Driving Shiny adoption by removing friction | Jonathan Lin | Data Science Hangout

Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12 p.m. U.S. Eastern Time. If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience.

I am so excited to introduce our featured leader today. We're talking with Jon Lin, advisor of data science and AI at Oventive. Jon, welcome to the hangout. I would love it if you could tell us a little bit about you, something you like to do for fun, what you do for work.

Hi everyone, I'm Jon. Let's go over the fun stuff first. I love dragon boating. One of the things we do at work is we organize a little dragon boat team here in Calgary. It's a big passion of mine, and I do coach the team, so I'm very happy about that.

Jon's background and early career

I had a really great start after university. I graduated in 2008. This was during the best time ever to graduate, which was during the Great Recession. Even better, I had a Bachelor of Commerce, which meant that my job prospects were a little bit thin. I was really lucky to have a job offer at Ernst & Young, which was kind of one of my first real out-of-university jobs. It was really around IT governance, so I learned a lot about IT governance, doing a lot of auditing on these public companies, and that's kind of where I got my big foundation is kind of being disciplined in that.

But at the same time, this is where I started to pivot a little bit more into analytics. I got a big taste for analytics doing kind of what we call journal entry testing, but looking at these big public companies to see if there was any evidence of fraud. And that was done through these kind of, I'd say, old-school software packages at the time. But that's where I started to really love doing kind of this data analytics work.

Through that time, that's when I actually got my first taste of R. R back then, this was around, I'm going to say, 2010, 2009. That was kind of an R GUI thing. RStudio didn't exist at the time. So R GUI was, for those who kind of remember, was a little bit rough around the edges. It was totally workable. It was beautiful, but back then it was really hard to learn. But that's kind of where I got my taste for R, and I really liked it.

Joining Encana and early Shiny adoption

Doing that data analytics work, I was really lucky. It pivoted me into that second part of my career. I joined Encana. Encana is an oil and gas company. It's now called Ovintiv. But basically, I got the opportunity to revamp an entire anti-fraud department using R here internally at the company. So that was my first real great application of R.

Really wonderful things started to happen after I joined. Number one, there was these pockets. So this is around, I'm saying, 2013. Little pockets of R users in the company. There was this one guy I remember very fondly. His name is Tim. And he was doing some really good production planning stuff in R. These algorithms for kriging, very difficult to do, would normally be done in Excel, or you had to pay someone to do it, pay an external company to do it. He just did it himself in R.

But one of the challenges that he had was that he had to put everything on a network drive. So the first pain that a lot of people have when they're starting to start up in R, they don't have the right infra. Well, they just put everything on a network drive. And they say, well, just go install R. And for a lot of petroleum engineers, that was already kind of over their heads. So that was a little bit challenging at the beginning for us to use R.

But a second really great thing came around. So we had the algorithms in place, but they were kind of harder to use. Then we acquired, the company acquired another team and that team, what they did, and I'm kind of looking at Kyle's messages, because that's what made me really happy. This team that we acquired, they brought an amazing talent. And one of the things they had was this Shiny app. Again, it was a beautiful app. It solved a wonderful business problem. We couldn't buy the software anywhere else. But it was solely dedicated, it was solely hosted on a network drive.

So for us, adopting that was again, you know, it's not so bad nowadays with RStudio IDE, we kind of forget the trauma. But for someone again, that was new to R and wanted to do something, it was really, it was really challenging to distribute and keep up to date. So that's when we got, that's when we engaged at the time RStudio. And Katie was actually our customer service rep back in the day. And we brought on RStudio Connect. And it was, it was a game changer for us.

Scaling with Posit Connect and Kubernetes

We put it on a virtual machine. We started off with 50 people, which we thought was a huge license amount at the time. And the amazing thing about when we started to bring on RStudio Connect is that people started just to jump on board with the tool. And people start to make their own tools. And we kept growing our virtual machine bigger and bigger and bigger. We actually got to a point and I love this part of the kind of the career where you start to experience scale. I know we don't, we're not like meta scale, but we started to experience the scale of the company with these apps. And we actually ran out of, ran out of the ability to expand memory on our virtual machine. We had half a gig of memory on a virtual machine and things were just getting catastrophically ending because there were things were super popular.

And so we were like, okay, we're at RStudio Connect. We've hit a growing problem. We've proved out what's our next step. We were really lucky because RStudio, the RStudio Connect team, the engineering team started to introduce Kubernetes. So Kubernetes, just for everyone who doesn't know, Kubernetes was a big game changer for us. It basically allows you to scale your virtual machines. So if you run out of space on one virtual machine, it adds another one, adds another one. And the other real beautiful thing is that it kind of just created little houses for people's apps.

So you can have an, you know, sometimes we have an R app, an R Shiny app, and it kind of runs away and it goes from like four gigabytes to like 50 gigabytes. And you're like, what the heck's going on here? So the beauty of us going to this sort of platform was that we could put everyone in their own little house. And we said, okay, everyone, you get a little bit of space. And you know, you would think that would be restrictive at the time, but what it actually encouraged our engineers to do is actually start to develop responsibly. It was like, oh, you know, I have a lot of space here. Why am I running out? And it got them to really think.

So anyways, long story short, that's where we are today. We put Posit Connect on Kubernetes back in 2021. We've grown from what I, you know, a very small number of licenses, which we thought was an aggressive proposition. We are up to 900 people. 900 people use our platform every day. We've grown from maybe two to three publishers. We're up to 60 this morning. I just checked. 60 people who have actually published things. So I think, you know, that has been such an amazing kind of background of my, you know, my career here so far using Posit Connect to really start to accelerate kind of innovation at the company at a grassroots level.

We are up to 900 people. 900 people use our platform every day. We've grown from maybe two to three publishers. We're up to 60 this morning.

Amazing. And there are absolutely some things that we can clarify for folks that I'm seeing in the chat, which is like, one, what is Kubernetes? And Kubernetes, I think, like John explained, is a tool that's going to automate the management of containerized applications. And like you said, it's like dynamic memory allocation. So you don't have to go and say, like, I need this much memory to this thing and this much memory to that thing. As you need it, it will create it for you. John, am I right?

Yeah, yeah. So I think that's one beautiful thing. Like when you're in RStudio Connect and everyone's on the same machine, you're always kind of worried that someone's going to cross the fence into someone's side, right? So like when an engineer was pulling like five years worth of data, that might inflict pain on the person that's, you know, doing half a year's worth of data. So Kubernetes is this kind of this construct around Docker images, where you kind of really create these isolated environments and you put guardrails up. And it does allow for that scaling.

And I think that was the big thing for us, because I didn't have to worry about trying to justify every single approval for more memory, because that's what IT does. Why do you need more memory? Well, I need more memory because we need it. Well, are they using responsibly? Yeah, kind of, but sometimes not. So this really helped that conversation out, because, you know, IT wants us to, you know, me being having an IT background and IT governance background, I want to be responsible, but I also want to enable people, right?

So having, you know, Kubernetes, one of the beautiful things we talked about, things can scale. It scales up, scales up, scales up during the day. And then at night, when everyone goes home, you're not stuck with a mega expensive machine, it all just gets kind of wound it down. And that's kind of a beautiful, elegant thing. And that's just made invisible by Posit Connect. Like it's so seamless. It makes, frankly, it makes my job easier. Because I don't have to worry about micromanaging and downsizing every other on the Friday night.

Dragon boating and navigating a sparse job market

Well, I would love to hop into Slido to have some questions asked live. And the first person I would like to ask is Noor, because Noor has two questions for you. The first one is a callback to your fun thing.

Yes, sure. Hello. Can people hear me? Yes. Hi, my name is Noor Sheer. So this is a two part question. I also remember 2008. So the important actual question that people care about is, what are tips for navigating a sparse job market, given your experience back in 2008? And the additional question is, what is dragon, what is dragon boating, my dude? Like, what, what is that? You can't just say dragon boating and just like stop.

Okay, let me answer the let me try to tackle the more difficult question first about navigating a job market. I will actually just punt this one down. I've been here for 10 years, and well, 13 years, actually. And the only reason I got this job was really, I went at the time, they were looking for someone to reinvigorate their program. And it was around MS Access. And it was this other proprietary tool. And I was known for that. But I think what happened, at least I was very fortunate, I had a very good not not like a better word of portfolio, but a reputation. And I think, for me, my personality of being kind of ridiculously enthusiastic about certain things and selective things helps out when you're when I get a chance to show up at the table. Because on paper, you know, I think one of the challenging propositions is posit connect can be substituted for 100 different things. And a lot of that nuance is lost on a paper.

Dragon boating, what is dragon boating? It's ridiculously competitive. It's a historical sport has its roots in China, but really, really has been modernized for competitive racing. So you kind of get 20 people in a boat, you all sit by side by side, and you're kind of hip to hip. And you hold a paddle. And you actually hold it kind of top down like this. And you're actually pulling forward, you're kind of putting your paddle ahead and pulling yourself forward. And it's a very intense race. They're sprint racing. And it's usually sprint races are about 500 meters, and they take about two minutes. But it's 20 people all working together. And I that's why I really liked it here at the companies, because there's no, you know, you could do other team sports, but they're all have an independent component to it. Dragons, dragon boating is just one of those things where there is no singular star, everyone has to work together.

Creating an internal handbook for adoption

Let's hop over to Logan who asked a question in the Slido. Logan, if you would like to ask your question live, we will help you unmute.

Hi, John. I'm Logan. I'm at a community college where folks may not want to always adopt new tools, unfamiliar tools. So I have a question for you. It seems like you've done some introducing of new tools. When doing that, how do you support all of your users with different skills and ensure adoption across the board?

You know, I think for me, so one of the successes that when we were bringing on Posit Connect to grow to that scale, one of the things I think is a relatively large challenge is that people tend to just deploy tools and they just kind of let it be. And I think that is a recipe for failure because it's never a build it and they will come situation. So one of the things that we did that I was really happy about was to create what I call a handbook. I created an internal handbook. And the reason I created an internal handbook is because every company does things differently. You know, you look at all the great kickstart examples that, you know, let's say Posit has on how to do something machine learning related or is Shiny related or API related. You know, there's all these things that you read up on the internet, but people don't know how things work internally.

So every time I got a question, so this is kind of how I built it up organically. I had this little website. Now it's a portal website, but I used to have this little Hugo website that I just published. And I would be like, okay, someone gets me a question. I would go into my handbook and I'd say, oh, okay, this is a question. Let's document what the, what the inventive way of doing something is. And this handbook has grown from, you know, it grew from simple things like how do I even get access to Posit Connect, which I have opinions on, all the way down to how do I, why am I running into this really, really weird error message?

And, you know, adapting these, when you develop a handbook for your own company, I think people start to go there and trust it. You know, it works for you. It works as an active person for you to really help drive that sort of engagement because you start to, you actually are, I think a lot of people here are empathetic to people's problems, right? And having people find an easy way to solve, to get an answer to a solved problem, kind of really helps drive that virtuous cycle. Because I'm now getting pull requests on my handbook from people across the company that I've never even heard of, and they're just getting on to Posit Connect.

So yeah, that was probably the big thing is just for me, the biggest success has been the handbook. I spend time with people. I love talking to people. I love seeing people succeed. I've always wanted to be a teacher and I feel like this is just a, another outlet to kind of live that dream.

Fraud detection methods

You said you worked with fraud data and I've been involved in lots of stuff like that. What types of fraud were you working with? I'm aware of lots of different types. I'm curious. What were you doing?

You know, so journal entry testing for fraud detection was really looking for unusual patterns. Basically you look for unusual patterns of movement. So, you know, probably the best way to do it is like if you thought about clustering something, right, and you cluster the way money moves in a company and you see, you know, there's money going into a subscription account and you are getting cash in from someone else, right, there's usually generally applied patterns that you would see flow throughout a company. And it's when people start acting out of the ordinary where those patterns start to emerge.

So, you know, for me, one of the really powerful things is just the classic, was the classic clustering methods, right? Because you could see the pattern of how data moved and you got to say, there's a little bit of a group here that just doesn't seem to fit. And that would kind of involve, you know, that would involve the dive in on stuff, you know, so that was kind of one aspect. Another aspect of a little bit of fraud detection is that we were always looking for fraud triangles, like, you know, could someone set up a conflict of interest? Could someone set up their own company and start paying themselves? You know, so if you start to, you know, understand the standard way of the way the business works and hopefully reasonably controlled, those sort of patterns start to come out. So that's how I normally conquered anti-fraud detection at the time. I wouldn't say it's, there was never a magic button. I think everyone here knows there's never a magic button to detect fraud, but it kind of helped you triage and know what to look for.

Getting people to read the handbook

So the question, I love the idea of creating a handbook or collating best practice and advice into one place. My question was, how do you get people to actually read it? Because I find that if you write the handbook, then people just know that you're the person to come and ask rather than actually reading it.

I 100% empathize with that. And I think that is the double-edged sword of having an open door is that people feel like that they should just ask you. There's kind of two things. Number one, yes, inevitably, people are going to come to me, but because they trust me, I love that as a signal. I love that people are willing to come to me. But yes, I also have a day job and I also need to get some stuff done.

I think one thing that I started to do was I answered the question. So I would, I would always reply to an email or a Teams message. I say, here's, oh yeah, here's the thing. Here's the answer to the question first. And then I would do two things. Number one, I would just point them to the handbook, very friendly. Hey, you know, I answered, this question has been answered here. And then I actually also go back and say, hey, I, you know, A, did you see this? And B, if you did see it and it didn't seem to answer your question, is there a way I can improve the documentation? Because people, people love to give feedback, but you also see, you also signal that you care.

The second thing that was actually kind of fun early on a days, we, our team here at, you know, our data science team, one thing that we started to do was we were experimenting with early rag, you know, retrievable augmented generation on the large language model side. And we actually took our Quarto handbook, which is written in Markdown, which is perfect for AI. And we put it into a vectorized search. And we also started to say, Hey, you know what, did you, just so you know, you let me know how this works out. And we just put it in the query, put in a little screenshot of the same thing again, not being overly, not being in a passive aggressive way, not to say, you don't go talk to the bot, never talked to me again, but just to say, Hey, just so you know, this exists.

And it's kind of like when you start to introduce Posit Connect or you introduce any tool, it's a long marathon on culture change. And the culture changes. I want to be transparent with the knowledge. Here's the knowledge. Here's how you can access it. I don't demand that you use it, but if you want, don't want to wait for me, cause I don't want to be in a bottleneck. Come here next time. Maybe you'll get your answer.

How to learn R

How did you learn R? What Coursera courses could you recommend, or would you simply recommend reading books by Hadley Wickham?

So I'm going to speak because now it's so different about how to learn. I, let's not, I'm not going to go into that side because that's a very deep, complex topic on education and enablement. But, you know, I, if I look at my bookshelf over there, I've got a few, I've got maybe five or six books on, on R. Hadley Wickham was one of them. I did a lot of online. So one thing I am fortunate here, I don't really, to be honest, I think the R stuff was quite a bit traumatic because it took me a lot of personal pain to go through. This was in the age before good YouTube tutorials and, and those things. So I read a lot of books and I was, you know, just kind of grinding through it.

I think the only thing that really accelerated my work wasn't going through the coursework because the coursework is great, but it's also kind of generic. When I wanted to learn how to use R, there was no resources for it. So I ended up just applying it directly to my work use case and just seeing how well it stuck and then starting to write it down. Like, so let's say we talk about R again and I was doing it early in audit days, I would write a procedure. I would take something that I would learn from Hadley Wickham or many of the brilliant people out there that I'm trying to learn from and just do simple stuff and just try to document it.

You know, I think you can't, you can read all you want. I kind of take this like 3D printing. You can read up all you want about 3D printing, but until you see the magic of something coming out and completely collapsing because you set up something wrong on the printer, you won't ever really learn how to use a language. So I'd say that because R was a challenge for me to learn, especially back in the day, Python was a little bit easier, but still a little bit hard. I think my only real thing is at some point after you read all these great things, you have to start applying because you could stay in learning mode forever, but I don't think it's going to stick.

Getting leadership buy-in

How would you recommend getting a company to buy in or get started with not just, you can download it to your computer setup, even if you only have a few users? And this comment says, n equals one or two, especially if they're very stringent with security.

You know, I think there's kind of two angles. Number one, for me, knowing how to answer IT questions and being, even nowadays, I'm becoming more disinformant. So, getting some, like, say, if you're in a high security situation, always just talking to, like, I think large language models are, like, the big deal right now, right? Like, I'm sending my data out to a large language model. How do you, you know, what are the security risks? Are people going to train on my data? I think engaging the right stakeholders on the IT security side has got to be a big part of it. On the IT security side has gotten me a lot of wins and buy-ins, because it feels like I'm being transparent.

The business side has always been, I empathize with you guys, with whoever asked that question, because it's always been a challenge. And I think for me, I don't know if my strategy is considered kosher, but I show and I don't tell. Because I think a lot of people, when people, it's hard to imagine what a better case would be like. You know, you could say I'm going to develop an app, and it's going to optimize workflows, and it's going to make decision-making faster. You know, those are all kind of great business-like things, but they don't really show any emotion, right? So, it's hard to compel to the person above you in the tree to do it.

So, I've always loved the show-don't-tell mentality. You know, work within the confines you have. Like the Shiny app that we talked about years ago that was on a network drive. We said, you know what, it's making changes. It's making people's lives better. But there's a lot of stuff I know I'm missing, and I don't want to do it wrong. I don't want to be Shadow IT. I don't want to circumvent installation policies and do stuff like that. So, for me, I like to show and I don't tell to get kind of that imagination going. And then also express an interest to be better, because I think it shows a little bit of professional responsibility that you aren't going to cowboy things at all cost. You're willing to go at a reasonable pace, but you still need to have something to show. And I think that's been my biggest success.

So, I've always loved the show-don't-tell mentality. You know, work within the confines you have. So, for me, I like to show and I don't tell to get kind of that imagination going. And then also express an interest to be better, because I think it shows a little bit of professional responsibility that you aren't going to cowboy things at all cost.

It's been for a lot of my projects, I'll be honest. It is a grind. It can be a grind. You know, you think you're doing the right thing. You know, we had, you know, Posit Connect was one of them. Doing the handbook was one of them. I wouldn't get buy-in to do that if I just said, I'm going to make a company repository. That sounds great in theory, but people don't see it until they actually start to feel the pain that you're solving. So, that's my advice. I love to show and not tell, but don't show something that's perfect, but just show enough that it gets people excited.

Removing friction to drive adoption

For me, so if I reflect on trying to get people on, I think you the biggest thing was empathizing with the pain. You know, there's a lot we're all trying to solve our own problems in our own way. We're very we're in the weeds on our own stuff. And sometimes to look at what an engineer is doing and you're you know, you're seeing they're making great stuff. I want them to focus on that. So for me, I think it's a little bit more of a service, right? Like when I say service, I mean, I'm in service of the person. I love just to sit down with people, say what they're working on, see what problems are running into, because it's never for me, you know, it's never the answer isn't always like, oh, positive connect, just throw it there. You're done.

Early on, when we deployed Posit Connect, there was, I appreciate everyone's response, everyone's desire to maintain cost control, but it kind of gets a little bit too much. So people are like, oh, I don't want to deploy onto this platform, because it's going to take a license and that's going to cost money. So they see the dollar and they just kind of take it like personally, like, oh, that's going to come out of my paycheck. And I think people, when you take away that pain, so aren't my decision, I made an executive decision at the time, Posit Connect, the moment you come in, you get a license, don't ask me. And that's removed the friction for not, not the user who's coming in, because they don't ever really care. Oh, if I access this, but it's the publisher, the publisher is like, who am I going to get this out to? Well, if I have to get this out to people, then I have to care about the licenses and I have to care about the security. No, you put it on our platform. You don't worry about licenses. You don't worry about security. I've made the decision to onboard people for quote, unquote, you know, frictionless, you know, and we already have security figured out.

So if you kind of show that you get to focus on the fun things and you deal with quote, unquote, boring stuff, people gravitate and people really come to you as kind of an enabler of things versus being kind of a restrictor of, of tools.

I really like that being an enabler, not a restrictor. That's something that, um, John, when I first met you, that you said was like, make the right thing to do the easy thing to do. It's also something that Elliot Murphy said, who's a VP of engineering here at Posit that like, if you want people to not be constantly like circumnavigating the barriers that you've set up to protect them and protect the company, then make the easy thing to do the right thing to do. And then people won't have to create shadow it or, or get around your things in creative ways because they will be able to do their job without having their hands tied.

Documentation habits

Do you block time each week for documentation? Do you set aside time for that?

I believe it or not. I don't. I don't either. Does anybody tell me in the chat?

You know, so I, so there are times where I think, you know, it's good to have strategic thinking, like getting things done methodology. I think it's wonderful. And one of the things in getting things done where, you know, it's a really kind of a framework for like freeing up your brain, cause you got a lot of stuff stacked in there. One of the things that getting things done says is like put time away to plan out your week. And I love that. I think documentation's a little bit different. I think I build it into the process where someone asks me, I look in the handbook, is the handbook answer it? No. Okay. I will answer it and I will update the handbook. I accept it as part of my debt cycle. I don't want to treat it as a separate thing because it kind of comes like technical debt.

Making your doc, making the documentation easy to write, put it in, doesn't have to be perfect, right? Just put something in there. And if someone comes in and says, okay, this, John, can you explain this a little bit? I'm like, oh, you know what? Someone's reading that. Let me expand it a little bit more and make it a little bit more human. For me, I don't put time away. I just put it in. If it's rough, that's fine. It's not writing a novel. I'm not writing something that's getting published on the great internet. It's just for internal use. And if it's kind of rough, it's okay. And I can refine it later, but at least someone has something to jump off of. So I think that's a long way of saying, for me, I don't, because I just, I made it a habit just to write and do it right away.

Driving Shiny adoption among engineers

It has to do with your approach and techniques that you use to getting people not only to use, especially with like oil and gas engineers who might earlier on have been more reluctant to use things like R and Shiny. What are some of the techniques that you use to maybe get them to trust the tech and some of these technologies that help them do things a lot faster?

For me, so if I reflect on trying to get people on, I think you the biggest thing was empathizing with the pain. You know, there's a lot we're all trying to solve our own problems in our own way. We're very we're in the weeds on our own stuff. And sometimes to look at what an engineer is doing and you're you know, you're seeing they're making great stuff. I want them to focus on that. So for me, I think it's a little bit more of a service, right? Like when I say service, I mean, I'm in service of the person. I love just to sit down with people, say what they're working on, see what problems are running into.

Here's a great example, you deploy something on Streamlet, everyone, a lot of people love doing Streamlet here. And they eventually, you know, you scale up and then you run into some issues of Streamlet because you're trying to do stuff and you're sitting with the problem and empathizing with people, for me has been probably the biggest changer. And then on top of that, because, you know, you've done the right thing. But what is the right thing, in my opinion, it's making sure that you're deploying on a platform that is scalable, that you don't have to worry. You know, there's there's a lot of things that if you take the worry out of people, then people feel like, oh, you know what, talking to you has given a great sense of freedom.

Here's a good example. Early on, when we deployed Posit Connect, there was, I appreciate everyone's response, everyone's desire to maintain cost control, but it kind of gets a little bit too much. So people are like, oh, I don't want to deploy onto this platform, because it's going to take a license and that's going to cost money. So they see the dollar and they just kind of take it like personally, like, oh, that's going to come out of my paycheck. And I think people, when you take away that pain, so aren't my decision, I made an executive decision at the time, Posit Connect, the moment you come in, you get a license, don't ask me. And that's removed the friction.

So if you kind of show that you get to focus on the fun things and you deal with quote, unquote, boring stuff, people gravitate and people really come to you as kind of an enabler of things versus being kind of a restrictor of, of tools.

Overcoming fear in meetings

Do you have any advice on how to overcome fear and feel confident enough to ask questions or voice opinions among coworkers who come across as confident and have a lot of opinions in the daily standup meetings?

I think there, well, this is a tough one. A standup meeting where someone has strong opinions. I think, honestly, there's a lot of people. I also always have to remember, there's always a lot of smart people. And every time I kind of get in this mode, like I kind of feel it now in my back, right? There's someone really smart and they're saying something and it could go both ways.

I have found, and I've gotten feedback on this, that, John, you need to be more brave and just talking, just picking up the phone and calling. We can be really, we can protect ourselves and our opinions behind the screen, behind the office, even being remote to a certain degree. And I don't think all of that is bad. I think we're personally, I love working remote and I think it's great, but you have to know when you have to break down the wall to make it human again. And if it's just this kind of one-way communication and it's not a two-way kind of collaboration, I think you can feel like that.

So for me, the thing I'm always working on is just picking up the phone and calling or FaceTime or something like that. That to me has been, it's the hard thing to do. But I find that most of the time, the person on the other side, like I'm getting a lot of wonderful questions now about, well, is Posit Connect ready to do things in production? It has the stereotype of being this. I'm like, actually, no, it's not. And let's talk through it, right? And I think that helps, but you only, you can't get that just by spamming and pushing it out. That works to a certain degree, but you, you got to call, you got to call. And I say that to myself every day, I have to call.

Staying current and choosing what to learn

How do you pick the topics to learn more about? LLMs have displaced ML as the hot topic, but there are still many advancements in traditional ML to keep up to speed on. In a related note, how do you pick how far to go down the rabbit hole with new topics that you're trying to stay up to speed with?

So the question was, I think the way I interpret this, Alex, is how do you know what to prove? So I've been honestly very privileged. I've been lucky to be able to follow a lot of the interests that I have. R was one of those interests. I always like to say, I personally don't overthink about what is best for my career. I kind of think about what interests me.

So let's talk about dragon boating, for example, in this data. And let's talk about how it works at Waze back. I always wanted to do an iOS app. That was something I've wanted to do for the last three years. So I started to pluck away at it. And then dragon boating came along. It's like, oh, what could I do? So I'm like, well, I'm interested in dragon boating. I love coaching dragon boating. I love the mechanics of it. But how could I apply a new technology, a new way of thinking to it? So one of the things I smashed together was, well, could I use a vision model, i.e. taking stuff from my data science background, just taking my interest in dragon boating and this new thing, and try to combine it all together.

And for me, it's been, I think for me, being interested in pursuing something that I cared about has been much more fulfilling. Because then what happens is that you talk about this stuff, right? You talk about, oh, you know, I tried this iOS app thing. You know, I can see the limits of it, but vision-wise, it's good. And then someone's like, oh, at work, hey, we want to deploy, we want to start deploying data to iOS. Oh, we want to start doing vision stuff here at the company. And you're like, oh, you know what? The interest I've had can kind of propel me into the next problem that the company has.

So I think I've been privileged, I'll be honest. Being able to follow what interests me has been very fortunate in setting up the next steps of my career. How deep you want to go, I think I'm going to sound like an economist, but it's really about that marginal benefit. How much pain does it take to get to that next level? And is it worth it? You know, for me, you know, I could talk about my Georgia Tech master's attempt. You know, that is kind of that same thing. You do some learning, are you going to get more benefit? If you do see it, keep going, but you'll know when to pivot. And that's okay to break earlier than the goal, because eventually you'll find your way through. That's kind of my way of saying just follow what interests you, because this is your life.

That's kind of my way of saying just follow what interests you, because this is your life. You only get one.

AI experience and solving business problems

Is it truly necessary to have deployment experience when you are looking for a job? Or if you're looking for a job in AI, is it truly necessity to have like deep AI and LLM experience? Or is it still possible to learn on the job?

I think that's a tough question because I think there's a lot of gatekeeping right now on some jobs that are quote unquote AI related. You know, and I'm going to take a step back. I actually don't know if it's the tool that is going to get you hired. It's how you communicate the business problem. You know, we all know how the intricacies of a gear on how a model works or how a large language model thing works. We kind of know the gears about it, but it's never about that. It's about solving a business problem.

So for me, like I was just speaking to an HR advisor and we got excited and we didn't get excited about the AI thing we're going to use and how to do it. But it's really about what are we actually going to solve to make people's lives better? Everything else that you do supports that mission. And if you put for me, if you put the mission on the tool, the ML, the model, the technology of choice of the moment, I think it's going to put you on this treadmill where it's really difficult to keep up. But if you are solving the problem, you have suddenly many, many ways of exploring that. So for me, that's been how I approach it. Not so much about the tool, not about the gatekeeping. I always come in with saying, how can I solve your problem? Not your problem with this. With this exact tool.

So maybe the advice could be if you have a portfolio or a resume highlighting the ways in which you solved problems and what the business problem was and what your solution was with the tool as an incidental like this was just the tool that I used to do it.

Okay, fantastic. Well, it's the top of the hour. So I will say thank you so much, John, for joining us. I hope that you had a good time. It was a wonderful session. I love the questions here. I love that I couldn't keep up with the chat. There's so much. It's so infectious. I love it. This is why I'm grateful for everyone to asking such thoughtful questions and seeing everyone's face. It gives me the energy to go back and be a little bit more productive. I'll try. It's been a high right now. So thank you.