HH73 - Happy Hour #73.mp3

Harpreet: [00:00:08] Hi. Yeah, what's up, everybody? Welcome to the Art of Data Science. Happy hour. It is Friday, March 18th. I'm back. I'm hosting the session today. Shout out to everybody who took care of the session summary over the last few weeks, Antonio and Ben and Mark. You guys were such, such, such a help if I miss anybody who helped take care of it. My apologies, but I appreciate you guys so much for for taking care of the office hours. I'll be back for the next couple of weeks. But you won't want to miss who's going to be hosting in April. So in April we've got Katie St Lawrence taking over, we've got Ken G taking over and Kiko is going to be taking over. It will be a fun time, so definitely be sure to to to come and hang out. I'm still waiting for the insurance to do some work on my basement and get me back set up. So I'm still without any place to record, without any recording equipment or anything of the like. So hopefully you guys don't mind my low quality audio either way. I'm super, super excited to be here. Shout out to the people that supported the show last week through Buy Me a Coffee. Rest assured, I'm not using that stuff to buy coffee. It's being used to help buy new equipment and pay for the editing costs of this podcast because that is getting a little bit out of hand.

Harpreet: [00:01:32] Shout out to Gina and Makiko and Naresh who helped out. As always, there's a link right there in the podcast that you can podcast show notes that you can use to support the show or just say hi. So thank you for doing it. I'm happy with that too. By the way, if you guys have any things that you want me to talk about or anything for me to discuss, feel free [00:02:00] to shoot me a message on on my email address. The Theartistsofdatascience@gmail.com kind of kind of going through a bit of a dry spell for content ideas. I think that's because just a lot of stress going on, a lot of stuff happening. Hopefully get a chance to tune in to the podcast episode that was released today. Got the one and only Joe Reece released that episode today. So I had a good time with Joe, so do check that out. This was a previously recorded episode. We actually were. All extras are previously recorded obviously, but this one was streamed live on LinkedIn. So if you missed the live stream, this would be opportunity to check out the the show itself. So huge shout out to everybody that is joining us, Russell, then Auntie Gina Makiko and the rest.

Harpreet: [00:02:47] Jacob, I'm so excited to have all you guys here, man. Thank you all so much for being here. Huge cheers, everyone drinking some fermented grape juice. So let's kick this session off with a question. So I'm curious. Right. What? Where do you go after data science? If you wanted to make a pivot out of data science, where do you go with their skill sets? I'm asking this question just because I'm curious to see what other options might be available to people. Some people might enter the field of data science and they think that this is exactly what it is doing. They're like, Oh yes, I'm going to be building models. I'm going be doing all this amazing predictive analytics and predictive modeling and all this great stuff. And then they go to work and it's essentially just analytics and dashboards and things like that, which maybe doesn't align well with the thought it would be. And they might want to make a pivot out of data science. Where would they go? What are what are some options that people have? Let's go to Venn then. Let's go to Mexico. And Russell, I'm going to put you on mute for the time being because there's some coughing coming from the keyboards there. Van Gogh for.

Speaker2: [00:04:00] And [00:04:00] told that I'm highly qualified to be a Walmart greeter. So I've got a backup plan just in case I ever have to have to do something else. I think from a data scientist, you can go straight to CEO, and I think that's kind of the interesting pivot that you can make, is that you can go from data science to pretty much any of those C-suite leadership roles and be successful because data is so important. So many business models are now being built on data in one way, shape or form. They're being built on machine learning or rebuilt on machine learning. And so you can I mean, if I was going to, like, apply for a real job someday, which required me to put a shirt on, that's nicer than the one I have. I would. Yeah, I'd probably try chasing a CEO gig down.

Harpreet: [00:04:54] That's interesting perspective like straight to CEO. How much experience would you would would a data scientist need to go from let's just say. They're beginning of their career, maybe their first role till the CEO role.

Speaker2: [00:05:10] I think you need a lot of experience on the leadership side because you're leading an organization. So you have to have gone from I led a team to I led a business unit or group to I've led an organization of cross-functional different mentalities, different objectives. I've handled investors. I've handled the board before. But it's kind of funny. If you come out of a startup, especially as a data scientist, you've done a lot of that. It's insane how much? As soon as people realize you can do data, you can present, you can do the visualization side of it. All of a sudden, everybody wants you in on investor pitches and high level meetings, and any time that they have to present any sort of data, they're all, Hey, let me send you an invite to this. And so you get some pretty broad experience [00:06:00] after like four or five years working in a startup, you've pretty much gotten most of what you need from the investor side and from the the external facing side of it. And if you run an organization for three or four or five years, you can obviously you're not running, you're not going to run Costco or something like that anytime soon, but you could go into a small midsized business and be successful.

Harpreet: [00:06:24] Thank you very much. Then another point of shirts that I'm trying to get to a place in life where all I wear is either black or white T-shirts and that's it. Zero fucks given no matter where I go. Shout out to Danny. Man the house. Slim. Slim Danny. Split in half. I wrote that secretly this morning. That absolute love that. Good to have you here, Danny. So the question that we're just kicking off with, it's. It's. You know, let's say you broke into the science of the thing for a while and kind of figuring out that, okay, this might not be the right fit for me. After all this time and energy and effort to break into data science now come to realization that it might not be for me. Where do we go after that? That's the topic we're talking about. So I'll give you a couple of minutes to think, Dani. We'll go to Mexico, then we'll go to Danni, and then Russell's got some comments. So we'll go. Danny Russell By the way, those of you tuning in, either on YouTube or on LinkedIn, if you guys have questions, go ahead and put them right there in the comment section. I'll be sure to add them to the queue. Go for it.

Speaker3: [00:07:26] The three that come to top of my head are. Content creator because we know people here who are content creators, but seriously, they're like, I think it's funny because. There is there's a couple of different gaps, right? There is how data professionals understand data science, machine learning. Right. And there are gaps there. Right. Then there's the next layer of air gap between people who are in tech and [00:08:00] versus people who are out of tech. So I think a lot of content creation sometimes it's like we're creating content for people like us. But I think there is like a huge gap in opportunity for creating content for people who are not like us. So for example, like my parents, they are all bringing in a world that is increasingly like data science, machine learning driven. They get these credit card offers and my dad swears that they picked him for it. I'm like, No, Dad, it's you fit the persona of a schmuck. It's not that they point at you as a schmuck, right? You know, but like the thing is, there is something to be said for the fact that like there are increasing sort of structural inequalities, right? It's not just inequality in terms of wealth and income, but it's also an understanding of like kind of the the systems for that ruin people's lives.

Speaker3: [00:08:53] So I think there is a huge opportunity there, especially for people who are sort of like multi faceted and creative folks, which I think everyone on this call is. Frankly, I think the other second area of opportunity is prog management. That's a really, really, really big thing, you know, and go and that kind of aligns with like vinz you could become a CEO definitely. If you want to create like an ML or a AI driven startup or an AI like native product, and you're good with talking with people. That's really valuable because a lot of traditional engineers, they have they operate in like one serve world or one sort of epoch, shall we say, which was like pre data and pre ML. And that sometimes influences product decisions in a way that is bad, you know. And the third part is, is part is pivoting to more of an engineering role, which is what I did. I pivoted to ML engineering and then I pivoted to Ops. Now it [00:10:00] seems like I'm kind of stuck between like an ops and like data engineering and infrastructure, but it's super valuable because at the end of the day my internal users are data scientists, so it's helpful to have that mindset of who are my people that I'm walking in their shoes to create better experiences.

Harpreet: [00:10:24] Kiko. Thank you so much, Danny. Let's hear from you. By the time I get to see you again, as always, it's been quite some time. So it's a pleasure to chat with you again.

Speaker4: [00:10:33] Everyone. Nice to see. Nice to see you all. I just woke up, so I'm sorry if I. If my thoughts don't sound too coherent. From. So the question originally was like, you enter data science, you kind of come in, you realize that it's not a good fit for what your career aspirations are or where you want to go in the future, for example. Right. So I've worked with quite a few people who've had the same experience where they came in. Maybe they're doing data science in terms of the machine learning, predictive models, putting things into production, coding a lot, and they didn't. They just didn't like it. They might have been very good at it and good at procedurally doing the thing and solving problems. They enjoyed solving problems, but doing all the like fitting models and. Doing all of that sort of stuff just wasn't really going their way. So there are a few different options, right? So one of them is. We can all agree that having these sorts of skills are really valuable regardless of whether you whether you physically code the thing and apply them, or if you're leading projects to implement some sort of machine learning to make a feature smarter or make your product more data driven or anything like that. I feel that having done a stint in that technical, technical area of data science is always going to be, well, let's call [00:12:00] it career capital for whatever you want to do in the future.

Speaker4: [00:12:03] So if you did want to go down an engineering route at least, or if you're doing engineering for data scientists, you would know exactly what the data scientist would need. And likewise for the business side, if you wanted to go and lead projects and manage teams of data scientists, you would also know what the data scientists need. So it's kind of like I felt in earlier in my career that I wanted to be as technical as possible and at different points. I got to the level of frustration where I was just like, Why are we even building a model for this? Do we even need a machine learning model? Why don't we just apply some simple heuristics and be done with it and move on to the next problem? So at different points in time, whether. Whether you think you will not get bored of training models and doing data science and visualizing data and explaining things to stakeholders and everything, I think everyone gets to the point where either you you've had enough of it or you really question whether you want to continue doing this to the next 5 to 10 years. So I think it's quite natural to be thinking about it and it's a really great question.

Harpreet: [00:13:09] Often. Thank you so much. I appreciate that perspective. A lot of good thoughts coming in. A lot of good. A lot of good ideas. Happy, too. Happy to hear this. Let's go to Russell. So just for everybody that just joined in, I just kind of kicked off the question and I'll explain why I kicked off this question after we're done with this discussion. But the question is, let's say you did your thing in data science for a while and now it kind of maybe you realize that this isn't for me, or maybe you realize that this isn't all I want to do. What comes after the science is kind of the gist of the question What comes after data science? What do you do if you wanted to pick it out of data science? So let's go ahead and hear from Russell because he had some good thoughts on that. I'd like to go to Joe Louis after this. Shout out to Kenji in the building, which he can. And if you guys have questions on LinkedIn or on YouTube [00:14:00] by on, feel free to drop the questions and the comment I will get to them. Let's go to Russell.

Speaker2: [00:14:07] Thanks. Yeah. So there's been great comments so far. And Mickey kind of touched on what I wanted to say, which was project sorry, product management. So depending on what the consumption product is, I want the business or any of the audience is seen managing that and being a pivotal sorry, a pivotal translator and catalyst for the data and how the consumers consume the product and the data science team or teams that create it. So if the audience needs a change or they're not understanding something so well, it's often very difficult for them to translate that into a message that the data science team can.

Harpreet: [00:14:53] Action.

Speaker2: [00:14:54] Efficiently. So having some kind of catalyst and translator in the middle is really invaluable. And that's kind of the space that I've been in for a while. So, you know, I'm not a I'm not a data scientist. I'm probably was more a data analyst than a data scientist anyway in the first case. So I, you know, I use data science in my day to day job, but I definitely want to classify myself as a data scientist. But if I had to pin myself down to a single role, it would be that data catalyst, data translator and it really is key and it's very rewarding often. So being able to translate that frustration or thoughts for innovation that's coming from the consumers to something that the team can digest and implement quickly.

Harpreet: [00:15:41] And efficiently is is very.

Speaker2: [00:15:43] Good. Other than that, a wildcard solution may be sorting out Zoom backgrounds from Akiko. She's always on point, and if there's ever an opening there, Mickey, go put me down for a please.

Speaker3: [00:15:57] Absolutely. I have no idea what service I am [00:16:00] offering, but as long as it's legal and I can monetize it, I'm there. I'm there for you. I'm here to serve.

Harpreet: [00:16:08] A I driven background for for right. You log in and zoom. The camera is going to read your facial features. It will understand your emotion and then appropriately give you the background. So, Joe, do you want to tackle this question for Cozad? Because I know you kind of made that pivot yourself as well. Then from Joe, we'll go to coast to.

Speaker2: [00:16:32] Oh, it was a question basically like what's next after data science?

Harpreet: [00:16:36] Yeah, pretty much. It's kind of headed in that direction.

Speaker2: [00:16:39] Yeah, nothing's next after data science, unfortunately, your life's over and. Sorry, I'm just kidding. No, I mean, I think it's just. It's just a progression, really. I view it as kind of like, you know, like anything else in life. Like, it's it might be a detour, it might be a stopover, but at some point, you know, you evolve and you kind of move on to other things. But I would say it's also very additive. So unless you absolutely hate data science and never want anything to do with it, you're going to probably pick up some skills and knowledge that you can use for whatever's next. So, you know, tech in general is an interesting field where I see people either they're in it for a long time. You know, I've been in data for a long time and to me it's never felt like work. I just kind of like it. I nerd out in the stuff in my spare time and it's, it's, it's always been. But other people like I know, I know a lot of people like one one guy I know he he's kind of he peace out. He's making furniture now for the most part. Like he just doesn't want anything to do with this industry.

Speaker2: [00:17:43] So I think it's just one of those things where either like you kind of transition to some sort of like, you know, adjacent role where you can apply your skills to a domain, you know, you go do something else, maybe raise sheep or something, I don't know. So it's just totally depends. So but [00:18:00] I don't I don't know that you can't really say like, oh, there's, there's, there's nothing left. Like, this is kind of the end all be all like this is the only thing you can do. I think it's the cool thing is you got some really cool skills. So when, as Akiko says, pivot in the chat here, it's, it's what I mean if you look at I think the more interesting people I know and some of the more successful people I know, it's not like they set out to do anything in particular, right? They just applied one skill set and have to move into something else. And then then throughout your career, what you're going to find is like when you look back, it all makes perfect sense as you're going forward. It makes zero sense and it's probably very scary in that way, but that's fun.

Harpreet: [00:18:36] So yeah, I guess we got the most transferable set of skills. It's not like being an accountant, right? Or all you do is spreadsheets and stuff. We had a good set of transferable skills. Let's go to Kosta and Kenji. And by the way, if anybody has questions either here or on YouTube or LinkedIn, please let me know right there in the comments and I will add you to the queue. Go for Kosta. So this kind of throws me back to something we were talking about last week, right where I think someone asked about how do I know that a sense is right for me or something like that, right? To me it comes down to Why are you doing data science, right? Data science, machine learning. These are tools, right? I'm I'm an engineer. I'm a robotics engineer. I've learned mechanics, I've learned electronics, I've learned software. And now a bit of machine learning. And I'm using that to do stuff right. It's a means to an end. For me personally, it's about the mission, right? For me, the mission is robots. I need to see robots in the real world. Part of that is solving the vision problem, the perception problem, right? If I get bored of this particular set of tools that I'm working in, I change the part of the problem I'm working on. Right? It might take a bit of a shift and maybe a bit of a jump backwards, but I can go back to my electronic skills, go back to the mission itself.

Harpreet: [00:19:53] Now the question becomes, am I a board of the mission? If I'm bored of the mission, that's a much bigger question. Right? [00:20:00] And then I go, What's my new Y? What's my new reason for doing stuff? I think I've seen a lot of data scientists and developers in general that are in this space because the space is big, right? There's been this big push for data science is this huge gap. We need to get loads of data scientists in. And I've seen a lot of people that'll just turn around to me and say, Oh, I'm in data science because I love working with data. Which is a fantastic answer, right? Because you're doing what you love. But then there's a higher step of guidance. If you have a mission that drives why you're working with a particular toolset. At the end of the day, let's not forget that these are tools, right? Like, for all, you know, scenes and transformers might be completely obliterated by the next thing around the corner. We just don't know what's going to come along. Right? So I try not to get too aligned to, oh, this is my tool kit and I love my tool kit and I am Man with the hammer kind of thing. So I think ultimately that's the question we've got to ask ourselves is what's our mission and why are we working in a particular problem set? Right. It's a part of that mission. And I guess sometimes we get disenchanted by it as well.

Harpreet: [00:21:18] I'm really curious and I heard this as a rumor, so I'm not sure if this is true or how true this is. But Joe Redman, the guy that came up with Theola, right. Eventually, he apparently kind of said, yeah, I don't want anything to do with object detection because of the way of how it was being used. Right. So all of these networks, all of these advanced technologies that can be misused. What happens if we see that? Like, what if I see our robots are in the real world and they're all being used for destruction? I don't want anything to do with it. Right. That's when you've got to do a bit of self assessment and really figure out what you want in life. That's a much bigger question than just, Oh, I've got these transferable skills. It might literally mean [00:22:00] that I go do something completely different. I might look into my music or something else, right? So it does come down to Why are you doing what you're doing right? If you don't know why you're doing data science and you're just doing it for the sake of it, you're more likely to get bored and find something else, if you know what I mean. Kosta, thank you so much. Let's go to Kenji. By the way, if anybody has questions again, let me know in the chat here on LinkedIn or YouTube, I'm monitoring it. All came. Good to see you again, man.

Speaker2: [00:22:30] Yeah, good to see you, too. First, I love all the backgrounds that are going on right now. I prefer my non virtual organized chaos in the background. But you know, to your point, what's after data science? This one's really interesting. I think for me, before data science, I was always interested in entrepreneurship and I got interested in the data field because I saw there were a lot of parallels between data science and entrepreneurship. Right. You're thinking about a situation quantitatively. You're building things, you're collecting information and you're producing either a product or recommendation or whatever that might be. And I think the natural exit to data science is also something entrepreneurial. This might have been said already, and I think Joe touched on it a little bit, but like this skill set that you cultivate the the mindset the way data scientists think. I think it really lends itself very well towards building a product, creating a platform, whatever that might be. First, you have the technical skills and you have the ability to scope and understand these problems, which to me is like two sides of the equation. The other thing you need is like knowing the right people, which I think everyone here, we're all pretty good at connecting with each other and forming a network. And then the last one is financial resources, which frankly the data science profession also affords. So it's pretty you know, it's pretty interesting to me that that [00:24:00] to me, what you have most of the tools that you would need. I can't think of too many other professions that set you up as well as data science does, to pursue something completely on your own entrepreneur. It could be a data related product, but running an organization, starting something, you know, building a product, building an MVP and and accelerating it to me again, is this very natural next step.

Harpreet: [00:24:25] Absolute love. I can't thank you so much if you guys are listening to this on the podcast posit. Go back and make sure you watch the YouTube video because these backgrounds are way too much. These are amazing. Yeah. So Mexico is talking about content creation as a next step in data science. Right. So for me, you know, I haven't been a quote unquote, practicing data scientist and like the traditional sense for the last six, six, seven months. So I move to Comet and my role at Comet is essentially what's called developer advocacy, right? And this is a really unique, I think, space that's well suited to me personally, where my specific knowledge fits in quite well there, because it's a combination of not only the data science aspect of stuff, but it's product, it's marketing and it's engineering all kind of combined and rolled into one type of role. And I mean, like. Job is to create content about our product so that it eases people's experience using the product. So there's that aspect of it and then kind of raising awareness about the product and the marketing aspects of it. And I'm really been really been fascinated by this role of this developer relations, developer advocacy type of path. It's been like an obsession of mine. And this is going to be something you guys are going to hear me talk more and more about quite a lot. It's also a quick, quick announcement like I'm leaving Comet as next week.

Harpreet: [00:25:57] My last week at Comet is. On [00:26:00] March 25th and I'll be going over to pachyderm, where I'll be stretching my skills in a completely different direction. It's still MLA ops oriented, still ML ops related, but it's more on the data engineering kind of aspect. But my role there is going to be developer relations and this type of role, I think such a high leverage role because to be successful in this role, not only have you had to have been a good practitioner and you have to demonstrate that you've been a good practitioner, but you have to learn to build that, to learn to sell. This is one of those unique opportunities where you get a chance to do a little bit of both. And the fact that this type of role crosses product engineering, marketing, and it's high on content creation. It's it's fun. It's interesting. So I definitely urge you guys to look into the developer relations or developer advocacy type of path. That sounds interesting to you. I've been. Knee deep in or neck deep, rather, in study and research around this field. Through here, we do a lot more content around that in the coming weeks. But yeah, like. It's been a great ride at Comet, actually. We enjoyed working there, had a great team. It was a super tough decision, but ultimately there's some intangible facts that that led me with Pachyderm, one of which was that it's been my dream to work at a Silicon Valley startup that's backed by Silicon Valley investors.

Harpreet: [00:27:29] And I grew up in Northern California, so that has always been a dream of mine. And so that's happening with with pachyderm. And that just that to me just means a lot. And yeah, I feel like I'll be able to have a huge impact there. But why did I make that change from individual contributor, individual practitioner into this type of role? Because I feel like the impact I have as just one individual in one company doing stuff for that particular company is not as impactful as it could be [00:28:00] in a role where I'm kind of guiding the industry in a sense, and being a thought leader for the industry and helping develop and set best practices for machine learning and ML ops for the industry at large. I feel like this type of role allows me to have more. More impact on the future of developer relations, developer advocacy. That is the path I am going on. Looking to see if there's any questions coming in. There's one question coming in on LinkedIn. Somebody is asking why it's hard to get your first internship in data science. Is that a question that anybody wants to tackle other than it just is? Yeah. It does not look like that.

Speaker5: [00:28:46] Let me add something new to what Harpreet just said.

Harpreet: [00:28:48] Yeah, yeah. Please, please, it. Just go to Kevin posted. Yeah. It just is. Let's go to ten and post up. And by the way, like I said, you guys got questions. Please leave my thing in the chat, in the comment section. We'll be sure to get to them. Can go for it. Then posted the.

Speaker2: [00:29:07] So I've sort of a weird two part answer. So the first I think is it's hard to do anything for the first time, right? You need proof of work, you need some. Why is someone going to hire you? Right. They want to see experience. They want to know that if they're going to bring you in and pay you money, they want to be sure that you're going to at least do a suitable job. I think the barrier for internships is significantly lower than than a full time job. You're looking at maybe three months versus potentially five, ten, 20 years. I would push back on that, though, and I'd say that. Some you know, for some people, it isn't that hard to land their first internship. And I think it's really important to study those people to figure out what they're doing differently than what you are, to be able to learn those things. So I think the first thing that a lot of people who are really [00:30:00] doing well and learning these internships are doing is they're finding experience in other places. Maybe they haven't had a previous internship before, but maybe they've done some volunteer work, maybe they've done research, maybe they've done these other things that someone can look at and say, okay, well this is a corollary for internship experience. This is real experience that other people who are also applying to this internship don't have.

Speaker2: [00:30:22] Why wouldn't we look at that? Another thing is a portfolio, right? If you have a really strong portfolio compared to everyone else in who's applying for that internship, you're probably not going to have a hard time getting internships because you've already differentiated yourself. Know another thing if you're like, Wow, you know, it's my freshman year in college. I barely know that much programing. I want to use this as a learning experience. Communication and reaching out is another way you can differentiate yourself if you're using different channels to get in touch with people. If you're connecting with people on a different level, maybe the content of your conversations or what you bring to the table on that end, you're also probably not going to have too much trouble getting an internship because you're differentiated in that sense. So I think about rather, rather than thinking about what you're doing the same as other people, think about what you're doing differently from them, what makes you stand out, because at the internship level, a lot of the candidates are very similar, right? Like most of them haven't had internships, most of them have have had an X, Y, Z classes, and they don't have that much to show for it. Just a small differentiator can make it from very difficult to relatively easy.

Harpreet: [00:31:32] Ken, thank you so much. Schachter Ben Taylor in the building. Good to see you, Ben. Well, at least inside the cavity here. Let's go to the question that we're that we're at right now. Why is it so hard to get an internship at Data Science Southcoast lab? We will go to Danny. And if you guys have questions again, I think Gina has a question. So after Costa and Danny, go to Gina's question. Cool. So I see kind of four [00:32:00] parts to this, right, are four separate factors that are creating pain points and getting internships and first jobs. Right. I think I'm more focused on first jobs than internships necessarily. So the first and foremost thing is that that science is not as linear a a learning path in the sense that a lot of the time it relies on both technical expertize and multiple fields, as well as some domain knowledge. Right. So there's a bit of a weird thing where if you've been in a domain, let's say I've worked in medtech for like four years, right? I develop enough knowledge in terms of the problems there are to solve in that space. Right. Some of that domain knowledge could make me an expert in medtech and data science and the confluence of those two areas. Right. So over time you build that up. So it's a lot harder to do that early career. That's one can't solve it. That's okay. Let's not worry about that. The main thing you've got to worry about is what are the things you can solve and can fix? The second thing is proven track record. Right. It's it's it's hard to prove your track record.

Harpreet: [00:33:08] But arguably, I agree with Ken. It's easier to prove your track record in something like data science than it is to prove your track record with mechanical engineering or electronics or robotics. So something else like that, right? I can't build a whole robot on my own. I might be able to build a little toy, you know, a tiny robot or it's a big investment cost to go get, you know, 3D printers and stuff like that, to show my mechanical design skills right to fruition. Right. I might be able to get CAD software or something like that, but with with data science, there's data available for free. There's tools available for free. You know, you pretty much got 300 bucks on GCP and it's not terribly inaccessible, right, in comparison. So, yeah, I don't think there's huge barriers to entry in that sense. Right. So go ahead and build that portfolio. I think that's that's the big thing, if that's what you're missing. But then there's the other two sides to it, right? I think a lot of companies are still [00:34:00] in a space where they're uncertain on the value provision that data science poses for them. Right. So a lot of companies are keen to hire their first hire as like a senior ML engineer or a senior data scientist or someone like that who can come in and they don't know what value is actually going to be added. So a lot of companies, while they might have data science rolls up, they're kind of still only looking for those senior people who can kind of prove out the value of the role and the value of that kind of field for that company.

Harpreet: [00:34:30] Right. So I don't think there's as many lower entry level data science roles as there are the senior space, because the maturity of companies data readiness. And the third thing is kind of along that lines is a lot of companies that are data ready don't necessarily have the built up maturity in their data team right now. In order to have an intern or a entry level engineer, they need guidance like not from necessarily even a data science perspective, but a how do you operate professionally? How do you deal with clients? How do you deal with internal party office politics and stuff like that? Right? Like there's so much that you learn on operating, particularly in large corporate environments, right? But even so in startups, right. How do you know boundaries in terms of this is how work should be versus I'm getting exploited over here, but that does happen. So essentially it comes down to is a company ready to take on an intern or a grad and actually give them adequate experience? Right. So that's the other side of it. It's that maturity of the companies is there. And given that this whole data science as a field is still at this stage, it's quite an early, early career stage, right, that there's career maps. But this is what a graduate should have and skills should have. Isn't established industry wide, right? This is what an entry level should [00:36:00] have. That's not a that's not established because that's not established. University degrees aren't 100% sure what to aim for in terms of saying this is what we're going to teach a data scientists and expect them to know graduating from it.

Harpreet: [00:36:14] Right. So the degrees that we have now, there's very few bachelors of Data Science, there's always masters of data science. I'm starting to see now, I think Sydney University and a couple of others are starting to offer bachelor's of data science here in Australia, but we're only starting to see that coming through the woodworks essentially. So I think over the next four or five years what we'll find is it'll settle to like the mean where we go, okay, this is kind of what we all expect industry wide of entry level data scientists. So your bachelors degree courses can focus on that, give them a specialty on there, and then you'll start to see that pipeline smooth out. Right? I think it's early industry stage is part of the essentially part of the problem. And in that if you can make yourself stand out. With a great portfolio initiative is what gets you ahead on that front. So yeah. So thank you so much. Let's go through it. Let's go, Danny. More than. And then some great questions coming in on on LinkedIn. One of my former colleagues, Kyle and has a question was that he helped me help me do some good workout. So I can't wait to get the kind of question. But we'll go to go to Danny, then we'll go to Eric. Then we got Gina's question that after Gina's question, we will go to Marcelo and then Kyle's question both coming in from LinkedIn. Awesome.

Speaker4: [00:37:35] Thanks Harpreet. I'll try and keep it short because we have many things to talk about. I think if you're having trouble trying to find a data science internship or any sort of internship role, nine times out of ten, you're probably looking in the wrong places, whether it's the, let's say the party on the other side wants someone who can instantly start doing work and producing value, or [00:38:00] if they're looking for people to actually mentor and coach and grow and kind of put a bit more trust in their skills as well. So really depends on this supply and demand dynamic on what people are looking for. So I think that's one thing. The second thing is probably everyone has trouble trying to differentiate themselves and differentiate yourself doesn't necessarily mean like have the best. Portfolio will Kaggle, Kaggle results or anything like that. It's you have to stand out in a way which is like at a very high level, but in a general sense, like different to other people who are going for the role. And it has to be in a way which puts some sort of faith in the person who wants to bring you on as an intern that you'll actually be able to help or that you fit that culture. So that's another thing to think about. But definitely there's there's a lot of a lot of my focus or my value proposition that I give to all my students. And my following is essentially how do we bridge this gap between what is needed in industry? And I want to break in or if I want to do a career switch or anything like that. So I've been thinking about this a lot, but I don't think I think we'll cover it another time because there's just there's a ton of stuff that we can talk about.

Harpreet: [00:39:22] And do you do this virtual internship as well? So go ahead, plug that in and tell people about that. Where can they find more for sure.

Speaker4: [00:39:31] So thanks for the opportunity, harpreet. So I run data with Danny. It's essentially you can check it out at Data with Danny Dotcom. My plan was essentially to provide this sort of internship experience through a series of courses, essentially. And my plan was to do multiple courses covering some of the skills that we we learn, technical skills that we learn as data scientists on the job. So the first one that I taught was SQL. So my first course is serious SQL. You essentially [00:40:00] learn SQL through doing many, many case studies as realistic as possible, some of the things that we all do on our jobs. And the plan was to have another future course in Python and other stuff, machine learning, statistics, etc. But I've just kind of focused on SQL as a major pain point for a lot of people coming in, especially when you need your force to use a database. You've never used the database before. A lot of the data that we pull when we're data scientist is coming from a SQL database. So it's really, really critical to have those sort of fundamental skills. And outside of that, I do a lot of teaching for O'Reilly as well. So I have some SQL stuff on O'Reilly. I have an upcoming interview video masterclass sort of thing that I'm trying to produce with them. So that's my current project right now.

Harpreet: [00:40:49] Then you do. The serious sequel is like serious stuff, like because you got people in Docker containers and working with people and stuff like that. So that's very much closer to the real world than just kind of just doing it like I guess a or whatever. Yeah. Awesome stuff. You guys check out Danny stuff. Eric Simmons, you had your hand up. Yeah. I had a couple of quick little things I was thinking about how we started this off, talking about what kind of comes after data science with the internship thing. It's like what comes before data science? A lot of people talk about transitioning because they've got work experience or whatever. But you know, I have to think about also like what kind of internship are someone looking for and what kind of internship are they ready for? Do you really need to get a data science data science internship like in quotes? Or can you get a marketing internship and bring a data mindset? Because I would say, you know, I haven't ever had a data science internship, but I would say that I worked with data long before I worked in data. Right. Because, you know, there are lots of opportunities [00:42:00] for that. The other piece kind of, I think along with what Danny was saying, was talk to people like if someone sends me a message and a resume, I'm not I'm not going to help them. But if somebody sends me a project that they worked on or that they're working on and they ask me for feedback, I am all ears, you know, they didn't even ask me for an internship in our first interaction suite. But then if they want to talk to me about it later, that's great. That's, that's way better because then I've at least seen some of their work and talk to them and like cultivate a little bit of a relationship. People I think so much. Michiko go for it.

Speaker3: [00:42:31] Yeah. I guess there are two or three points I want to add in. There are first off an internship is not it's not a prerequisite for success later on in one's career. Like I'm sitting on some of these panels for MailChimp and these interns are like scary like, oh my God, the kids are are all right. They are doing great. From a skills and experience perspective. Sometimes when decisions are made, it's not even about the skill set of the individual. It's sometimes the consideration is literally like, are they closer to graduation and will they accept a full time offer from us? Because I think companies who have a very healthy expectation relationship with their internship programs, like they understand that it's a time commitment like, you know, we're not bringing on an intern with this understanding that the intern is going to provide immediate value. Sometimes they do like they pick a really cool project that's been kind of scoped for them and they run with it and it becomes a cool new feature and that is awesome. Or the MVP, like a new tool that that is fantastic. But it is a time commitment of the senior engineers or the partnering engineers. It is an investment. Part of it is to do the talent pipeline, but also because it just makes us kind of like better in general as an engineering community or a data science community, right? So [00:44:00] when we look at candidates, sometimes we are literally going like, okay, there are this batch of amazing people we get to only like we have a batch of like let's say ten candidates.

Speaker3: [00:44:11] We only get to pick two like we don't get to say, but could we please, please do offers for all of them and then get like five engineers? And it's like, no, no, the budget is four too, right? And a lot of times it's literally like this person is closest to graduation, so we'll be able to pull them in. And those conversations sometimes come up. But like sometimes it's not really like about the intern skill. But the other part to you, right, is if these people come back later on, like, we'll probably want to hire them. And like for me personally, I would not have been able to get in to like the MailChimp internship program. Like I look at the, the people that were we're sort of considering and it's like, wow, like I'm really not, I was not there like at that age, right? But I was able to get in as a full time engineer. And a huge part of that is because of some years of work under my belt, you know, some experience with key stakeholder management, sort of broad exposure to other areas. And a lot of it is because at the engineering level or the data scientist level or like the senior data scientist, whatever, a lot of times you're expected to kind of scope your own projects.

Speaker3: [00:45:22] So the requirements for being hired as a full time or whatever, it's just it's going to be different from an internship. So I would say, like if someone can't get like an internship, like out of the program, it's not like the death kill for like the rest of your career. That's just not really the way it works. Right? But also too, sometimes there are factors out of one's control. And the reality is that data science is a super, super popular field. And it's also, unfortunately, the dump bucket for like everything. So whether someone is interested in in data engineering or ML ops or like dev ops or even project management like, a lot of people [00:46:00] just go straight there. And sometimes like by actually choosing something different, like Eric mentioned. If you if you go into like more of a marketing or something else, it's less competitive, but you still get the same experience. And you can still brand it. And honestly, that's I think that's a strategy that if I recall, some of us on this call have used of getting into less competitive programs and then transferring to more competitive ones, because sometimes it's just easier to do the transfer over as opposed to doing like the direct line. So for me, that was the case. So it's not the end of the world if you don't get an internship, it's totally cool.

Harpreet: [00:46:42] I'm Akiko. Thank you so much. So. Shout out to Mark. Mark is in the building. Mark, you want to talk about internships so quick? Yeah. So what was the original question? I have a lot of thoughts and internships. That's very top of mind right now, but I want to make sure I answer the original question. The original question was, why is getting a data science internship so hard? That was the question, definitely. So I'm actually actively interviewing interns right now at FAMU. We have a roll up and so now recruiting is just pulling me in randomly. I'll come to work and be like, Hey, you have an interview today? I'm like, Great. I enjoy interviews. And similarly Michiko said, Yeah, these interns are scary. They're so good. I interviewed individuals last week and one of them the notes I literally wrote, I feel like I can learn so much from this individual. Please hire them. Right? So like learning so much from from an intern. Joining us is depth of skill. But I think something that I think a lot of at least early career like I didn't realize at first was cool. You have the technical skills, you could do an analysis, you may build a model or write some great code, but there's a huge difference between someone directing you to do that and you execute on that and you [00:48:00] being able to identify opportunities and scope it out yourself and just handle yourself through that process. And so for me, that's a huge difference between being early career, getting my first kind of entry level data science job and now being a senior data scientist is actually my technical skills may have improved, but what's improved way more and what kind of got me the promotion was my ability to self manage myself.

Harpreet: [00:48:23] And so interns are coming in not only with their, their kind of technical skills was there, but if you can show your ability to take a project on your own and instead of coming to me saying like, what do I do next to actually be like, here are my thoughts and I engage with that. I can provide way better mentorship to you, and that would be a much better internship relationship for that. And so some of the top things that I'm really seeing for applying for internships, at least through the pipeline when I review projects and whatnot, is your ability to clearly you can do the data that's great, but your ability to clearly communicate it to others, that goes a long way. So I'm currently so for context, my job, we're currently hiring interns right now, but we're not actively posting it because if you put a post out at 100 or 203 hundred, level, people are just going to scramble to it and it's going to make it really hard. So we've just been actively going through LinkedIn referrals and stuff like that, things that say like when people reach out to me, I get kind of two reaches out. One of them is like, Hey, give me a job. I ignore those that there's nothing I can do to help you with that. Then there's some people we're like, Hey, I'm interested in who? Mu Here's a blog article I wrote and I read the blog article and I'm just like, Oh my God, this is exactly what we're looking for.

Harpreet: [00:49:45] I go straight to my manager and give like a quick pitch on Slack. Like, Why do we need to hire this person? And then I refer them over and they get interviewed. And so thinking of other ways to best set yourself up, it's actually not the technical skills, in my opinion. I think it's [00:50:00] being able to communicate your value even if you're at an intern level that felt very rambly. But those are my thoughts right now. Actually. Mark, thank you so much. Let's go to Danny Mob. But before we go to Danny, might want to give you guys a hint of questions to come. Gina has a question about knowledge grafts and how knowledge grafts are using business. Marcello has a question about AutoML and what that means for the future of data engineering, machine learning engineering. My former colleague up priced Kyle on has a question about software engineering, starting a machine learning engineering, converging to software engineering and with a future that looks like Bhavin has a question about ten x growth for AI and ML and something about the quantum future of quantum computing. And then Anoxia has a question about what makes a good portfolio. So a lot of good questions come up in the pipeline. Make sure you guys stick around. Let's go to Danny that after Danny, we're going to go to Gina's question.

Speaker4: [00:51:03] So I wanted to add just one thing on Mark's point about releasing the job ad for who move for a data science internship. So I had a really good buddy working at Amazon who had a similar sort of problem. So what they did was they put up two roles side by side. One, it's exactly the same position. One was a data science, entry level data science position, and another one was an entry level by engineer position. The sort of skills that they need were exactly the same. Only the title was changed and he got absolutely slammed for the entry level data science position. But there were probably 10% of the volume was going for the BA engineer role. And I think what a lot of entry level people coming into our industry, what they don't realize is maybe the bi engineer skills of having a lot of SQL, being able to do data visualization, build some pipelines, communicate the value that you're doing, try and solve some business [00:52:00] problems. That's the thing that people are looking for. We're not necessarily looking for people who can build crazy models or know all the deep learning algorithms and other things like that. But deep learning is really cool and all, but just something to to throw in that with it.

Harpreet: [00:52:16] Awesome. Thanks so much, Danny. Let's go to Gail's question.

Speaker3: [00:52:21] Hey. So thanks a lot. We do have a lot of questions in the chat, so I'm willing to table this one. But my question had to do with knowledge gaps. And I personally, I hear so much about knowledge graphs as the way to go and how insight, how exciting an area is. And I honestly don't know that much about knowledge gaps, and I also don't know that much about therefore how they're applied in business. So maybe speed round, I'm not sure. I don't know if that's possible for knowledge graphs, but that was my question.

Harpreet: [00:53:01] Great question. I wish David Knickerbocker he'd be a good resource for that. Let's go to let's go to Jim for knowledge, kick off the discussion on knowledge graph. Eric, do you think this would be something like be able to jump in on as well? I'll wait until afterwards.

Speaker2: [00:53:18] Knowledge left the knowledge graph. Yeah, that's complicated. It's a big question because there's a lot of different use cases for it. You can embed domain knowledge and knowledge graphs and you can actually train deep, deep learning models with some basic domain knowledge using a knowledge graph. You can even start overlapping with a very basic causal graph because knowledge graphs can contain causal relationships where they can propose them. You can discover knowledge graphs in a number of different ways. There's a you can go through text analytics and you can begin to mine text for causal relationships and build knowledge graphs through text. [00:54:00] There's a ton of different applications. The hardest thing about every single one of them is when the knowledge graph gets too big, how do you validate it? And so that's the that's been kind of the stumbling point is that you you get to a point where it's so big that it's unmanageable and unverifiable on the edges. You know, you look at the main center that everybody's going to be using and there's enough use cases and enough people playing with it to validate. But then you get to the edges and weird stuff happens and you get, you know, especially if you start relying on it for anything. And even something as simple as just document search, if you create a knowledge graph based on your HR or document library. You can ask the knowledge graph some questions and get some real interesting responses about some unintended policies showing up just because there is a relationship between harassment and whatever health care question you were asking somehow. All of a sudden you're getting recommended for harassment training. So there's yeah, it's got a lot of positives, but it's really complex and there's just a lot that can go wrong when it gets big.

Harpreet: [00:55:17] Thanks so much. Anybody else have any thoughts or insights on the knowledge gaps.

Speaker5: [00:55:21] If I may? I have not had opportunity to use them other than in code, but then I wanted to ask you a question about them. And maybe this will help Gina. It seems like. They could be automated on the output side or manually derived on the input side. I've even come across people seeking my help and they were cleverly using them as inputs for deep learning networks. But is all that accurate?

Speaker2: [00:55:55] You're asking me? Are you asking?

Harpreet: [00:55:58] No.

Speaker2: [00:55:58] Sounds like Ben. Yeah, it does. [00:56:00] I know. Yeah, I'm guessing.

Speaker5: [00:56:02] Let me say then from Reno. There we go.

Speaker2: [00:56:07] Well, Ben's from Reno, too. This is confusing.

Speaker5: [00:56:11] Ben, I feel like I've hurt your feelings, but trust me, it was unintentional.

Speaker2: [00:56:15] I'm not qualified to answer that question about the graph stuff, so I will default to Ben. Yeah. You can use them to train deep learning models. It's one of the ways that and the biotech field has been using this pretty regularly to train deep learning models with the type of domain expertize. And when I say that, it sounds a whole lot smarter than it really is, but the knowledge graph itself creates a really crude ontology for domain knowledge, or at least embedding some domain knowledge into the deep learning model. And it doesn't have to learn as much because it's not relearning all the basics. And you can actually improve the ontology or improve the knowledge graph and make the deep learning model train a little bit faster. So there's yeah, you can definitely do some things with a knowledge graph and use it to train deep learning models. But it's again, it's really hard to build any sort of ontology and validate any sort of ontology, and you can do it in an automated fashion. But at some point you have to figure out how to say what's accurate and what isn't. And the less data you have around a particular part of the knowledge graph, the more uncertain it gets.

Harpreet: [00:57:34] Thank you so much, Gina. Hopefully that helps.

Speaker3: [00:57:39] And I need some I think some self study along these lines will will help. I did listen to Kang's podcast with then from back in September and you guys touched on that somewhat and a lot of other great topics to. So thank you very much, guys, [00:58:00] for those thoughts.

Harpreet: [00:58:02] Shout out to Kensington Stables podcast recently crossed the 5000 subscriber mark on YouTube and 200,000 downloads. God damn, that's insane. Good job, Kenji. Shout out to Kyle on in the building. Kyle and I used to work for their industries. Kyle took my prototypes and made them into a reality and help drive business value. Kyle, we're going to get to your question after we get to Marcello's questions. So sit tight, man, but it is good to see you again, by the way, because one of the smartest kids that I know, he's fucking amazing. You actually all connected, Kyle. All right. So Marcellus question coming in is coming in from LinkedIn. With AutoML evolving, do you think the future of this career will be centered in the ingestion, data engineering and or deployment machine learning engineering or AutoML will never completely replace the function? I was hoping Ben would be here. I would love to. Get them perspective on that. Mark. Yeah. So regarding the AutoML, I was recently reviewing a vendor and as we said, like a pitch with them just to kind of see what's about like a potential use cases for it.

Harpreet: [00:59:20] And one thing that really popped out to me that I didn't consider until I went through it, kind of like the sales pitch was that when you do AutoML, the ML component is generated by the vendor and stored by the vendor. And so it creates like this potential bottleneck or this potential reliance on a vendor where your IP is tied to them. And so say for instance, like the vendor got sold to someone else or you have to go switch services or for some reason that's going to be a huge pain making that switch. And I know you can export kind of like the parameters and all those things, so there are ways around it. But [01:00:00] that was an edge case that I considered. I'm still trying to learn about like, is there a way to kind of get around this where we want to like leverage? Automl But we're not stuck with them. If you want to do more kind of complicated things, so not necessarily a full answer, but something that really popped up recently that was very top of mind. Mark, thank you so much. Let's go to Mexico.

Speaker3: [01:00:24] I think as long as you have broke startups, you're not going to have full adoption of AutoML. I know that's like super salty, but like so, so, so AutoML. Right. So these things that we talk about, like future generation model training, model testing, observability, deployment, retraining, the closest thing we've gone to that are like. Like Comet ML, maybe Neptune. Google's Vertex II may be a stage maker, and you still need so much there just to make things work right. So even like I think the part that is surprising to me, but it's kind of like hopeful in a way, is that even in conversations with like senior and like Prince staff and principal, like ML. Engineers, data scientists, data engineers, software engineers. There's not really this sense of like, all our work is going to get automated out because a lot of these decisions are really hard. So like even if you were to literally like do push button, like let's say we have a push button to deploy like a recommendation system versus like a forecasting system. And let's say there's like certain latency requirements on like the front end, like it has, one has to be real time, the other one is batch like even kind of humans disagree on [01:02:00] what is the right like which what, what design pattern do you use. So, so, so where I see kind of things like being helped, right? So for example, like a lot of times AutoML is literally like try a bunch of different model architectures and try a bunch of different like tuning parameters and pick the one that works.

Speaker3: [01:02:18] Like I think that's kind of the state where it's at right now that's obviously like super helpful, you know, but like you still have to somehow like input. For example, if you want to like make sure that your model is not biased, you still have to determine what those segments are. Right? And there's still going to be like data regulation and like management requirements and also goes back to the whole system designs like super hard. Like it is it is really hard even like senior staff, like they struggle with this. So I just don't I see like certain parts of low hanging fruit being automated. But even then, like that knowledge of how things work, like one or two layers down, that still has to be with a human because the minute something breaks and like you need like support to go on it or you have to troubleshoot, you know, like we're still helping engineers like literally troubleshoot their GitHub issues. And that's like GitHub is one of the things that like literally every single engineer and analyst can kind of agree on hypothetically. So if we then start going into the more abstract areas like model training, deploying like a container versus serverless versus whatever. Yeah, it's a hot mess. I feel good about job prospects for myself, like in the next 40 or 50 years, at least as long as I'm on this planet, assuming it doesn't get nuked by a bunch of competing superpowers. But all said and done, I think it's still going to be like a good career, you know, good career there.

Harpreet: [01:03:51] Thank you so much. Point of clarification, though, Comet Mel is not a AutoML platform. Comet is an experiment management platform. So [01:04:00] essentially just to help you manage and keep track of your deep learning experiments. Also the new product just launches model production monitoring. So that helps you essentially monitor the drift concept of data drift and things like that. I still am an employee of Comet for the next week, so shout out to Comet. Even though I'm leaving Comet doesn't mean that our relationship there is is finished. You will still see me supporting them in various forms shot at the comet. You guys should try it if you are doing deep learning. So, Eric, then Tom, then Danny, and then we'll go to Kyle's question. Yep. Just something small. I'm going to take a little bit of a different tact and assume this person's maybe asking about AutoML because they're wondering if they should learn to code or if data scientists need to code or if it's all just going to be automated. Somebody asked about that on LinkedIn recently, and I kind of had a thought about it that I'd never really considered before. And that was, you know, we get all like all in a huff about this with data science, but nobody's ever, like, you know, does a doctor need to know how to do surgery? Like, nobody cares.

Harpreet: [01:05:10] Like, some doctors know how to do surgery. Some doctors don't know how to do surgery. And there's going to be a need for both types of doctors or whatever. Right. Same thing with if you if you need to if you know how to code or if you can use these ML AutoML tools or if you know Alteryx or whatever kind of to make ego's point about like as long as there are underfunded startups or whatever, there will always be a need for a wide range of skills. And so if you are way into AutoML and or you have the opportunity to pick up the skills like awesome. If not, then that's okay because there's a company out there that's going to need your level of skill and just keep working at it and then you can find that right match. You haven't seen this week of a lot of debate. Maybe it's just my feed popping up. I seen quite a few posts talking about AutoML. [01:06:00] Yeah. Just learning the code. It's just a good skill to have. Let's go to Tom. And I think Russell wanted to have some input as well. So maybe Tom Russell and then Danny and then Kyle's question. So we'll do that or Tom Russell. Danny. Then we'll go to Kyle's question.

Speaker5: [01:06:15] So I think this is a cool question. One thought. I think most of the people here that can code can create their own AutoML platform given enough time that that's my confidence in the level of people that are here. Second, we've seen something like this in the past where managers thought, Oh, great, we have these finite element analysis packages now. I don't have to hire an expensive mechanical engineer to do this analysis anymore. Well, guess what? They found out just because you had a fancier tool didn't mean you didn't need to have a really smart person running it. So throwing AutoML in the hands of a non data scientists, I think that could be a recipe for disaster. But wait, there's more. There's so much work in the pipeline that requires a lot of careful, thoughtful coding. But most of all, and this is what completely gets me, Greg and I love to give this talk on return on data. Greg and I. And it's with this spirit. If. Why do we have such an emphasis on how many machine learning models get into production? When 80% of the work and insights can come from that 80% you do before you get to the predictive modeling. For example, if you're doing your job and you're really looking at multiple models and you're always looking at linear and logistic regression now and you've done good feature scaling and reduction, you've [01:08:00] got a good pereda a feature importance. Well, to me, that's more actionable data to your greater organization than what you predicted, because they can go act on the knowledge of that future importance. So I applaud what Mexico is saying. Yeah. And by the way, I love Dennis Rothman, but he's really into Transformers and he thinks they're going to take over all of data science. And I'm just like, I love Transformers, too, but I don't agree with you there because there's still so much grunt work to just find out from the data what features are most important to the business. Let's act on those instead of just chasing after predictions. Let's do prescriptive analytics. I'm off the soapbox.

Harpreet: [01:08:48] Awesome. Let's go to let's go to Danny and Russell, then go to Kyle's question and Kenji has a question. So that would be the order of operations. So Russ or Danny Russell, then Kyle's question.

Speaker4: [01:09:04] Oh, good. I think you swapped the order around between me and Russ, but I'm happy to go.

Harpreet: [01:09:08] Yeah. So.

Speaker4: [01:09:11] Oh, good. Oh, good. So I think AutoML has a place definitely for in terms of like. It's there for a reason. It's when you know exactly what sort of data set and what sort of problem you're trying to solve. And you just need a really damn good model to be generated by that process and you want it to be regulated, or especially if you want to share it with regulators as well. That's another point that just flew into my head. But a lot of times when we're trying to like identify the problem and work with business stakeholders to actually figure out what the heck they want to predict anyway and what sort of actions they want to drive off the model. The intricacies of building up like a really good, good target variable is very challenging. [01:10:00] And then over time that target variable will change and you need to monitor it and then out the back of it. Once you have a really good model, then you have to think about, okay, how am I going to use it? Will I do I run an experiment? Do I try and embed it into some systems so it starts serving recommendations to people? How do I know whether my model is performing well in production? There are all these sorts of factors around the whole machine learning process, which is definitely not covered by AutoML, and I think that's where we can kind of lean in with the expertize of we know how to build models and we know how to use data and analyze data and analyze models and all of that good stuff to actually drive better outcomes using this new technology. So I think AutoML has a place, but there's still people and processes and strategy around what gets done around all the ML.

Harpreet: [01:10:57] Thank you so much, Debbie. Let's go to Russell and then we'll go to Kyle's question.

Speaker2: [01:11:04] Thanks, Aubrey. So coincidentally, I posted something today on LinkedIn that was about no code, low code or pro code. And whilst it didn't explicitly reference or HTML, I think that's distinctly one of the parts that I was trying to get to with the post. So firstly, it was a22 way post. So the the one direction.

Harpreet: [01:11:27] Was.

Speaker2: [01:11:29] Not everybody needs to code. It seems like data science and coding is the zeitgeist, is the dairy girl thing and everybody wants to go towards it. And I think that's kind of a mistake. We shouldn't be pushing everyone towards that, you know. Yes, it's great. Go towards that. But we also going to need people that can do jobs that don't require coding. So that was one very basic thing. And the other way that's more in line with what we're saying about AutoML here is that also smells great. But as Danny was saying [01:12:00] and other people here, you need to understand the code that's beneath it. If there's something that goes wrong with it, you need to be able to trace the issue. So I likened it in an analogy to say electronic calculators being used back in the day and mathematics.

Harpreet: [01:12:16] Skills dropping.

Speaker2: [01:12:17] As a result because the school syllabuses were allowing the pupils to use calculators to do their arithmetic and they their ability to do mental arithmetic dropped so that we took the calculators away from them. They were kind of lost. And I think it's the same here. If you have anybody, you train them to use AutoML and they don't have a grounding in the code and something goes wrong, they're going to be lost or say the tool crashes or you take it away from them or move somewhere else to another company as a data science that doesn't use AutoML and they need to create the models from scratch, then they'll be lost. So really great tools. But you need to have the grounding and the methodology and the practices to be able to create them.

Speaker4: [01:13:03] Yourself.

Speaker2: [01:13:04] To be able to use them effectively and stable.

Harpreet: [01:13:11] Awesome. Russell, thank you so much. Let's go to Kyle's question. Kyle and my former colleague Price, the guy who took my ideas and production them. Good to see you again, man. Well, hey, man. Yeah, yeah. Good to see you. Hi, everyone. My name is Kyle. By the way, your model is I've I've improved it so that we have more explanatory model. People can now feed in more input as they want dynamically, and they can see the profit margin based on suggested multiplier, which is interesting, right? Hey, man, that's awesome. Yep. So my question is this. I have a strong software engineering background and I'm currently currently working as a data architect in two organizations, Price Industries and Slalom. As I learn more about statistics and data science, I'm studying actually I'm about to finish [01:14:00] my master's degree in statistics at UBC. I feel like that data science is essentially that data science. Data science that most of the companies actually do is essentially 80 to 90%. Software engineering. Plus 10 to 20% of some stuff, let's say statistics or data science knowledge. So my question is, do you guys agree that the machine learning engineers or machine learning field will be eventually converging to software engineering? I see a lot of observations, especially in companies that senior and senior software engineers are able to pick up the machine learning fundamentals, and they switched to my shoulder in the engineers and I see that a lot. Probably I see around me, I even even myself, I'm in the same route. So I wanted to see what what everyone's thoughts are around that. Good question. Let's go. Let's go to then. So sorry. Let's go then van on that and then anybody else has an input here. Just go ahead, raise your hand. I'll be sure to add to the queue.

Speaker3: [01:15:09] I kind of I used to think that it would, but I now I think this idea of convergence is almost imposing a false dichotomy in that like. So I've worked as a data analyst, I worked as a data scientist, went to ML Engineering then now went to ML Ops. So living along in the software DevOps side and I feel so much better about my life know so I think. So I think at the end of the day, I think you're still going to have personas of work that people like doing and are attracted to. I. Like if you toss me a data storytelling analysis, experimentation design thing. And someone was [01:16:00] like, Here's $10,000 to do it. I would be like, Yeah, I'm not doing that. You know, I'll pass that high value opportunity to someone else on this call, to other people on this call who are much better at it because I dislike it so much that I am willing to take that that ten k like project, give that, give that money to someone else. And if they were like, oh, here's two K to do like an engineering posse or you're probably going have to work overtime. I'd be like, Yeah, I'm a sucker, let me do that, because I just genuinely like the work better.

Speaker3: [01:16:38] But I think in a lot of companies like you do need that explore. So there's a couple of things that I so a couple of things I see missing from companies that do just treat it as an engineering exercise. One, when stuff goes wrong with their data, like there's changes in their data distribution. They have no idea why. The second thing that goes wrong is when they're actually trying to deploy models for testing, even if it's like shadow testing or like a full on multi arm bandit thing, if they don't have any like fundamentals and statistics like they might, for example, and we've had this happen in a health care setting, they might not get the appropriate sample sizes to actually run those tests, because the reality is most companies are not at Google Scale like it's not like they can run the test for one day and get like statistical significance, like within an hour. Especially in like in like manufacturing, like health, tech and other areas. So I think you lose a lot there. I think also to that, like the benefits do, I see a lot of the amazing data scientists I work with that like like I don't have is they.

Speaker3: [01:17:54] They're willing to kind of go explore and they're willing to generate lots of ideas. [01:18:00] And like, they love reading research papers and all that. I also love learning, but my learning style is just way different. Like, for me, it's how do I optimize? How do I create? The least worst architecture is what they say, right? As a as a system design. And I'm willing to kind of help them out with like the more engineering stuff. But some people, I think they they gravitate towards that. And for me as an engineer, I'm like, I can respect the value of that exploratory work. That's statistical sort of deep knowledge, that appreciation for experimental design. I can appreciate it and also not want to do it. And the best part is like in a lot of companies, they also appreciate that distinction. Some people would argue that's not a good one, but I think a lot of companies, they will continue to invest for that as long as universities are like pumping out PhDs or better yet, even for cutting edge research like robotics, computer vision, all that you can't always think scale. Sometimes you literally have to do something manual in that area to even get a product that is worth like prioritizing.

Harpreet: [01:19:11] So that's why I think maybe maybe I should frame my question in another way.

Speaker2: [01:19:15] I don't I agree.

Harpreet: [01:19:16] That data scientists will never go away. We need statistician. We need someone to make the decision. We need someone to make the call. My question is whether ML engineers who are in between data scientists, let's say statistics statistician and software engineering, well eventually convert to software engineering and the data scientists will be serving more pure statistical and data science role and not be required to understand system design and software engineering and the software engineers and will be and machine learning and ML engineers will be basically converts. That was my question. I never got the statisticians. They existed for hundreds of years. Data scientists will exist for a hundred years. But what about the ML engineers who [01:20:00] are currently booming in the current decade? Right. That was my question. A lot of people are are interested in ML engineers and that's a big buzz word. But what's what's the future of ML engineers? Is it convergence to software engineers or are they will they be a distinct field going forward as well? Let's go to. Let's go to then. Then Danny, then Kenji. And then. Then after Cannes, we'll see if there's any other input. If there's no other input, then we'll go straight to Ken's question. Let's go to Danny then Ken.

Speaker2: [01:20:35] So I think what we're all missing and it's kind of funny that we're all in data science, but we're all missing sort of the fact that data science is in software development. And so what ML engineers do right now is going to evolve. We're going to start having because when you think about software development, the valuable thing that you create is the code, and the code is what does most of the functionality it does. So the platform itself doesn't get any better until the code gets better. But models use data and that's a completely different paradigm. It's not the code. And so when you talk about an ML engineer, it's less about what they code and more about how they architect for that new paradigm, because that's totally different. I mean, think about it. The platform gets better as you feed it, better data. It's not about code anymore. Your model becomes more functional as you feed it. More data, better data. And so an ML engineer is not focused on code. I know they write code and they have to be super technical and they have to have a strong understanding of architecture.

Speaker2: [01:21:41] But you have to realize that what an ML engineer is going to be doing over the next probably ten years is more focused on improving models, not using code and architecting increasingly less code reliance into the platform and creating essentially a [01:22:00] model that handles all of the functionality. And it improves as data improves, it improves as the architecture improves, as you come up with potentially novel architectures, but you don't have to write a whole lot of code to get any of that to happen. And the platform is really the thing that software developers are going to be focused on is building the platform so that anybody who's paying money uses the platform to get access to the model or to get access to the data. And so the platform itself kind of becomes irrelevant. It's almost like coding is one of those things you want to commoditize because it's not adding as much value, it's just a bunch of APIs. So that's and that's where you want to think about machine learning engineers is what we do with them now is not what we'll do with them over the next ten years.

Harpreet: [01:22:55] And thank you so much. I will go, Danny. And then to Kenji.

Speaker4: [01:23:00] I also think that the whole ML engine software engineering like it kind of grew out of a need of having more robust systems for machine learning. Essentially before we had any of these larger platforms and frameworks, we would just kind of code everything ourselves. So it would be like end to end pretty poorly written code depending on your on your software engineering skills. But it would work, right and it would bring value to businesses. So then business essentially went, okay, let's get this bike really good so things won't break anymore. But over time, as Vin was saying, I think the the focus will probably shift to something different over time. It might be like building scale scale systems which scale even further or maybe more robust systems which don't need as much maintenance or different things like that. I think the some of the ML that we're seeing now is how to hook in with the [01:24:00] data engineering teams to do like fully automated machine learning, like fully, fully automated, like building target variables, making the drift calculations and making it all work within the database environment. So that stuff is something which could work down the end, but it's like no longer machine learning engineering and it's like a mixture of data science, machine learning, data engineering all put together.

Speaker4: [01:24:25] I also think that I've worked with a lot of software engineers who wanted to transition into ML engineering because it's like the hot new thing and they're really proficient at it. But there's still like a fundamental gap which is missing of like the machine learning knowledge. It's not necessarily like math and stats or anything like that, but it's like, how do you build a good model using data? How do you use it to solve problems? How do you know if something's gone wrong? I think that's the gap for all of the software engineers, like pure software engineers and designers or system design folks who want to move into that space. But it's surprisingly not difficult to learn that stuff. You just need to build a model and try it and then over time you just figure out what's what went wrong. So yeah, I think it might not converge, converge in the end, but they'll definitely be like more and more overlap over time.

Harpreet: [01:25:17] Yeah. I think that the convergence to me, like, just, just just thinking about out loud, it sounds like the convergence is in like off to the side of things that I feel like is kind of where that convergence is happening. So maybe not a full on convergence, but it's a little bit of a stopgap, I guess, like a better word. But Ken, let's hear from you.

Speaker2: [01:25:38] Yeah. Really quickly. I'm not sure if I even really agree that there's going to be aggressive convergence. I mean, if you look at the larger trend of positions in the domain, I mean, like what, like seven, eight years ago, you pretty much only heard about data scientists, right? And then the field was broken down and fragmented more. You have data engineers, you have data analysts, you have ML machine [01:26:00] learning engineers. You have all these different, more specialized positions. And I think the trend in the industry is generally go going towards more specialization. So you're going to I would expect that we'll see like more engineers having more tools that are specifically relevant to them and less relevant for the average software engineer as well. So in my mind, like I think that there's a probability it goes any direction. There's no way I can predict the future. But my bet is looking at the larger trend and saying, Hey, we're seeing more specialization. There's probably going to be a better definition of what a machine learning engineer is in five years than what we have now. Very much like how some of these other roles have been hopefully better defined over time. So again, that would be how I would go about thinking about that problem and that as someone in your situation as a student looking to advance your career, that's something you might also want to pay attention to is like, what would I want to specialize in if I wanted to specialize among these positions? Like which one gives me the most flexibility or which one? If I go down that path, can I have the greatest return on going forward?

Harpreet: [01:27:14] What do you think? Any follow up questions they're going to be follow up add ons or anything? I think I was always curious of this question. I think I even asked this question to you, Harp, a couple of years ago. But I think the answer to that, I'm hearing is most of the people agree on that. It's going to be, if not more, if not less, it's going to be only more specialized. And that's what I also thought as well. But a lot of a lot of folks are a lot of people around me who are who are software engineering background, who moved into ML engineers surprisingly realized that what ML engineers do are are most of ML engineers not the most advanced, let's say PhD level of data science are pretty easy to pick up and learn. And that's why they were. Well, last Friday we were [01:28:00] having a beer night and they were thinking of, oh, maybe it'll be it'll be converted into a software engineering eventually. And I think I agree with you guys, it's going to be even more specialized if not. Yep. Thanks for. Thanks for all the valuable thoughts. Thanks so much for coming by. We're going to have to connect offline. I want to hear what you've how you've built on top of that model that we had to for sure. So let's go to Ken Ken's question, Canada question. Good question here. So I'm excited to get that.

Speaker2: [01:28:33] Yeah, I mean, this one's, I think, pretty interesting. It's more of a like a brainstorm type of thing than than it would be a question. But I mean, with what's going on globally with Ukraine and Russia right now, I'm interested in how data science or data scientists uniquely could provide value to everything that's going on there. So something that I've been very interested in this whole conflict for a while. I read this book called Sandworm that talks about the state sponsored hacking that that Russia has been doing and essentially like. Crushing the Ukraine and testing cyber weapons on them since around 2014. And there seems to be a very clear application of of what people can do in the in the cyber and in the hacking communities around how to help the situation there. We also saw an incredible outreach of help during the COVID pandemic from data scientists and aggregating data and making it more tangible and useful to people. And so my thought is, I mean, obviously, this is very different than the COVID scenario, but how can data be leveraged to help improve the quality of information? How can data be leveraged or data science will be leveraged to help save people's lives in some sense? I think that that's a kind of important and and interesting question to [01:30:00] to bring up in a discussion setting.

Harpreet: [01:30:05] Great. Great question. Ken, I'd love to hear from a spokesman to get to the brainstorm. So going then. What do you think?

Speaker2: [01:30:17] I. I've got some like an N'DIAYE type thing here, so I can't really. I know that sounds like a strange answer, but I can't really I can't really answer. And I'm really sorry. It sounds horrible. I wish I had something to contribute.

Harpreet: [01:30:35] Yeah. Anybody else? Like I said, it's a good question. I wish I had some ideas here. Let's go. Let's go to Mexico and then Mark than anybody else, if you want to jump in. Just go ahead. Raise your hand.

Speaker3: [01:30:50] So there were two interesting trends that were going on. One was crypto donations to both sides of the conflict. I'll let Mark talk more about that one. The one I will sort of bring up is documenting war crimes. So digital forensics and the deepfakes that have been starting to circulate and that is really important, especially because I don't know if people saw this, but I think over at The Hague, they had. There was the ICJ. So normally they're responsible for prosecuting crimes like Bosnia, Rwanda, you name it. Normally those kinds of cases land up there. And I think there was a. I want to go back and see this, but essentially there are 50 members on 15 members on the court. It was. So I think 13 versus two or something like that with Russia and China basically voting against sort of one of those [01:32:00] statements that they make as if things will be better. And it's kind of important because in a lot of times a lot of these war crimes, they do indirectly get documented either through like GPS locations or through photos or things that are shared on social media. But a lot of times, I think this is actually one of the issues that Facebook has had, honestly, because they've had some instances of human rights abuses and all that, they're not always able to catch it.

Speaker3: [01:32:29] Or if there is like videos or things like that that get passed around, a lot of times it's using old footage that's that's been stitched for propaganda. So I think that's kind of an important sort of area that like data science can help. One of them is like being able to discern if a video is in fact a deepfake. Another part is documenting potential forensic crimes that happen and doing some crawlers in that, and also providing better documentation, better speech translation. Because the thing is like there's a lot of different perspectives and narratives going on. Even within our household, we have a lot of debates about this because some people in India, they have a slightly different perspective on Russia's behavior and actions. Some. People. Family members? Yeah. Like, it's just very geo specific. Right. So I think that's an important area. It's unfortunate because frankly, it's a little bit after the fact. But I do think that when there is this whole narrative of going on that like, for example, a theater that's been bombed, filled with families, you know, I think that kind of veracity of information becomes super important. So sad, but it's needed.

Harpreet: [01:33:56] Mark, we. Yeah. So [01:34:00] thanks for bringing up this question, Ken, because I think there's been a lot of people's minds too. It's like, how how can you help considering this huge humanitarian crisis? I think one thing that I've kind of struggled with, what is is that when I became a data scientist, one thing I said to myself that I would never do is I'll never use data to like that can be used to kill people. I refuse to do any defense type of work. And I've even had clients where like they'll say like, Hey, we're proud to be veterans and I will double check with them like, yo, that's cool. Are you, are you doing defense or are any of your clients defense? Because if it is, I can't do it right. So that's been a challenge because a lot of the allowed, like the crowdsourcing there is like a mix between humanitarian aid, like Red Cross and like getting supplies and like getting bullets and be able to trace like where it's going to where. And that's been something I've been really struggling with going back to like what you can do within that vein. And I think Michael really already highlighted this. Well, I'll tie it back into kind of like the Web3 blockchain thing is data mining. There is so much information out there, there's like propaganda or Twitter or like where things are moving.

Harpreet: [01:35:17] And I think someone's been highlighting this kind of crisis in this war is the role of cryptocurrency for funding things. And there's there's two kind of arguments, right? So there's one supporting Ukraine and quickly deploying money to them without worrying about bank transfers and all that type of thing. But also for the Russian people who are kind of innocent is where their country in the leadership is like completely going all in. And this is some of them are evil where again, I may be falling for propaganda myself, so I need to double check that. But I was hearing soldiers now where the father is doing a training mission and no, they're actually going to war. So there's these things called bank runs [01:36:00] where everyone's going to to the bank to pull out money, and therefore there's no money in the bank. And so that's one of the big draws of cryptocurrency, is that you don't have to worry about bank runs. And so if you're just a citizen who's, like protesting against the war now, you're impacted because of these sanctions and whatnot that are outside of your control. And so blockchain has this interesting dynamic in cryptocurrency where, one, you can subvert these sanctions.

Harpreet: [01:36:28] Either you're innocent bystanders stuck with your government or the government itself. On the other side, you're able to deploy kind of transactions. And so I think something I saw I forgot was like someone from MasterCard, they raised like $10 million through entities and like 30 minutes and was able to send it directly to to Ukraine NGOs. And what's really interesting about that is, you know, if you if you donate to an NGO like just through like your credit card or something like that, you have no idea how it's used or where it goes. With cryptocurrency and blockchain, every transaction is safe for the public to view. You can go on other scan to IO and view those those kind of send and to from kind of thing. So there was one project where they did a Twitter thread was like, Hey, we raise $5 million. This is how is deployed. Here are the contract IDs. You can look it up for yourself. We sent this much here, this much here, and we're setting up a wallet for this other NGO. You can check back in ten weeks. Right. So things you can do is like if you're really I do have some links, I can go find them. Is this where I think like network analysis can become really interesting, especially we're talking about graphs earlier is that you can trace kind of exactly where the money is funneling from and who's it going to.

Harpreet: [01:37:54] There's this phrase kind of like the blockchain doesn't lie. Everything's stored on there. Vinge [01:38:00] raises AIS, There's probably some ways to lie. I just don't know about that. And we can now validate kind of where my money is going. Is it actually going to the people that expect it? But also you can you can quantify how much of this cryptocurrency is actually subverting from from these sanctions. That's a very long winded answer. But just trying to give you a perspective like what's happening with cryptocurrency and how data scientists can mined that data to actually fact check people. Mark, thank you so much. That's great to me. That's a good question. I wish I could think of some some idea that could that could help with this. I mean, this thing that is coming to my mind is, can we help match refugees to potential? Host families collecting information on host families, collecting information on refugees, maybe finding recommended families for them to go to, or maybe recommended geographies based on the skills that they have. They can move somewhere to just quickly get kind of up on their feet. Not too sure. But if we keep on this.

Speaker3: [01:39:20] Fun fact, there is a kid who did that. He put together a website to do that. He was also apparently the same kid who put together one of the most popular covert trackers. So for people who are like, what is what is my perfect internship project? Honestly, sometimes it's just getting something out doesn't even have to be like the most amazing web app in the world, and it's like a very stripped down version of, like, Airbnb. So he did do that.

Harpreet: [01:39:55] I mean, how about the call to the leaders out there? [01:40:00] Because, you know, a comet, a lot of our engineering staff is Ukrainian and I know that. In tech, there's a lot of Ukrainian folks intact. What can we do to help? Sponsor them. Bring them over to the States or Canada? I don't know. Opportunities. Job opportunities for the people that's here. Gina.

Speaker3: [01:40:27] I just posted a couple links in the chat, but obviously for people just listening to the podcast, there is a coalition of companies Adobe Arm, BBC, Intel, Microsoft and.

Harpreet: [01:40:42] To.

Speaker3: [01:40:42] Form a coalition to develop an open standard for trust, for tracing the origin and evolution of digital content. And that caught my eye. That was actually February 22nd of last year, and I had a long time ago worked on an anti counterfeiting project when I worked at Hewlett-Packard, specifically in the realm of pharmaceuticals, because at the time they had plants in Puerto Rico and they did work with pharmaceutical companies. And so it's just a fascinating area and it applies tracking and tracing applies across practically any industry you can think of. I'm sure we're all aware now that counterfeits go way beyond just like electronic products or handbags or clothing labels. There's just you can't believe the amount of stuff people can't counterfeit. And a lot of times it is in China or other places around the world. But this new initiative, well, it's not so new now. It's a year old. But the point being that that could be a way or an area to look.

Harpreet: [01:41:55] Into.

Speaker3: [01:41:58] Can and anyone else who is [01:42:00] interested in that. So I initially found this. I can't remember how I saw the initial announcement. But anyway, on Microsoft's blogs, they talk about this so one could search and easily find it.

Harpreet: [01:42:17] Awesome. Gina, thank you so much. I know there's a couple other questions on LinkedIn coming through, but we'll go ahead. We'll have to table those for next week. So I hope you guys do come back next week and ask those questions. I promise to get to those first and foremost on the on the show. One of them was about quantum computing. I was hoping that Mr. Kokila would be in the building, since he's our resident quantum computing expert. But we need you next week for that. And then this question about a good portfolio. We'll talk about that next week as well. So, yeah, hopefully next week I'll be at a co-working spot. I'll probably. My house is proving to not be a good place to stream from because I keep getting blinded by the sun. So yeah, we'll do it from the spot next week. And then in April we've got three co hosts or guest takeovers that can be taken over Shady St and so we'll be taking over and Kiko will be taking off as well. Hope they can get somebody else in there to help out as well. And you guys are you guys are awesome for being here. Thank you guys so much for for taking time out of your schedule to join me. It's been good to be back. So good being a full officer. I think I haven't done one probably since early February. So like at least at least a month since I've opposed it.

Harpreet: [01:43:41] So felt good to be back wrapping up my last week at Comet next week. Just getting some good content into the pipeline. Even though I won't be working there, I will still be having positive relations with them, will probably still see me help and support them and whatnot. [01:44:00] I'm excited to be headed over to Pachyderm doing some awesome stuff with them, so keep an eye out for that. Be sure to tune in to the episode that was released today. With Joe Reese, the boss of the data Nerd Herd. It's a great episode, so be sure to check in and and and tune in on that. Ante, thank you so much for hooking it up on on on the buy me coffee guys. Remember to check the show notes. If you're tuning in on the podcast, there's a link there that you can go and buy me a coffee. I'm not actually buying coffees at that stuff. Uses it to get new equipment and pay for some editing because you haven't had some haven't had sponsorship for the podcast for a while. So that being said, if you're a CEO or CTO or a founder and you want to get some exposure for your show, reach out to me or get a sponsorship. Thank you so much for taking time to be here today. Remembering my friends. You've got one life on this planet. Why not try to do something?