is Mining the Social Web. And I thought that that might be a book that in particular resonates with our data science audience. So, what do you cover in Mining the Social Web? 

Matthew Russell: 00:41:41

Yeah, Mining the Social Web, so to contextualize it, the First Edition of the book was published before any of the big tech companies that are social platforms had gone public. So Facebook, Twitter were the two big ones really, at that time. MySpace was still a thing back then as well. And the thought I had was, “Okay, this is not just a fad, this is the new normal.” Although, it was unclear maybe if it really was, but it seemed that way to me. And I thought, okay, if you start to scale this and think about all of the privacy considerations, and everything that advanced technology could bring to bear on the data that people seem so eager to provide these platforms, where does that lead? And so, the point of the book was not so much to voice an opinion on that and cast judgment on whether it’s good or bad, so much as it was to be very pragmatic and show people in a very hands on way, what you could do with the data. So, we’ve organized that book into chapters one per social network, essentially. So it’s very, you could cherry pick chapters, if you were only interested on Instagram, or Facebook, or what have you. You could just cherry pick a chapter, but you could also just read the book cover to cover and it tells a good story, kind of a crescendo and complexity as you run through it. 

Matthew Russell: 00:43:16

But the general thought process was, let’s just make people aware of what’s possible here. And let’s do it in a way that back to patterns, like every chapter has a format. So, I think in one of my talks, I described the format as aspire, acquire, analyze, summarize. So aspire, well, I need a hypothesis, I need a question I want to answered, there’s something I want to learn. I’ve aspired to figure something out. Acquire, okay, let me go get a data set and maybe even a set of skills I don’t have, or some software that can help me to actually have the tools. Back to having a workspace to solve a problem. So, I’ve got a problem and now, I’ve got some tools, some skills, some data. Let me analyze that data. Let me get my hands dirty. Let me see what I can learn. And then along the way, summarize. Well, a summarization in this context is really just a checkpoint of what you’ve learned in a concrete enough way that you could structure that as like a miniature experiment. And then of course, back to our KPI for data science being, “Well, let me maximize my number of experiments per unit time. And let me try to increase my rate of learning.” Well, that sounds like a pretty good pattern and in the abstract, that’s the way each chapter flows in the book. 

Jon Krohn: 00:44:49

Cool. Aspire, acquire, summarize. Did I miss one or is that it? 

Matthew Russell: 00:44:53

Aspire, acquire, analyze, summarize. 

Jon Krohn: 00:44:56

Analyze, summarize. I knew I missed one. Cool. Yeah, that’s a really great kind of framework. I can see that you in general, as a thinker, you’re adept at creating these kinds of structures for presenting information or solving a problem. And so, it’s cool to see it here showing up in each individual chapter for each individual social media platform in your mind in the Social Web book. So, I’ve alluded to it already and then you did as well by mentioning a talk. You’ve given some big talks. You’ve done several talks at a rally conferences before the pandemic hit. And a talk of yours that particularly interested me was a TEDx talk that you did on having multiple objective functions. So, we talked about this earlier in the episode about having, for example, for the Strongest iOS app, there are multiple objective functions that we’re catering to. The time that a person can invest in their fitness, what their fitness goal is, and things like avoiding injuries. So, that TEDx talk, it brings that concept to the audience in a fun tangible way and that’s the Zombie Apocalypse. 

Matthew Russell: 00:46:13

Yeah, yeah, so yeah. So, I think the title of the talk was something like How AI Can Make Us Stronger, Faster and Harder To Kill. And becoming stronger and faster and harder to kill, that is a multi objective optimization problem, because what makes you stronger, may not make you faster, what makes you faster may not make you harder to kill, right? So at the end of the day, any sufficiently complex problem, you have to make some trade offs and you can’t max out every single variable because it’s either intractable or really impractical to try to do so. So, as like a human, I really like this idea that if I could be ready for the zombie apocalypse, I sleep well at night, I feel good about myself. And I just sort of lead in the talk, with a little bit of humor and maybe a little bit of pragmatic thinking on just resiliency in order to get into the AI. 

Jon Krohn: 00:47:16

Well, however, likely a zombie apocalypse is something else that that kind of thinking really prepares you for is something like the CrossFit Open. So, we talked about earlier how your platform, how Strongest is the leading platform for fitness events. And so, things like the Rogue Invitational, things like Water Palooza, the CrossFit Games, which is the championship, the annual championship. These have lots of different kinds of tests of fitness. So you have strength tests, you have endurance tests, you have speed tests. You have tests that effectively are testing your mobility or your gymnastic ability. So, you have this broad range of capabilities and one of the greats of the sport of fitness of CrossFit is Rich Froning, who was a four-time men’s champion in the sport, and he’s still active, and he’s from Tennessee. Have you ever met Rich Froning? 

Matthew Russell: 00:48:13

Yeah. We’ve done a great deal of work with Rich Froning. 

Jon Krohn: 00:48:16

Of course, you have. 

Matthew Russell: 00:48:18

I was in CrossFit Mayhem last week. 

Jon Krohn: 00:48:20

Oh, wow. His gym. 

Matthew Russell: 00:48:22

And yeah. Cookeville is 90 minutes from Nashville, give or take. 

Jon Krohn: 00:48:27

Oh, man, I didn’t piece that together till I got to think about it. Yeah. Wow. So, of course, Rich Froning. Well, for listeners who don’t know. He’s like Wayne Gretzky is to hockey or Michael Jordan is to basketball as Rich Froning is to CrossFit. And I saw him do an interview a little while ago, I think it was a recording of an interview while he was still active as an individual competitor. And he was talking about how you have to be careful in CrossFit because making yourself too strong can limit your speed capacity, or making yourself too flexible can have a negative impact on your strength. And I had never, I didn’t realize that. I kind of had this idea as a relatively, as a completely amateur and just recreational crossfitter have this idea in my head that I can optimize all these things at once. That I just need to find a way in my day to be doing long runs while simultaneously doing heavy back squats and gymnastic stuff. And somehow, I’m going to emerge as better than ever at all of those. 

Jon Krohn: 00:49:36

And so in a way, it was kind of a relief to hear him say that, I can’t be at a time like now where I’m the strongest I’ve ever been, I can’t expect no matter what I’m doing to be as fast as I wasn’t my fastest a few years ago when I was training for a marathon. So, anyway, so that was kind of a relief for me. That’s kind of, it’s an example of this multi-objective function that you were talking about for avoiding the zombie apocalypse. But beyond it being something that’s useful for CrossFit or the zombie apocalypse, this is a computer science idea, right? It was a machine learning idea. Do you want to talk about it a little bit more than that context? 

Matthew Russell: 00:50:20

Yeah. A multi-objective optimization problem is one in which you’ve got many variables and as an optimization problem, you’re trying to maximize whatever the function is. And if you think about just some curves you’ve seen, like in algebra books or standard, like high school textbooks, you’ve got your ups and downs and your peaks and your valleys. And so, you’ve really only got two variables if you’re like in the Cartesian coordinate space, though. So, you’re really one variable, right? You’ve got this X value and you’re trying to maximize the Y value in that example. But as you move into more dimensions, so you think about three dimensions. So, imagine you’re looking at a topographical map where there’s mountains and valleys and like the Earth’s surface. Okay, well, how would you find the tallest point on the Earth’s surface or the deepest valley on the Earth’s surface if you were using a machine to do it? Knowing that there are lots of local minima or maxima along the way, you could get stuck on a certain peak or in a certain valley, because in any search algorithm, there’s really two things that are always fundamentally trading off against one another – exploration and exploitation. 

Matthew Russell: 00:51:49

So, if I explore more, I can cover more surface area and potentially, find a new path that leads me to a better outcome. But if I found something pretty good, I can try to exploit it and get every, squeeze out every last little bit of juice, but I can potentially get stuck. And I mean, that’s true in life and in the way you think as a person. It’s true in computation, as you think of satisfying multi-objective optimization problems. But then when we sort of come back to machine learning and AI, well guess what? If you have a feature vector with 100, or 200, or 300 features, and when you start to deal with human language, you’ve got a lot more features than that. Well, now we’ve got this hyperspace, potentially hundreds or thousands or more features, and you’re trying to navigate all these hyperplanes and maximize or minimize some outcome. I mean, that’s where the state of the art constantly leads us is better techniques to do that, right? That’s what all the rage about neural networks as of late has been is just finding ways to navigate, like multi-objective optimization problems in hyperspace with lots of dimensions. 

Jon Krohn: 00:53:15

Super cool. And you know a lot about this, because you studied this in your Master’s. So yeah, I’ll give a bit of an intro to that. So, out of all the incredible things that we’ve already described about you, Matthew, we haven’t even gotten to your background, which is, it’s not, well, it isn’t the most common path that we have into becoming serial CTO, serial technical book author. And I think it might open some listeners minds to another path that they have as an option, certainly in the US and probably in any country, which is that you can gain your education as a part of military service. So you, for example, got your education while in the Air Force. So, you’ve got a Bachelor of Science in Computer Science with distinction while also on the Parachute Team. And then you did a Master’s in Science, in Computer Science and Machine Learning, again, with distinction. And in that you focused on these multi-objective functions, these multi-objective problems, so I’m sure that there are a million things that you can describe to us about that at background. But I just love for you to share with the audience how that experience, yeah, how that was useful to you later in your career and why other people might want to consider this military education route themselves. 

Matthew Russell: 00:54:48

Yeah, I would love to say a little bit about that. I grew up in rural West Virginia with my great grandparents, so this was like the greatest generation right? They were in their 70s when I was a little kid, just to put that in perspective, like multiple generational gaps. And we were living somewhere around the poverty line. There was not a lot of opportunity and so, my best path to have a future was, there were only so many possibilities that I could see. And the best possibility I could see was military service. And then I learned, well, you could essentially go to a military school, like the Air Force Academy or West Point or the Naval Academy, get a world class education, learn to be a leader in what’s really a leadership laboratory. Go do some great things for your country along the way and who knows what’s on the other side of all that, depending on how long you want to stay in uniform, and what you learned along the way. So, for me, the Air Force Academy, like getting into the Air Force Academy was a multi-objective optimization problem, because you had to have certain test scores, you had to have a certain level of civic participation, you had to have a certain level of physical fitness, you had to have a certain leadership aptitude, you had to get congressional appointments. 

Matthew Russell: 00:56:23

So for me, that was an interesting problem space that said you’ve got to be a reasonably well rounded person to even get in the front door. And then by the way, once you get in the front door, there’s a lot of other people that did it, too. And, so you’re going to be surrounded by some amazing people while you’re there. And that was just obviously, a lot of people will say in the military, that’s one of the best aspects of it is just the quality of people you get to spend your time with and the camaraderie around that. So, I had this chance to be a well-rounded person and sort of take that to an extreme. I’ve studied Computer Science. Went on and did a Master’s and studied Computer Science, again. The problem for my thesis that was really what opened my eyes to machine learning and multi-objective optimization problems in genera, was, so this was back in 2004, 2005 timeframe. The idea was, well, what if we could route swarms of unmanned aerial vehicles that were very, very small, maybe small to the point of insect small or bird small. So, back in 2004, 2005, there is no such thing. You could have read my Michael Creighton’s book Prey and sort of got some ideas, but this was not a thing, really. It was a bit futuristic. 

Matthew Russell: 00:58:00

So, the Air Force was working on practical applications of this early, and I had a chance to tap into an Air Force Research Program and produce a thesis on the topic. We basically did two things. One was, well, how do you even simulate this with software, in general? How do you even simulate something like this, regardless of what it actually does. That’s a high performance computing problem, especially back in the early 2000s. And then the really interesting part to me was more of the algorithmic aspect of it. And the idea was if you had this swarm and there were multiple objectives, right? So, some of these UAVs might be providing camera coverage to an area. Some of these UAVs might have lasers on them. Some of them might be taking other sensor measurements, right? You’ve got multiple objectives. Well, how do you cover a surface area and provide a certain level of sensor coverage in a way that’s completely decentralized, right? You don’t want to be able to, if you were to knock one of these UAVs out of the sky, how does the swarm reconfigure itself in a completely decentralized way and continue the mission? 

Matthew Russell: 00:59:26

So that was, we took genetic algorithms and applied them to solve what’s called the vehicle routing problem, a particular instance of the vehicle routing problem, which is similar to what FedEx or UPS or delivery services do, just in a different way. A very complex computer science problem, well-studied and we kind of advanced the state of the art just a little bit and published it. And that’s what really opened up that this entirely, entire new world to me in a new way. 

Jon Krohn: 01:00:01

Very cool. And now fast forward to 2022, and this kind of technology exists. So, the work that you did was a stepping stone towards having these unmanned aerial vehicles, these UAVs in swarms with multiple objectives. And being able to handle exactly the kind of situation you described where a subset of the swarm is brought down. Very cool. And I can only imagine that your experience in the Air Force was critical to some of the kinds of traits that you look for in your employees. And that no doubt you develop more in that time, things like integrity, grit, communication capabilities. Is there anything else from that time that has had a big influence on your later career as a technologist and then now, entrepreneur? 

Matthew Russell: 01:00:59

I think it was also in the military, especially in military school, that I really appreciated the whole mind body concept. It’s not enough to just use your brain to solve problems and it’s not enough just to be healthy and physically capable. There’s this deep relationship between the mind and the body and the balance between the two. That sort of struck me in a particular way, because you’re sitting there cramming for a test or a final or something with significant consequences. It could have been a mental test. It’s your aeronautical engineering final exam. But guess what? The same week, you might have had a physical fitness test that you should have been preparing for, staying on top of. 

Matthew Russell: 01:01:59

So for me, that’s an important part of life. And back to Strongest, if at the end of the day the goal is to make the world a fitter place, there’s a lot of people that are hurting and suffering. And I see a lot of the chronic disease and the chronic pain is not always, but there’s a lot of self-inflicted pain in that way. And if we can help people see that and use their body in a more productive way, use their mind in a more productive way, live a better life that’s really what it’s all about. When you look back on life, I think you’d want, I don’t want to look back on a lot of fear driven behavior and have a lot of regrets about didn’t do this, or didn’t do that because I was scared. I mean, definitely, for me, I stamp that stuff out. I try to find fears and attack them wherever I can. But then more broadly, it’s like, “Well, yeah, what did I build and was it scalable and durable, and to what end? Did it help people?” For me, that’s what Strongest is. It’s a vehicle to take these ambitions and really deploy it at scale in the world. 

Jon Krohn: 01:03:24

Super cool. Beautiful message there to deliver. Is there anything that given you’re relatively young, but you’ve already achieved an absolute ton. The military service, authoring many editions of four completely different books, CTO of several different tech firms and now, an entrepreneur behind what has already a remarkably successful early startup. Is there anything that you wish you’d done sooner to attain the success that you’ve had? 

Matthew Russell: 01:03:59

Yeah, there may be a couple. Two things that immediately come to mind. One is that I wish I would have built a better foundation of business fundamentals much earlier in life. And another is that I wish I had appreciated feelings and emotions and things that have nothing to do with technology or engineering, but are just quintessential to the human experience on a deeper level. A lot of my life was kind of working around feelings and not fully understanding emotions. Not really being able to put a precise label or word on certain emotions. I think looking back, wow. One, I’m glad I’ve started to figure some of that stuff out. And two, I know I’ve got a lot more to do there, but I wish I’d had a better grasp because I think that’s back to fundamentals. Those are just fundamentals being human. And then on the capitalism piece of this, business fundamentals. Well, capitalism, think about that for a minute. An ism is like we’re going to systematically build a world around something. Well, capitalism is like, well, we’re going to build a society on this premise that it’s to some extent all about the money. We’re going to disagree on religion, and politics and lots of things, but we’re going to agree on the roles for making money. 

Matthew Russell: 01:05:36

I know that’s super high level, but it never really hit me that, I never thought about the word that way. But eventually, came to understand, well, all of these technology problems, all these engineering problems that we’re overcoming are in service to a more abstract set of problems that are going to be formulated as business objectives, whether it’s my company or someone else’s company, that are in service to solving some other set of problems. And if you follow the money, and start to understand how money and power are often related and all of the nooks and crannies and twists and turns that can take you on, I think it’s just good operational context, as you’re making decisions along the way. At the end of the day, if you’re going to be in business, you’ve got to make money, you’ve got to turn a profit. You’ve got to have something that creates more value than it consumes. That is trickier than it sounds. It’s easy to solve an engineering problem in isolation. 

Matthew Russell: 01:06:49

I used to think that the smartest people were engineers, right? I think, wow, the scientists and of course, the brilliant. I’ve come to appreciate just how smart you have to be to make money and actually make more money than you spend with a good cost basis along the way. For me, that’s been a huge part of personal and professional growth as an entrepreneur. 

Jon Krohn: 01:07:23

Yeah, I’ll echo that completely. I, now with coming on, I guess a decade, out of academia and in the capitalistic world, even though academia to some extent is capitalistic, you’re sheltered a bit more than you are starting a company, of course. And I think that a lot of scientists and engineers share that perception that this work that we’re doing with linear algebra and calculus and machine learning models, probability theory, this is really complicated and we’re really intelligent for doing it. And of course, you are, but being able to make those mathematical foundations work in the real world, on an application that somebody hasn’t already figured out how to do efficiently, that you can sell to somebody else, that you can generate a profit on, it really can be difficult. 

Jon Krohn: 01:08:22

And it’s interesting how, I know there are people out there who study business and economic as degrees, but it seems to me that at every level of education – elementary school, high school, undergrad, grad school, having increasing complexities of economic understanding, of pitching company ideas of marketing an idea, of figuring out how to price something, so that it’s profitable. These kinds of skills, most of us, only learn them after academia. And I’m not sure why the system needs to be that way. Even if you think about people who are in a more everyday level, like not everybody needs to go out and start a company. But just in terms of the economics of running your life profitably, most people don’t learn that. And either have to kind of come across it haphazardly or have somebody teach them a friend or family member, or they’re just racking up credit card debt because you just think I can spend. Somebody is giving me the authority to spend. I’m going to buy more things. I’m not going to worry about what’s that’s going to be like 10 years from now. 

Jon Krohn: 01:09:35

So anyway, so I’m trying to echo some of your ideas and add some extra in there. But definitely, understanding business fundamentals, being able to navigate this capital ism, that all of us swim in everyday, usually valuable. And yeah, I certainly have a ton more still to learn in that space. So at the end of every episode, I ask for a book recommendation. And I have a feeling, Matthew, that given your rich background and all the rich ideas that you have, you’re going to have a really interesting book recommendation for us today. 

Matthew Russell: 01:10:13

Yeah. I like to collect quotes. Quotes, the best quotes, I think, are little bits of wisdom that have profound meaning. And a quote that comes to mind is, “If you want to go fast, go alone. And if you want to go far, go together,” right? I think that’s an African proverb. And going together means there are other people on the bus with you. You’re on somebody’s bus, they’re on your bus, you’re on the bus together. And for me, the most gifted book that I’ve ever given, right? When I have new teams, I will buy copies of the book for every member of the team is called Crucial Conversations. And so, there’s a definition here. A crucial conversation, I alluded to this earlier, it’s one in which the stakes are high and opinions vary and emotions run strong. So, when you think of any problem worth solving in life, any situation you actually want to be in, in life, probably involves those variables. 

Matthew Russell: 01:11:25

And when you’re interacting with another person, you have to be able to navigate that. And I think that is one of the most important skills in all of life. No matter what you do if you can navigate a crucial conversation, if you argue with someone, we’ve all heard that even if you win an argument, you still sort of lose the argument, right? Because when you argue with someone, you win it. You sort of stripped down someone’s perception of reality in a way that probably hurts a little bit or is uncomfortable. Sometimes, you have to help people see things more clearly. Some things you need to see things more clearly regardless of which direction the conversation flows. If you know it’s a crucial conversation, and you can start to recognize that it is a crucial conversation, the outcome is going to be a lot better. And you may not be able to control the outcome, but you can control the process and you can make sure you’ve improved the process along the way by knowing this and that probably does lead to a better outcome. So, it would be my number one book recommendation, Crucial Conversations. 

Jon Krohn: 01:12:53

I love it. I knew you were going to have a great one. All right, Matthew, clearly, you are a fount of valuable knowledge and inspiration, entrepreneurial ideas, scientific and engineering ideas. How can listeners follow you or get in touch with you? 

Matthew Russell: 01:13:12

The best way at the moment would be LinkedIn. Just reach out, send a connection request or follow me on LinkedIn. My email is matthew@strongest.com, M-A-T-T-H-E-W @strongest.com, so also that’s a fine way to get in touch. 

Jon Krohn: 01:13:33

Super cool. And yeah, so I’d recommend if there are people out there who are, first off just interested in asking Matthew questions. But in particular, if you’re interested in learning more to potentially invest in the pre-seed round for Strongest, if you’re interested in fitness and AI that could be something that you’re keen on. And if you’re not interested in investing, but you’re still interested in fitness and AI, maybe you’d like to be a heavy beta user of the platform to be able to provide a lot of feedback and make it a better product. So, those are particular things that you might want to reach out to Matthew for. All right, man, thank you so much for being on the program. I’ve learned so much from you. It’s been a wonderful episode as I knew it would be. And hopefully we’ll get you on this show again sometime in the future and we can hear how the Strongest journey is coming along. 

Matthew Russell: 01:14:23

That sounds great. Thanks for having me. 

Jon Krohn: 01:14:24

What an inspiring entrepreneur and data scientist Matthew is. He has tons of success in the rearview mirror, but no doubt orders of magnitude more lie ahead for him. In today’s episode, Matthew filled us in on how cloud native web development with a platform like Heroku makes data intensive applications like Strongest, horizontally scalable to any size. How Terraform makes life easy by automating infrastructure management. How setting up a test harness for running machine learning experiments can enable you to iterate on your machine learning models increasingly rapidly. The integrity, grit and communication skills he looks for in the engineers and data scientists he hires. How multi-objective functions are practical for countless real world applications, such as the strongest fitness platform and swarms of unmanned aerial vehicles. And he covered how he wishes he’d better understood emotions and business fundamentals earlier on in his career. 

Jon Krohn: 01:15:24

As always, you can get all the show notes, including the transcript for this episode, the video recording any materials mentioned on the show, the URLs for Matthew’s social media profiles, as well as my own social media profiles at www.superdatascience.com/545, that’s www.superdatascience.com/545. If you enjoyed this episode, I’d greatly appreciate it if you left a review on your favorite podcasting app or on the SuperDataScience YouTube channel. I also encourage you to let me know your thoughts on this episode directly by adding me on LinkedIn or on Twitter, and then tagging me in a post about it. Your feedback is invaluable for helping us shape future episodes of the show. 

Jon Krohn: 01:16:01

All right. Thanks to Ivana, Mario, Jaime, JP and Kirill on the SuperDataScience team for managing and producing another wicked episode for us today. Keep on rocking it out there, folks, and I’m looking forward to enjoying another round of the SuperDataScience podcast with you, very soon.