HH74-28-03-2022_mixdown.mp3

Harpreet: [00:00:06] What's up, everybody? Welcome. Welcome to the Art of Data Science. Happy hour. It is Friday, March 25th, 2022. The snow is melting out there. As you can see, there's still piles and piles of snow melting. Spring is upon us here. Hopefully guys get a chance to be up. So that was released today and episode with Brittany Doerr talked about her book, It's Bigger Than Leadership, really great conversation, chat with her. She's still in college and published a book. I think she's like a junior or something, call it. So it's a really great book all about leadership, especially for younger folks as well. So definitely tune into that episode if you have not already other great episodes in the pipeline as well. We're celebrating two years of the show on April 8th. April 8th will mark two years since I released the first batch of episodes of the podcast, and one of the first batch guest is right here in the room of the right here in the room we recorded right before that COVID thing kicked off for the first time. So it was a great, great conversation. So let's let's let's kick this conversation off. You also got to say, though, last day at comment today, definitely, definitely sad to leave but I'm on to the next thing that's the next opportunity will be amazing. I had a great time working at Comet. I think they are defacto machine learning experimentation management platform out there. So if you are managing experiments, definitely check out Comet. Looking forward to continue to work with them doing content and other other things. I'll be moving on to Pachyderm B Developer Relations Manager over at Pachyderm, so I'm excited for that.

Harpreet: [00:01:50] Take another kind of pivot in my career, moving from data science practitioner now to kind of a role that I feel like I could really impact the [00:02:00] industry at a much larger scale. Developer relations at Pachyderm. So I'm pumped for that. It's going to be a great journey. So before it started there, April 4th is the first day there. But yeah, dude, let's, let's get into this. So I want to keep off with is, is about data science and competitiveness. I saw a post on LinkedIn earlier today. I can't remember who it was that posted, but I actually liked it and I agreed with the perspective he was providing there. He's asking, he's pretty much saying our data science jobs getting less competitive. And his argument was, were they even competitive to begin with because it was just ill defined kind of a job description. A lot of people were applying people that were not even qualified for it to begin with and they're just getting expelled. So is it considered a competitive field that people who are applying for the job that aren't qualified or whatever? So lots of different ways you could take that take this conversation. But but let's start there is data science less competitive? Is it becoming less competitive to get into data science? As always, take it off with Ben and then we'll go to Russell. If you guys got questions, go ahead. Let me too, right there in the chat. Feel free to type it out on LinkedIn, on YouTube, wherever it is that you are joining from. And I'll get to your questions then go for it.

Speaker2: [00:03:25] I feel like you just handed me a match and some gasoline that was competitive. I'm going to be completely honest and to preamble before I get mean when I say that there's different levels of maturity. You know, you get your early stage maturity companies who they don't really understand who they're hiring and you can't blame them. You really can't. There's nobody in the organization who understands data science. So how are they supposed to hire a data scientist when they don't know what a data scientist is and they don't know what they're going to be doing with data science? Then [00:04:00] you got the ones and these are the ones I really blame where they've had a data science group for a couple of years, but they still don't know how to hire. And then finally you got the really mature companies who some of them have great hiring processes, some of them not so much, you know, they just haven't matured their process and they're hiring the same person over and over again. And so when you say is it competitive depending upon where you are in that maturity cycle, it's it's either.

Speaker3: [00:04:32] A.

Speaker2: [00:04:32] Lack of kind of an ignorance of what it is that you should be hiring. And that's why you get a ton of people who aren't qualified. And that's it. If your job description doesn't make any sense, what do you expect? Somebody's going to take a look at it and go, That's nobody. No one has 18 years of experience in Kubernetes. It's just I'm sorry. They don't it's not possible. So don't ask for it. You know, ten plus years of data science, there's like eight people who are available with ten plus years of data science. And, you know, that's it. Everybody else is either running a business or Andrew being in that sort of a category where you're not hiring that person, you know, especially not for 130,000 a year. And that's the other part of it. If you offer a salary for that, only a junior data scientist straight out of college would take. I don't care what you put in the job description, no one else is going to. No one's showing up except for people who are going to take that salary. And so companies, you know, when you say, is it competitive? Well, if your company doesn't know what they're hiring for or if you do something that's backwards, as far as the hiring process is concerned, yeah, you're going to get unqualified candidates if you just use common sense hiring. Reach out and network with people. Build some sort of network at conferences, do open source contributions. [00:06:00] So you build up a community that way, you know, maybe hire someone that could be a community relationship. I've heard of people that do this, like community relations harpreet I don't know if you've heard of these people that that build communities. You can actually get talent from those communities. You can hire people that you already know, that you've already worked with. Why do people do this the hard way?

Harpreet: [00:06:24] Yeah. Yeah, definitely. Community builder type of roles are emerging in it because there's more and more tooling now for machine learning engineers and data scientists and things like that. Shout out to Susan Walsh is in the building.

Speaker4: [00:06:40] I'm in the right time zone. That's why.

Harpreet: [00:06:43] Yeah, I was speaking to Kate earlier, like I think yesterday, day before, and she mentioned that she was going to be hanging out with you and Ben and Scott Taylor.

Speaker4: [00:06:52] And we have spent the afternoon together. It's been so much fun.

Harpreet: [00:06:57] I am thoroughly jealous. I wish I was there. I wish I wish I was.

Speaker4: [00:07:02] I'm sure that you will see what we got up to at some point.

Harpreet: [00:07:05] Yeah, yeah, definitely. Susan So you want to tell us anything about what you guys were up to? What did you guys get into? Yeah.

Speaker4: [00:07:15] So Scott was impressive. He came with a whole kit and we had individual interviews and we also did like a group chat that was videoed as well. So it's going to be really cool just talking about stuff. So yeah, it should be fun and something special that I can tell you about.

Harpreet: [00:07:34] Oh, okay. Well, I'm excited to see what's what it is. Speaking of Ben Taylor, he's also in the building right now. Ben, what's going on, man? So, Russell, I think you're frozen, but if you want to chime in here, if you want to talk about is data science more or less competitive than it used to be to break into? That's a topic of discussion, by the way, if you guys are joining in on Zoom, I see ten of you guys so far I'm sorry, on LinkedIn. Go ahead. Let me [00:08:00] know if you guys got any questions and we should smash that like there.

Speaker2: [00:08:06] Yes. I'm sorry. So?

Speaker3: [00:08:07] So I echo Ben's comments entirely. Maturity is a key factor to this, but also AIDS. I started out by saying there's kind of two streams of understanding of of data science, I think. And this affects.

Speaker2: [00:08:24] Both.

Speaker3: [00:08:24] The immature and the mature companies, as far as I can tell from a lot of things I see posted that, you know, you got the the data.

Speaker2: [00:08:32] Science, the broad spectrum data.

Speaker3: [00:08:34] Science. You know, everyone wants to be a data scientist. And then you have some areas where it split down into data analysts, data engineer, data scientist, statistical analyst.

Speaker2: [00:08:47] All of those different specifics that are.

Speaker3: [00:08:49] Split out. And if the company is mature and mature enough to know those, then it's probably less competitive because.

Speaker2: [00:08:57] There's more nuance in what the.

Speaker3: [00:08:59] Job description is and the people they're going for. That being said, I still think there's a lot of poor recruitment campaigns out there, just like Ben said, whether they're excuse me, that they're underselling the role, the salary wise. And they're yeah, they're asking for too much. You know.

Speaker2: [00:09:21] You might as well be asking for ten years of experience of being 18.

Harpreet: [00:09:24] Years old.

Speaker3: [00:09:25] For a lot of the things they're putting in this. It's just an impossible.

Harpreet: [00:09:30] And impossible thing to.

Speaker2: [00:09:31] Fulfill.

Speaker3: [00:09:31] I saw a role advertised on LinkedIn recently that was looking for a head of data analysis, and it was for.

Speaker2: [00:09:38] About 55 to 65000.

Speaker3: [00:09:41] Per year. I mean, this is sterling, so.

Speaker2: [00:09:43] Not quite.

Speaker3: [00:09:44] Dollars. And the role level was.

Speaker2: [00:09:46] Identified as.

Speaker3: [00:09:47] Mid to senior management level and there's just so many contradictions in that. There was people that applied for it though, and I think.

Harpreet: [00:09:54] That's part of the problem.

Speaker3: [00:09:56] The people that are looking for entry level roles and see those types [00:10:00] of things and think, oh, well, you know, that's at a management level that I could go for and a salary.

Speaker2: [00:10:05] That's maybe good for me at this point in my career.

Speaker3: [00:10:07] Even though the actual requirements are way out of range. And it just leads to a self-fulfilling prophecy.

Speaker2: [00:10:13] Of this.

Harpreet: [00:10:15] Mire of.

Speaker3: [00:10:17] Poorly, poorly written job.

Speaker2: [00:10:20] Descriptions.

Speaker3: [00:10:22] And I imagine they'll just get a load of candidates that are not what they're looking for.

Speaker2: [00:10:27] But, you know, the old the.

Harpreet: [00:10:29] Old.

Speaker2: [00:10:31] The old phrase, you know, you pay.

Harpreet: [00:10:33] Peanuts, you get monkeys is kind of true.

Speaker3: [00:10:35] And then if you want the right person.

Speaker2: [00:10:37] Firstly to find it.

Speaker3: [00:10:38] All properly, be prepared to pay market rate for it. If you're not going to pay a market rate, you're not going to get.

Harpreet: [00:10:44] What you want. Russell, thank you so much. I appreciate that. I like that. If you are taking notes, expect monkeys. That is a great way to put it. Shout out to Ben Taylor. Gill is also joining us. Good to see you. Lovely. Thank you so much for for tuning in. Look, if you guys got questions about anything, please do let me know. Actually, we're talking about hiring in data science. And I think you wrote a post just a couple of days ago, maybe yesterday, something about your process for hiring data scientists. Can you talk to us about that?

Speaker4: [00:11:21] Is that a question for me?

Harpreet: [00:11:23] Yeah. Yeah.

Speaker4: [00:11:25] Gone. So I think the. Thanks for having me. I appreciate that. Hey, I don't know. I guess I don't know if I have a very clear process, but I think I'm kind of organic by nature. So the pulse that I had yesterday was about questions that data scientists should ask in interviews. And I don't necessarily ask all those questions, but I do ask some of them in every interview I have. Just to get a lay of [00:12:00] the land or the resources they have, the process it has, and the kind of role that I'm. Interested in. And can I leave my stamp on it or not? You know, but the other I think the other polls that you're talking about. How I hire a data scientist. I believe that was a comment that Vince Post had about how to hire or something along those lines. And in those cases, I actually work with a lot of junior data scientists and. I usually go by whether they have a STEM background, but I make sure that I walk project with them so they kind of go over some of the code, some of their GitHub work and how they've chosen this project, how they've laid this out. And by the end of that project, I do have a good sense of where what that person means and where it needs some work. I'm I don't care much about what languages they know because I know kids are smart these days. They are. Their skills are quite transferable. But it's more about what have they done on a couple of projects and whether some of these skills are teachable in the kind of work that I'm looking for or the project I'm looking for.

Harpreet: [00:13:30] Awesome. Thank you so much. Appreciate that. So that question is coming in from from LinkedIn went from Robin. And he's pretty much just asking Canada looking for analytics opportunities if remote opportunities. Look, you can get remote opportunities now from all over. There's companies hiring from everywhere now. So definitely just keep an eye out for those. Start looking on LinkedIn. Somebody asking What's the [00:14:00] best source for senior manager roles in data science in that space? I will say that we're talking about hiring data centers right now. We're not like job finders or what we call recruiters, anything like that. I would say use the search bar. Use LinkedIn search use. Indeed. I think LinkedIn search for jobs is pretty, pretty good. So definitely check that out or just scope out companies that you find interesting in a domain that you find interesting and just look them up and go to the careers page. Even if they don't have a job posted, just apply for them anyways. So definitely check that out. I like this question. Okay. So I feel this one too, then, because you kicked off this discussion about getting paid, somebody saying, all right, well, if I'm two years experience data analysts in the banking domain, what's a decent amount of annual salary that I should be expecting? This is specifically in Canada. So whatever then says just divide that by two because Canada pays shit for it.

Speaker2: [00:15:06] You know, the way you want to look at any sort of compensation, forget the number of years of experience that you have. What you want to look back on is the impacts that you've had on the businesses you've worked with and whether that's client work, whether that's actual like being an employee within a company. It's the same idea market yourself that way and then look at it as your compensation is really a percentage of what it is that you can bring to a business combined with how much that business makes. I mean, if you go for a startup, you're not going to make three or 400,000 a year. It's just not realistic because the startup is not profitable yet. So there is no way that a poorly funded startup is going to pay you a ton of cash. If you're looking at a midsize business now, you're looking at a different situation. So you could make a significant [00:16:00] impact there because they already have revenue. They have enough money to invest in a data science team. So you can kind of hear what I'm going through. You know, I'm going through a progression of how much money have you made companies in the past? Don't worry about years of experience or capabilities or anything like that. How much do you think you can make this particular company based on what they are? What's their revenue? What are their opportunities to monetize your skills? What kind of projects do they have in mind? I mean, do they even know are you going to be that person who shows up and really transforms the team and helps them figure out a direction? And all of that plays into what your compensation is. When you talk about finance, that's a great domain to be in. It's one of the higher end domains.

Speaker2: [00:16:47] So you're looking at some domains like finance or at the very top. Insurance is one of the ones that top health care is up there as well. Then there's ones at the very bottom. So get an idea of each one of those aspects and then get a salary. Forget the averages. Whatever you do, do not look at any average. They're all garbage. Any survey, those are bad too. But think about yourself in terms of value and think about the company in terms of that specific company. What could you do for them? And then go for a compensation package because salary is just one part of your compensation. You can do a flat salary. Talk about bonuses. There's two types of bonuses. You have guaranteed bonuses and you have performance related or performance based bonuses. You can have a significant amount of your salary or excuse me, your total compensation and performance based bonuses, companies like that, because if you don't perform, they don't have to pay you. If you do perform, everybody's happy. And then also look at stock. If the company has a track record of their stock growing year over year and be careful right now because tech companies are getting brutalized. So [00:18:00] stock is different now than it used to be. You may want to ask for grants versus options. Options are pretty much the universal thing. But if if the price that they staked at given to you at ends up being underwater, now you're losing money. I mean, you're not literally losing money, but the stock isn't worth anything to you. So you might think about grants. And so those would be everything that I would go through if I was trying to figure out what my compensation was for a job.

Harpreet: [00:18:30] Thank you so much. All right. Let's go to coast to coast. Got a question? This question is coming in on LinkedIn. And these questions are of the variety of how do I get a job and stuff like that. We'll definitely circle back up to that. We've talked about that stuff hundreds of times in one of the other 200 iterations of this app. So please feel free to go to the back catalog of the artist site Happy Hour, and just dig through that. But if we've got time, we'll come back to it. Coast to coast. If I was chatting with your classmate earlier today. Richmond. Richmond. Yeah. Yeah. You had a great time to chat with the cool guy. Yeah. Go for it, man. Yeah, he's a cool bloke. I think Ken Kenji knows him as well. Yeah. Interacted a bit as well. Yeah. He's a cool bloke. We studied our masters together and then now he's doing all sorts of stuff, writing for video and a few other things like that. Cool bloke. Yeah. Check out his. Check out his LinkedIn and his medium article if you guys want. The name's Richard Richmond. Are lucky allocate there's a plug Richmond. You're welcome. But in the meantime, what? Something that's always on my mind is, is growth and development, right? Like, I'm always trying to develop myself as a machine learning engineer and as a data scientist.

Harpreet: [00:19:49] Right. And I know where I'm focusing in terms of computer vision, but I'm always torn between, oh, I want to go read papers, learn the new stuff, learn about the latest platform [00:20:00] stuff, the latest engineering background stuff, the latest research that's coming out. But then I'm also like, Ooh, it's been a while since I've touched, you know, basic linear algebra and some of the, you know, Bayesian math and some of the basics, right? So I want to hear from people around the room because there's a lot of experience here. How do you guys balance the explore new stuff in studying and in and in keeping up with your field versus exploit the basics? And how do you balance that? And maybe a round table question is, what's everybody reading right now? What's your focus on studying? Do you have a pattern as an established data scientist? Yeah, I love it. It's a great question if you guys are going to research. So I spend some time every summer going back to basics like a month or two, just going back, refreshing up on the basics. But I do it in the most fun way possible. I've got these books that are not damaged because, you know, the basement flooding. I lost all of them, but they're the manga guides and the cartoon guides.

Harpreet: [00:21:00] So I had like the manga guide to linear algebra, to calculus, to statistics, to regression analysis, 20 different books like this that are essentially manga slash comic books, and they're just teaching the basics and it's a ton of fun to read. So I'll go through one of those in a week just by reading a few pages every morning. So that's kind of how I keep in touch with the basics. Like. In terms of learning new stuff. I kind of just like I'm very distractible in the sense that, dude, like, I think everything is so cool and so interesting and I'm just so curious about everything, but I try to just find the one. Like I will tend to sample a bunch of different things and then find the one that holds my attention the most and explore that until I'm tired and move on to something else. So recently that's been a lot of I've been going all in on jet like just spending the last two weeks. It's really trying to get [00:22:00] past the fundamentals and basics and get and just start learning a little bit more like, you know, all that fancy stuff. Like he goes talking about like squashing comets and rebasing and cherry picking and all that stuff. So that's, that's, that's what I was in for the last two weeks. And now I'm getting into a Docker and Kubernetes just because I'm going to have to have the more and more I go into this like ML space, that's competencies that I'm going to need to have to be able to speak intelligently about, especially like a developer relations position.

Harpreet: [00:22:33] So that's what I've mostly been in, technical stuff wise. I always try to mix in some deep learning in there as much as I possibly can, maybe pick up one one algorithm. Like this week I was, I was doing basic stuff, fleshing out my last project for Comet and it was just using some auto encoders. So that's pretty cool exploring different activation functions. So instead of the Revolut activation function, I was learning about the cell activation function, which you need to have the liqun normalization as the kernel activation function. So that was new stuff I was doing this week, but I'm also reading this sci fi book, apparently not really listening to it. It's called Exhalation and they're a collection of short stories and they are almost like a Black Mirror episode because they all have like this weird twist at the end. So it's pretty, pretty interesting. Let's go to let's go to meet on this. And then I'd love to hear from from Greg and Ben and Ben is in the building, but he is on a airplane. Ben had a safe flight.

Speaker4: [00:23:43] I mean, that's a good question. But I'm by nature, I don't seek new technologies on purpose. I've been in analytics about 15, 20 years now. [00:24:00] I my master's is in econometrics. So I, you know, I, I had that background. But out of curiosity, I actually when I started my career to learn new things, I jumped industries. So I'll go from CPG to financial services to digital to CPG to whatever. And by default I learn new technologies that they use and they apply. So if you are in a company that works in multiple industries, which I highly encourage. So if you're working with a vendor, let's say, and they work across multiple industries, you will be exposed to different technologies, to different models, to different methods of doing things. So my advice would be in the first decade of your career, right, to work with consulting services or vendors that work across different industries. And that way, by default, you'll be exposed to a lot of really smart people who are solving issues across industries. That's kind of been my Go-To. So I get bored very quickly. I don't stick around at any job more than a couple of years because I'm like, I'm done with this industry, so.

Speaker4: [00:25:26] But, but that's kind of but I also, like I said, if I want to learn some data, science is something that I don't I don't have data science on my resumé. And part, and partly because most of my work tends to be more on the traditional side of stats and econometrics than my most of my dependent variables are continuous, not binary. So I do tend to stick on that side of things a little bit more. But to learn like I'll be an associate faculty at Columbia [00:26:00] or any of these schools, they're very happy to have you. But I also learn a lot with these kids. There are a lot of companies who come in. You can be a part of capstone programs and you can also sit in some of those professor classes for free. You have a good connection with them. So I used to do that with if you have a good rapport with a professor, they don't mind you sitting in their class. So thanks. Those are things that I do, but I, I don't like. I learn so many languages by myself, but I don't seek and say I don't know the eight of us or this.

Harpreet: [00:26:41] Yeah. Yeah. It depends on what it is, what space and what problems are solved and things like that. Awesome. Thank you so much. Let's go to let's go to Greg. And then also, if we keep going, Vivian, I want to take a stab at this as well. I'd love to hear how people are doing this man. So this is a question is how are you keeping up? What are you guys studying? What are you learning? How do you reinforce basics, things like that? Go for it. Greg then. Then. Then Makiko and Vivian if she is down.

Speaker3: [00:27:10] Okay. So I know I caught the entail of telling of the customs question, but you summarize it in our predefined standard. Well, how do you keep up with with tech? How do you maintain the fundamentals? You don't forget them and what else are you working on, etc., etc. Hopefully I caught that. I guess for me I let many things drive me and I think NAVNEET You did mention that as well. One of the drivers for me is my professional work. So where do I work and what is my role for my role? I'm lucky enough to work on innovative projects like things that haven't been launched before. So it helps [00:28:00] me do two things, achieve two things to one, leverage the current tools and systems that are available on my in my work environment, but to also source certain tools that we may not have that can help us deliver more value. So the first one is a little bit easier because you can come up with different use cases at work and then tinker with these tools, existing tools, and then see if you can maybe hopefully bring some new insights or spin up some new ideas that help you get some competitive advantage. But oftentimes, you know, it's it's a tool or a process that is outside of your day to day that can help you achieve that. So working with, like, you know, just like, you know, those consulting companies with them can help you surface some of these techniques.

Speaker3: [00:29:03] But even before the tools, what I like to spend time on is really understanding like business. So I've been on that quest to better understand business strategy lately. Kind of like why? Why do we do things certain way and how do we go about it long term? How do we execute long term? Short term? What kind of tactics do we use in the tactics is where you will find out what kind of tools you need to use to execute those. So for me to stay ahead of the curve in terms of like what's happening, I connect to a lot of the I guess, you know. Newsletters. I know newsletters can have a lot of noise. Groups like think about Simeon sites, for example, when they announce a lot of these new tools [00:30:00] or new companies and things like that. So I have access to a lot of them, but I cherry pick what I want to spend my time reading based on what is aligned with how I can be better at work and how I can be better personally. So in terms of the fundamentals, though, this one is probably my biggest weakness in the sense of out of sight, out of mind. I'll give you a quick example. When I was before joining Amazon, I was doing a lot of like I've trained myself. I learned I don't like coding in the first place, but I find myself loving building dashboards but you know, beefed up my analytics skills and things like that.

Speaker3: [00:30:45] Move to Amazon. So now I'm moving where I'm telling people who are good at building dashboards what to do. Therefore you put me in front of a dashboard. Right now I will forget everything and how to do it. So it's like I have to rebuild that muscle if I get back to it. But do I need it now? I don't really need that fundamental. So my quest is to gain more fundamentals, right? So what are the fundamentals of strategy, for example? What are the fundamentals of data? Product management? What are the fundamentals of cross collaboration between teams, etcetera, etcetera? So trying to get gain more and kind of like deprecate I guess the ones that I don't need based on my goals and I and I'm pretty aware of your goals don't have to be static, right? You can change your mind any time, right? As long as you move towards it as you go. So it's an interesting thing. It's a dynamic thing. So it's something that you have to enjoy and also enjoy sharing with others too, what you learn. That's another growth point too, because once you learn and share with others and explain it in your way, that's like proof that you actually have a sound understanding, at least of the concept.

Harpreet: [00:32:03] Greg, [00:32:00] thanks so much. Get on the flip side, philosophical side of stuff there. I think about business and business strategy. What are you using to kind of brush up on that? Any books, any particular newsletter or podcast that you that you tune into?

Speaker3: [00:32:18] So yes, reading from books to interviewing people, asking for help directly, like doing one on ones. I even use my post to to get to get insights, right? So a lot of times I learn even on the job to write. So trying to understand what's motivating people. So there are a lot of resources out there that that can help you with strategy. And in my mind, I think strategy is one of those that you can, you can, you can, you can. There are so many things involved in that sentence, but at the end of the day, people run companies. Right. So when it comes to strategy, you have to understand what motivates people for the long term, in the short term. And that is something that you can either learn from reading an article, but it won't be enough. It's one of those you have to be in it. You have to test it. You have to make mistakes. You have to learn from these mistakes until you get better at it. So there's never the perfect resource, in my opinion, to learn strategy, you have to have a solid group with you to help you get go through that journey.

Harpreet: [00:33:40] I think so much I can recommend a couple of resources strategy first and foremost, which is this class all about strategy for data scientists that we check that out. Second, I interviewed Fred Maillard, who wrote a book called How to Be Strategic on the podcast called How to Be Strategic or How to Be Strategic for Data Scientists like a Strategic Data [00:34:00] Scientist. Fred Heller, definitely check that book out and that episode out. All right. So let's go to let's go to then I want to hear from Kiko and Vivian if she's down and then if you guys have questions, go ahead right there in the chat. Get to questions as much as possible.

Speaker2: [00:34:20] Don't say for strategy. I'm just going to give a plug. I don't know if every if everybody's got the same view as I do right next to me. But she works at Columbia. She might have some hookups. There might be a couple of smart people at Columbia who knows some things about strategy. Maybe one or two. So I might want to network with her and see if she can. You said you were working with people at Columbia, right?

Speaker4: [00:34:44] And the word on and off. It's been a few years I haven't done it, but I the capstone program is actually they look for people like us and I learned from them you don't really make any money. So just, it's just a lot of work. But I enjoy just working with kids. They have a lot of companies that come in that you cannot work with. Over comes there and NBC and a lot of smart companies come with really good big data problems that kids solve so you are a big part of it usually happens in fall but I can give you a couple of contacts. We look for folks like us. Yeah. But I haven't done it in a few years with.

Speaker2: [00:35:32] From a strategy standpoint. Greg, she's sorry. I hate this reverse thing. She's. She might be able to help you out with a few contacts, at least. There's some really smart people at Columbia. As far as what I read and what I do, this is going to sound really strange, but I get the best. Like Rabbit Hole deep dove information from Twitter threads on Twitter you can just stumble across [00:36:00] something on a topic that you're interested in, and most of the threads are questionable quality, but they'll point at resources and those resources are typically top notch. And then it's like I said, it's a rabbit hole. And if you're persistent enough, you can learn almost anything on these threads. And then following the people that that person says to follow, that they get their information from, and you begin to build just these networks and communities where now my Twitter feed is four or five different topics, and at any given time somebody can just drop an amazing thread of information or perspective that I can follow down. And it's every topic you could think of, you know, across the entire data science and machine learning field you get from time to time here, a couple of data scientists decide to go 12 rounds with each other and argue the merits of deep learning versus symbology. That's the recent one. That was fun. It's fun to watch really, really smart people like put the gloves on with each other because there's always something that you learn about both sides. And so it's there's so much to learn that way. And I hate to I hate to admit it, but I learned a lot from Twitter. That sounds kind of like something an ignorant person says, but you can actually learn quite a bit if you curate your feed properly and if you vet sources and look always for secondary, tertiary sources to support whatever it is that you're seeing.

Speaker2: [00:37:35] And that helps me understand trends because typically what people talk about on Twitter isn't the stale stuff. I mean, you have the here's how to break into data science, here's the basics that you have to learn. But on the other side, you have people like Jeff Pearl who are just constantly talking about some of the best, most important aspects of causal machine learning and providing detail. This isn't the stuff that you get in [00:38:00] a YouTube video. This is I'm answering a specific question that somebody asked, and that person is way smarter than I am. So I would never have even known that I needed to ask that question. And now I'm getting some information and some answers about and causal is one of the areas that I want to understand more about from an applied side. And that led me to figuring out that Microsoft is doing a ton of work. I found this two years ago. They're doing just a library's worth of work on trying to figure out how to put causal and machine learning together, because they're always talked about separately. And so they're trying to figure out how to merge the two fields, which is incredibly interesting. And I would never have found any of this if I hadn't been following people who find a question, answer it. And so I'm almost like off books right now. I'm doing a lot of research papers. I'm following people, following what they're thinking about, what they're talking about, what they're worried.

Speaker3: [00:38:59] About.

Speaker2: [00:38:59] Because that's another big thing is no one ever tells you what breaks in machine learning. And 90% of doing good data science is knowing what won't work and what you should never, ever, ever try to do in the first place. And what will break if you push it too far. And the other thing just a few months ago started going through some really formalized sources in the AI technology and AI policy space. And I've learned a ton from them. You know, there's everything from A.I. Pulse, the Aspen Institute Atlantic Council, a security initiative, you know, Stanford High, Wilson Center. You know, I've got like 50 of them, so I could go through it forever. But if you start looking for some of the more formalized think tanks and policy centers, it is amazing the amount of information that they publish and they publish it from a different perspective. [00:40:00] These are people who some of them are practitioners, but many of them aren't thinking about how do I implement? They're looking two and three years down the road and they're saying, okay, what happens if what do we do when? And those? I think for me, those have been the things that I've started to follow and started to learn because as far as practitioner, like a practitioner standpoint, I've got most of the information I need and I think what I've always seen as the most. Valuable is the types of questions I don't know I should have been asking and the perspectives that in this field I'm never going to get because those policy centers and those think tanks, none of those people work with us. We don't have any of those people accessible to us.

Harpreet: [00:40:49] Yeah. The Twitter has been amazing, 100% cosine that I've been I've been on Twitter more, I guess, regularly over the last few months, and it's definitely been one of my favorite places to go for information. I primarily go to Twitter for mostly marketing tips and copywriting tips and tips on how to write better and things like that. So that's been super helpful for me. Also, like he shot out, Bloom is pretty dope. He's got some amazing threads out there, so definitely check him out. Great, great tips then. Macgillis, let's hear from you. And then after is actually let's jump to some of the questions coming in on LinkedIn. If you guys want to, those guys quickly visit my LinkedIn profile, look at some of the questions. And if there's one or two things that you guys want to bring up in answer, let me know. Just let me know right there in the chat. Kiko, let's hear from you.

Speaker5: [00:41:46] Yeah. So I try to do a mix. So basically like kind of my criteria of criteria for learning stuff is it's got to either make me money or it's got to like help me build future [00:42:00] capabilities or it's got to feed my soul. I do not learn for the sake of learning anymore. Like, I don't know if it's because I pass that big 300 mark in life just hit me harder than expected. Then don't give me that. Look, I'm not three oh right now. I passed it. That was a few years ago. So I'm a little bit older than you think I am. I know you're giving me that. Yo, you're young and look. Yeah, but I would say, like, for data scientists, too, you know, it's rough out there. Like if you're especially for your junior because you don't quite yet have that filter of what is relevant for you as an individual, I think gets better as you get more, frankly, to the senior data scientist route or even as you get more niche. Like for me personally, I. I've actually been taking a book out of Mark Freeman's page of Mark Freeman's book, where essentially I reserve the last three or 4 hours for of the day for learning for my soul. So right now, I'm focused on on honestly getting better at communicating via writing and also figuring out the the fashion startup space and kind of incorporating a lot of the techniques into sort of my projects in my shoes.

Speaker5: [00:43:14] I keep promising I will post photos and I don't. I know, I know it's I know. Don't worry. It's it's it's coming. It's coming. It's going to be awesome. If it's when I'm not doing the feeding the soul work, it's either driven by projects at work. So for example, like we're doing a lot of stuff with sicced and ops and version control and that's kind of all I'm going to say. But that by itself already it's like a really, really big space to dove into, especially on the ML up side, because especially to what I find is that a lot like a lot of like you go out there and you think, you know, it's all been written, it's all been explained, it's all been taught. But [00:44:00] for example, when I was trying to find like good use cases of GitHub actions and in like an enterprise setting, I actually didn't find that much good written material. So sometimes the best material for learning is actually still the documentation like from the the package or the source or what have you with like a few sort of gems of like technical blog posts.

Speaker5: [00:44:19] So there's that big where it's like driven immediately by work concerns and then the stuff that's like the future capabilities. I used to make the mistake of doing like 80, 20, like 80% of all my studying was future capabilities. So things that sounded really hot, computer vision, deep learning, etc., etc., you know, but what I was finding was that if felt like I was constantly off keel, it didn't feel like I had a really stable source of understanding that I could really pivot off of. So now that mix is a little bit more like 60% kind of foundational tried and true techniques and maybe 40 to 30% more like quote unquote cutting edge stuff. I found that to me is very helpful. And what I personally do is I essentially set goals for the quarter. Here are my top three intents. This is where I want to like kick ass. This is where I want to, like, next level. And then I'll try to align sources of learning to that. Yeah. And then also just be patient too. Like, I kick myself a lot for not learning stuff. Like, get like. I'm still very bad. I get. And it's it's mind blowing.

Speaker5: [00:45:38] All the things that it can do or even things, for example, like cookie cutter, like it's a tool that's used in a lot of places. Really took me a lot of time to understand it. Like it's used a lot in web dev frameworks, but I'm not doing web dev, so things like that. And honestly, the best ways to learn are like signing up for hackathons. Signing up for, like, [00:46:00] those, like, boot camps or workshops that are like three or four days because you get condensed learning and it's good. And also writing articles. Like, I legitimately didn't think I could learn by writing articles until I spent 30, 45 minutes rattling off the difference between like a VM, a container and a virtual environment. So once I was able to do that, I was like, Oh yeah, I actually learned something from researching this like ten hour article. Kidding. It was like 30 hours to write in research. I just say ten because that's how much I scoped it for. But it's 30 hours. Yeah. So that was like a really, really good way to learn. I was I was shocked that all that documentation actually went through my brain. So yeah, yeah. Right now for me it's all about sicced ops. You name it. Platform.

Harpreet: [00:46:51] Writing is thinking. Writing is thinking, that is for sure. And it takes longer than you think it takes to write. I don't know how Vim pumps out so much writing content and it's all so thorough and so good. I'm like God to have two years like a super human. Somebody who's like a big part of my job is writing. That's what I do is part of devil's advocacy. I don't know. I can't pump out articles that haven't been out for years like Greg.

Speaker3: [00:47:21] No, I was going to say, it's like writing. It's like I'm fairly new to it. Like I listen to long form and I'm actually curating things that are already published. And even that takes me like 8 hours to go through four articles and then kind of like explain it in a sense, in a way that makes sense to me and also try to empathize with the audience that it makes sense to. So thank you, Harpreet, Vin and others who have given me some some feedback on on the on the new blogs that I that I put out. So it's it's been fun lately, but it's been. Thank [00:48:00] you. And I appreciate that. I appreciate that. And it's kind of like it's it's a super like it's a superb way of learning a lot like, you know, you learn about the history of Transformers, how we put language models to the top, you know, how now it's spreading to computer vision. Then what next? Right. So how can you combine two systems together? I mean, it's it's been very eye opening, right? So and I'm hoping that the audience will continue to learn with me, too. But I'm really enjoying it in Mexico. So I think you put the hours in, but at the end of the day, it will be very worth it. So yeah.

Speaker5: [00:48:48] I will say that I do find kind of interesting so and this might be because I came from like an anthro, an anthropology economics background. We're like history and context and the philosophy of why theories were developed is so important. But it's a little bit weird that like, I feel like sometimes people go straight to learning the tech and the documentation, but I feel like it would. And then, you know, everyone's like, I think there was like this good quote where someone was like computer science or programing. It is like the greatest expression of punk. Because it never matters. The history. People always just they go in. They think it started from right when they enter the field and they never think about that. There was this long history before it. And I feel like I feel like it's so fascinating because like at least in the anthropology world, there's even in social anthropology, there's such an emphasis on ethnography, on like asking questions on like understanding the stories, understanding like, like how things developed. And I feel like a lot of times when people are questioning, well, why does why was this technology like designed this way? Why do we use it this way? It's like. Instead [00:50:00] of assuming that like that fence in the middle of a yard was just built for no reason, maybe the better question is there is a reason why this fence was built that way. So maybe we understand what the problem it was trying to solve and how whatever was developed was a response to that. Like with data scientists, for example.

Speaker5: [00:50:21] And this is the thing that I found a little bit frustrating as a data scientist. Right. So let's take the example of like containers versus VMs, right? Any time you want to get an explanation about, for example, what a data scientist should use to encapsulate their project cleanly and safely, people always toss you this is the difference between a container and a VM. Well, guess what? Most data scientists aren't choosing between a container and a VM. They're choosing between a virtual environment and a container, or they're packaging something. So it's really interesting because the reason why that's written right, is because it's assuming an engineer in mind, it's assuming an engineering audience. But most scientists are not engineers. I mean, I'm sure they'll I'm sure they'll kick off a nice LinkedIn comment argument. But it's to me, it's so fascinating because I think the history is so important of why things develop. So when we're like retelling stories, when we're adding our own inputs, like, for example, like with your articles in your blog post, like I think that's so important because you're bringing like a unique perspective to that. You're helping to kind of explain the story of the why. And it's fascinating to me that more people don't do that. I love these like I love some of these people who've been in the engineering field for like 20, 30 years and they're explaining the history of like, well, how do we develop like. A server or whatever. You know, it's like, Ah, that explains it. That is what sets the stage for all this innovation.

Harpreet: [00:51:49] Remember, Greg?

Speaker3: [00:51:53] No, I was going to say, remembers when Twitter was coming out, like they put this like simple video about like how you [00:52:00] would just share, you know, you're sharing what you're doing throughout your day or I'm setting coffee. You know, I'm sitting at home reading a book hoping that, you know, your neighbor who didn't want to spend time walking and visit you would just read through your read their Twitter feed about what you're doing at home. I'm like, man, people are so nosy. Or Why do people want to share about so many simple stuff about their lives? And then now Twitter is this thing that, you know, is being used way more than what it was built for. Right. That's one example of what you're saying here, Michiko. Like as Ren was describing how useful it was. So. Yeah. Sorry. Sorry, Harvey.

Harpreet: [00:52:39] Yeah. No, I'll just say the history of it is so, so important. Like, I think you should invest a little bit of time in doing that. Like Kosta was talking about random stuff that we've been studying a couple of weeks ago. I was really, really going deep into the history of Silicon Valley and what makes Silicon Valley for me, and just hearing about all the interesting people that help make our lives fantastic. But there's a couple episodes that as data scientist, I think that if you're interested in history of ideas, you should tune into those. They're both on like student podcast. One of them is that Trevor Oliphant, who is the guy that invented numpy and sci fi, and he talks about that, that history of that package being developed. And you get such a rich history about how data science has evolved, how he took this package called numeric and converted it into Python and how like the history of it because he was in early Python days like Python one day he's talking about the history of Python, all that. The other episode with Peter Wang, who co-founded Anaconda with Trevor Oliphant, and talking about the history of why Anaconda is the de facto package manager for data scientists. We talk about virtual environments like a package management problem is something that is really hard to solve and we take it for granted. We just install Condor. But why? Why is Condor [00:54:00] the de facto thing? And you get the entire history of why it is that way, why it's so important in data science to have this package dependency problem is all super interesting stuff.

Harpreet: [00:54:12] Great conversation. Let's go. Let's go to you. Got your hand up. And then also there's a question coming on LinkedIn that that would be good to obsess about a computer vision or after after your comment to correct. Yeah. Just just kind of rounding out on that. That's a that's a fantastic call. I mean, one of the things that I almost forget that I do that I do quite regularly is I go back to like people who've been around for a while, like people I've worked with who've been in the industry before. Like, I mean, my previous company, I had two guys who had been in the robotics industry for 20 years, each as postdocs, and a third guy who'd been a robotics engineer for like 20 years. So just having them around, I was just like absorbing random bits of information that I didn't really realize I knew until this role where I'm working with more junior machine learning engineers who didn't quite get that same insight into this is why we use version control in this way. This is why we use a single source of truth for this and that, right? A little little nuggets of wisdom that I just I was lucky enough to find while talking to some of those more experienced guys.

Harpreet: [00:55:23] Right. But one thing that I find that I do quite regularly is I go back to books like any of Uncle Bob's books or pragmatic programmer, or there is this this is YouTube channel on YouTube that I think is thoroughly underrated. It's called continuous delivery. I don't know if you guys have heard of that one. I come across that one. It's it's this guy who's kind of been around a while and he's he literally he literally goes through this whole plethora of ideas like test driven development and just general good software practice. And he goes into this [00:56:00] is why it was created. This is the history of why we do things in this way. Right? And it was a really good way for me to like firm up some of my core engineering software skills just to remind myself, okay, these are the good things we do. Not because we do it, because we do it, but because it's the there's an actual reason why we got to that stage, right? So those are probably some of the things that I do consistently. But in any case, thanks everybody for all these interesting ideas. I think I'm going to try switching to that 40, 60 kind of split. So yeah, this is probably the most useful page of notes in my life for this month, for the next couple of months. All right. On that. I'm glad I'm glad to contribute to that.

Harpreet: [00:56:43] So question close up on the past. Our resident computer vision expert coming in from Paul Sanchez here. He's wondering how to effectively explain your ML projects during an interview with an ML engineer for a ML engineering intern position. This is specifically for a company that focus focuses on computer vision, so assuming it'll cover an image classification project from photos interview with Determined AI and I had to design a computer vision system and I've never done that before. I ended up getting the offer anyway, so I must have been able to bullshit my way through an answer somehow. Because if you're hiring for a intern and they did a computer vision problem, but what are the kind of questions you would ask them? What would you expect them to be able to explain? That's a that's an interesting one. So in the past, a lot of the time, what I've seen with internal level or grad level engineers who've come through with a computer vision project under their belt like either a capstone or a thesis project or something they've just done on the side, they consistently jump straight to the, Hey, I developed [00:58:00] this kind of a model for for this kind of use case, right? The biggest thing that I like to see is I actually like to see that they're able to understand what the what the purpose was. So if they go back to that classic model of explanation, right.

Harpreet: [00:58:16] Start super high level. Hey, here's the problem that I had to solve. Here's the opportunity in that problem. Here's the technology I used to solve it and then build it back up and then go, okay, why did I use this technology? Right. The main thing is I'd I'd be curious to know is how well they can articulate the reason they chose a particular network. And to be honest. Probably at a undergraduate level, or even if it's like your masters, but like you're doing a massive data science after coming from a different field entirely. Often the reason you chose a particular line of thinking like say, Oh, I started looking at VA is instead of Gans for this particular task, it's probably because someone recommended it like a lecturer at the university or something. Recommended it. Right. So can they articulate why they're using that particular approach and are they able to compare it to other approaches? Essentially, when you're looking at an internal level, though, how much are you actually going to know to be able to articulate that? Asking them that question is Why did you choose this versus know again or something else like that, or a super resolution at work or something? Why? If they can articulate that, then I know kind of where their skill levels are and their understanding levels are and how much coaching they're going to need. That's the that's the main thing you look for when you're assessing a project.

Harpreet: [00:59:40] Right. What skill does this person actually have? You're not that that's not the thing that says to me at least that's not the thing that says this person is horrible slash. Not horrible. Right. That's the thing that says if I hire this person, this is how much time I'm going to have to spend with them bringing up their basics on software engineering, the basics on machine learning or computer vision. Right? [01:00:00] Whether I want to hire them or not comes down to the kind of energy that they bring, how much they're willing to learn. You know, a lot of the more soft skill kind of components. Right. I'm really curious. I mean, I've only done a small one handful or less of interviews on my own, so I'm really curious to see what everybody else thinks. Definitely. So what I'm saying is don't jump right into defining architecture. Put a little bit thought into the why behind it. How did you come to that conclusion? How did you come to choose that specific architecture type of thing? I got random questions for you on computer vision because let's say, like, if if first of all, what constitutes an outlier in a computer vision problem? Are you talking? Okay. So if you're talking about anomaly detection, that whole concept gets a bit weird because you want outliers in the data because you're only trying to detect the outlier. Right. But I guess what it's that's that's quite complicated.

Harpreet: [01:01:15] It depends on how you're processing your features. Right. If you've got a feature chain that's going to focus on the textural information that you've got, then your outliers are, Hey, do I have data with different textural understanding of the process? But essentially what I look for is, okay, let's say I'm trying to solve a detection problem and I'm trying to detect, let's say, military vehicles and civilian vehicles as two separate classes right on the street. Essentially, what I want to see is a whole balance of data across all the different situations in which I can see them. Right. You want to essentially, I don't see many ways of doing that without simulation. [01:02:00] So you're trying to make sure that you're covering things so that you're not seeing where the outliers. Outliers are basically essentially turning into anything that comes up in your data that you wouldn't have seen. It's it's very, very hard to deal with that because it's such a broad world, especially when you're talking about like open field, open world kind of computer vision. Like, like we're solving this whole problem where we had to get a aircraft to kind of land itself and select whether it should land or not. That was part of our control loop feeding back to the actual aircraft controller. And one of the tricks, one of the things we found was like, there is no way we can conceivably train this thing to detect if there is a cat underneath there.

Harpreet: [01:02:45] What? I mean, like, at some level, computer vision is not the. Solution to that answer. So what we did was we stuck different technologies, right? So you use computer vision to tell you the stuff that you're not dealing necessarily with outliers. You're dealing with the facts you actually care about. Like, is this terrain okay to land on? Is it bushy? Is it Rocky? Do you have a lot of is it water? Is it grass? Is it. What's that terrain type? And then you use totally different technologies to detect transient things in your field of view. You use like radar and and all sorts of other things. So I kind of pulled that back to the first principles. What is an outlier in your situation in that particular context? The outlier is, oh, it could be a cat walking past a person, an ambulance. I can't conceivably train against every possible thing and have a network that could reliably do it. So I need a different way. Sometimes computer vision is just not the right answer. But what do you mean by outliers? Like what do you consider outliers? Know we have statistical tests to say, okay, a particular data point is the outlier using, you know, if you have a regression problem type of thing. Right. So I'm curious, how would that concept mean in computer [01:04:00] vision? Like how you measure that that concept is of an outlier, right? In that sense, it would be quite similar.

Harpreet: [01:04:09] Like you'd see like so let's say you're doing a cat detector or a cat slash dog classifier, right? Outliers are still going to be statistically those data points that just don't line up from a performance standpoint. Right. That cat that every network seems to think is a dog. Right. Essentially, it is the same thing. It doesn't make a difference what the domain hearing. And here's maybe a basic question that maybe I could probably use it myself. But does it matter the order in which your Marvel architecture encounters the data in those batches in a given epoch? Like I got to know that that. Intuitively, in my mind, it does. But I don't know whether I can explain or justify that stance. You know what I mean? Like, intuitively, it does. Like if you train, you know, the first half of the epoch on on just the same handful of images, your weights become less randomized, like we always saw with randomized weights. Your weights become less randomized as you feed it if you over fit to a particular subset, right? So it's like taking it over a fit network and then returning on top of that. In my mind, that's the closest that I can rationalize that. But yeah, that's the reason we always just use random shuffles for injuries that cause the big you. Greg, you had your hand up a while ago, so we'll come back to you there.

Speaker3: [01:05:41] Yeah. Yeah, no problem. I mean Kiko has me to help me understand and also to my question was what is the difference between a virtual machine and a container? So I find myself like go confused all the time. And then also, you know what would be like use cases where, you know, either [01:06:00] software engineer or a data scientist, which is one versus another. So that would be super helpful. But Michael has already been enlightening me in the in the chat, so I don't know if you want to give us a powwow or.

Harpreet: [01:06:16] Yeah, if you wouldn't mind to if you're available, that'd be great.

Speaker5: [01:06:22] I and I really wish this blog post was published by now. It doesn't need to be. And I think this is where. It's funny, there's so much data science content out there, but I feel like there's not a lot of. Quality software engineering translation to data scientists material, if that makes sense. Right. Like, and I think this is one of those cases where like so the most mind blowing on Reddit of all places because sometimes thread it's a cesspool, but sometimes it's got like insight and info. But the most mindblowing thing someone ever said to me was someone had written, right? Because someone was like, when would you use a VM or or container in the cloud? And someone was like, idiot. Like containers run on VMs in the cloud. That's like how GCP and AWS and Azure deliver their services. Because if you think about it, right, like as customers of like a database or GCP, like. Well, look, I'm not on like the t like pure sorcery in fireside, right? But if you think about it. So in terms of use cases, any time you want some kind of like layer of abstraction away from the physical computer. So on a computer like with a VM, you can determine the underlying OS and [01:08:00] more importantly, you can kind of specify like the resources.

Speaker5: [01:08:03] Now one benefit to like using cloud for data scientists, right, is that a lot of times as a baseline, you don't want to really think about what resources you're using to deploy your model. Right? Like you don't want to have to set limits on your Docker containers and all that. But hypothetically, you could you could have this arrangement where you have like a virtual environment on in a container, in EVM, on a physical computer, and you could have multiples of those and you could even have like a virtual environment and a container on a VM and you could do all sorts of things. It's just really like it's like onions. It's almost like layers of abstraction. And this is one of those things that makes me really think about how how there isn't like a lot of quality explaining like kind of software engineering to data scientists because a lot of times the information that is presented is a comparison of containers versus virtual machines. But with all of data scientists, that's not that that's not the abstraction level that they're necessarily or we are necessarily operating at. Right. The difference between container or VM is relevant. If you're more on the ops dev ops side and you're trying to figure out, well, how do I like publish or deploy a bunch of these models? And even then, you still want to make it as easy as possible.

Speaker5: [01:09:20] So in terms of kind of the use cases, it I feel like I don't feel like there should be a reason why a data scientist would ever need to spin up their own VM. But there's a lot of reasons why they would need to spin up multiple containers. And if they're deploying the containers to like Amazon or GCP, then it's going to be operating off of VM, usually configured by like our like cloud enablement team or something like that, but usually like set the limits and the main sort of selling points are if you have a VM, you can define the OS, you can, you can manage resources [01:10:00] a little bit more. Directly with the container. You can't necessarily determine like the Oz necessarily. And it's going to be sharing resources like with other containers on the VM. Yeah. This is an example of learning by writing, because honestly, there is no other reason why I would have ever cared about this topic, except I said I would write it and then I wrote it and I was like, Oh, this is a mistake. But I learned a lot.

Speaker3: [01:10:28] That makes sense. That makes sense. I'm looking forward to seeing your post, though. And for a container, for example. The way I understand it too, is when you want to guarantee little to no failure, when you know that the environment may change. I don't know if that makes sense, but kind of like you want to guarantee that where you're serving your endpoint, your your your model won't crash. Maybe. I guess my question to that is, is a container does the container facilitate integration with an application, for example? So a container would have an endpoint and then that API would connect to an application easily, can be easily managed, removed, etc., etc.. And then you have, for example, let's say you have an application that calls multiple models, right? So each model can be served the container environment and each of those are an API endpoint that is connected to the application. Is that is that a fair understanding of how they would work?

Speaker5: [01:11:37] Yeah that's definitely like that's a pretty well established that that is one of many acceptable patterns out there and I think something that like this goes back to the whole like sometimes. And so I think the general guidelines and and harpreet definitely correct me on this and, and Mark and other people on the call. I think the general guideline [01:12:00] is you want one application per container. Containers don't offer as much security as BMWs, but it's ideal if you have multiple applications and you want to talk with them. And that was essentially what Kubernetes was meant to solve, right? It was meant to solve this. How do we get these different containers to talk to each other? But yeah, that's a pretty well established pattern, I would say, because essentially what most people don't want to have to control is they don't want to have to control scaling up instances on compute. Yeah. And but it also depends. Right. So for example, another pattern is you batch, you batch predict like in, say, redshift, for example. And then a company can call up those values because at the end of the day, like a table, you can kind of treat it more like a hash as opposed to having to like cache the values. But there's a lot of these patterns which I'm going to be honest, I don't understand 100% well, which is why I'll be doing two weeks of a front end boot camp for the next two weeks, Monday to Friday 9 a.m. to 5 p.m.. So yeah, but that's 100% the use case.

Speaker3: [01:13:16] Thanks, Mr..

Harpreet: [01:13:17] Coleman. Marketing is ability to go from and see what's good.

Speaker3: [01:13:22] I'm not even going to find what Michiko said. Lava went over my head. But I do want to stay like a potential use case. Because, Greg, I know you like finding these use cases and digging in is a group that a company that uses Docker and Kubernetes very heavily is Netflix. Their culture is completely centered around that. And I was never I was talking to someone who used to be engineer at Netflix. And what they described was essentially is like Netflix, just a bunch of little startups masquerading as a large company, hence why they they like using using [01:14:00] microservices so much with Kubernetes to like bridge all their kind of controlled chaos together. And so, you know, I cannot speak to the technicalities of it and I will defer to Michiko and her amazing blog that's going to be coming out soon. But I think once you go through that blog and you're like, All right, I want to connect to the real world, know from the theory to like what's happening in industry. I think Netflix might be a really great use case to explore a little bit more in.

Harpreet: [01:14:33] Into the hands down the best blogs, engineering, blogs, Netflix. They got the best technical engineering machinery, engineering company blogs. Then any any input there to raise questions about containers. Because I see you you've got a nice container behind you to you know that.

Speaker2: [01:14:54] Anybody lost the container. Just describe the contents and I'll get a return to, you know, my, like, my VM knowledge is so old, I've been able to use a s, I've used Kubernetes. He's a bit, but it's almost like it's so easy. You don't really have to understand it until you break something. And then the best thing to do is call someone smarter than you and not try to figure it out unless that's your job. So that, that's really like my only input into this is just be really careful because you can oh god, the mistakes you can make. And especially when I was talking about security. Yeah, yeah. You can leave some stuff just way exposed on an VMs and and containers, you leave a lot of stuff exposed. So it's just, I don't know how much a data scientist really needs to understand it, like from a practitioner standpoint, but it's really good to understand what you can do with it so you know what to ask, because when [01:16:00] you're building models, building it for a particular environment, you don't do a whole lot of that. But it does help, especially when you get into optimization. You can do your ML engineers a whole lot of solid favors and save days of refactoring if you just understand the basics.

Harpreet: [01:16:21] That's like a little bit of history out there.

Speaker5: [01:16:24] So, Michael Eric, who? He runs confetti. He wrote a blog post called MLPs. Tooling is a mess, but that's to be expected.

Harpreet: [01:16:36] Yeah.

Speaker5: [01:16:36] That's a really great one. I had a chance to meet him, actually, him and a few of the other guys at Meet Up last week. That one's a really good one because I do think that like. It's man like tooling is a mess. And I do feel like a huge part of it is because there are so many different ways, like there are so many different ways to kind of meet some of the same needs for deploying models. And part of it too is also where companies are at in their maturity. So for example like so it's funny. So Chip playing, right? One year she wrote an article. Here are the top ten things geoscientists should know. She include Kubernetes on their right, and then a bunch of people like echoed that thing of like, Yeah, DeSantis should be full stack. Then a year or two years later, she writes. That was my bad day. A scientist shouldn't have to know. And I think part of it is that, like when you're at a company that, like, isn't super mature. Like, you do have to roll your own stuff, right? Like at the other day, you're getting paid to like do a job. And that job is to put out like machine learning models. And so sometimes people have to roll their own stuff. But like, and I feel like that's honestly what machine learning engineers are. They're designed to set roll their own stuff.

Speaker3: [01:17:53] Hopefully successful. It's like I think I think at least a high level understanding is important for data scientists. [01:18:00] Right. Like the other day, I was reading something Cuban and when I hear it, I hear kids. And then all of a sudden I'm hearing something about K three. So K threes or more of like devices, tools. So if you're a data scientist, you want to deploy on the edge and you want to leverage Kubernetes. You have to look at K three because it will accommodate the size of your model and make sure that your environment doesn't fail, etc., etc. I didn't know all of that, but you know, I'm thinking at least high level, you need to understand the difference between the two so you don't try to use K eight for an edge device, something like that. So cool.

Harpreet: [01:18:38] Yeah. And they're scientists, you know, but data engineers, machine learning in the years, that should probably work. Yeah, but the history, look, we used to live in a world where every time a business wanted a new application to have to go and buy a new server, then this thing called VMware came along, and that enabled us to get more value out of the existing assets. Right. And then after that, there's a thing called a hypervisor, and these are supposed to be newer, more efficient, more lightweight. But then it became these more efficient lightweight. And is this thing called containers? So containers have been around for a while and doctors kind of the people who have brought it to the masses. So. That's good history in a nutshell. There's a question coming in. I think I want to pass this question. I think it's a good question, Jane or Mickey, actually. Any resources to help solve MLK studies and interviews. I feel like you have a video about this, about case studies, interviews, machine learning, case studies interview. This person you wrote says or Rosie says that she has difficulties with that part and it always makes her fail in interviews.

Speaker2: [01:19:57] I guess I might be talking about [01:20:00] the wrong kind of case study. But what I do and this is I mean, I love doing this every once in a while. Watch a financial channel and MSNBC has tech check. Bloomberg has. Look at tech hour where they'll talk through and you'd think it's like, oh, they're just talking about investments. But they get CEOs on they get analysts on who will talk about the problems in the industry that need to be solved. They'll talk about products from an analysis standpoint, and you can sort of glean case studies of what companies are trying to do, what problems they're trying to solve, what their customers are looking to them to do, what challenges their customers are having in adopting their solutions. Why aren't they buying? Why aren't they getting the kind of penetration into their target market that they want? And you can from there kind of walk forward and say, okay, so when I talk about a case study, that's actually what they're talking about is here's a problem customers have, here's a problem the business has.

Speaker2: [01:21:05] And you can walk those forward and say, okay, now what would I do with data to help solve not obviously not the the whole thing, but what could I use data for? What could I use a model for? What could I do as a data scientist to help with this problem? And then how would I present that to somebody? You know, am I talking to a product manager? Am I talking to maybe even the CEO? Am I just trying to talk my CTO into this? How would you how would you do that? And it's it's a good kind of role playing, scenario based thing to do before you go into those types of interviews, because you'll be able to you'll be able to do a pretty good analysis. And at least when you get those random off the wall, hey, here's a use case you've probably never heard before. You at least exercise that muscle of hearing, breaking it down into a data problem, coming up with a real high level feasible solution, and then figuring out how to present [01:22:00] it.

Harpreet: [01:22:02] Then. Thank you so much. Akiko, any input there or anybody else if you guys want input here.

Speaker5: [01:22:12] Yeah. And I've actually yeah. I would also love to hear Mark's response to on because I know I think you guys are like interviewing right for data scientists and I'm all people I so when I hear a case study I kind of similar to Vin like I think of a business case study like the kind I would do if I was interviewing for like BCG or McKinsey. Now what? People will do is like system design interviews. We give you a thing here. So for example, some e fashion company I interviewed for like two years ago or whatever one year ago, they were like design a computer vision recommender for. Clothing based off of if you're a new customer or if you're a geo location or like near a store or your attorney customer, and we got some stuff in your shopping cart. So there it was like they were, to be frank, a lot less interested in the algorithm and they were a lot more interested in what solutions I would recommend to achieve that goal. So that's one kind of style this system designer reviews where it's meant to pattern off the engineering ones and then there's more like the take home ones. Yeah. So actually I would be kind of curious cause I've done both and whatnot, I think. In terms of like being successful. I feel like it's a combination of you should understand like basic machine learning algorithms, you should understand basic machine learning use cases and then you should know a little bit about the domain of the question being asked. But I'd be curious to hear Mark's take on it, because I know for us we've struggled on interviewing candidates a little [01:24:00] bit.

Speaker3: [01:24:01] Yeah. So for for whom? I would argue that our focus isn't fully on ML right now, and so we don't really interview for that of anything. Like we say, like I'm really heavy. Our concern is more so like, well, would you be happy here? Because we're creating some infrastructure to be like heavy ML stuff. So that's why it's more so on analytics and more so and not just like, like summary stats, but like actually like do you know your statistics and experimental design kind of stuff? But speaking from my experience where I did interview for ML Engineer Roles almost two years ago when I was interviewing, other times I was taking interviews just to brush up on my skills and I've had two ML models. I'll be completely frank, I suck at hyper was like, that's great bad models. So how I end up in these interviews and end up to the final round of interviews is that I'm really good at writing clear code and my thought process and the considerations I would take if I had more time. And so when, when I do do that and when I think about case study, I think like a take home exam or a live coding exam where walking, talking through things. And so, you know, I would argue don't get caught up with trying to create the most perfect model because you're going to be staying up really late and be miserable doing that instead. Really research kind of like, why? Why would you make certain decisions? Why would you split your data X, Y, Z way? Right. Why do you choose this model? Why did you create these certain features? Get really close to the whys and like the trade offs you made of of doing that, because that's going to show your sense of like the type of question you're going to ask me, work with them.

Speaker3: [01:25:42] Because when you actually working for the job, you're not going to be so many projects and be working with a team or at least have resources, fingers crossed. And so, you know, I would I've been able to get past that case by case study interview, even though I'm not the strongest ML person, simply [01:26:00] because I'm very methodical, explaining my thought process and why I chose certain things. And what's happened actually is when I go into the interview, I'd be like, Hey, yeah, my model sucked. Like, here's, here's, here's the points where it sucks at. This is where it's like, Oh, we're fitter or fitter where maybe. And like, if I have more time, this is what I would do. And that leads to a really fruitful conversation where I go back and forth with the mechanics, and especially if they're more, more senior than me, I actually learn a lot from them and be like, Oh, you approach it that way. Oh, I see that now. And those are the kind of things that, especially if you're struggling with them, you can leverage that as a as a kind of a selling point of like, wow, this person is really coachable and great to talk to. And yeah, they have some mistakes, but they were able to really recover and learn fully from them.

Harpreet: [01:26:53] Great tips, Mike. Thank you so much. I appreciate that. Great advice. Great insight. So this question for all the consultants in the space in the room here, because it's a question about starting a consulting. So what would be what would be the best starting point to start your own consulting business in data science and the machine learning space? I want to build a small team who can help solve business problems. It's good to to end first because I think you might have some expertize in this area. I might be wrong, but.

Speaker2: [01:27:25] Yeah, I've run my own consulting business for ten years. I would advise you not to get into this space, mostly because you'd be competing with me. So don't. Just don't do it. It's too hard. There's no money in this space. That's why none of us really. You know, there's no reason to come in. Don't. Don't do it. But it's hard right now, and I've got to be honest about that. It really is harder. When I broke in and built my consulting practice, it was kind of one of those things you could stumble your way through and you'd be fine. But now we're at the point where [01:28:00] there are so many small consulting companies, and then there are like these monster consulting companies that are out.

Harpreet: [01:28:08] There.

Speaker2: [01:28:09] That you have to find some kind of a niche, some kind of advantage, something, I mean, really just something about you, your background, the staff you can pull together, the maybe you can find an underserved niche. It's something connections go a long way to. If you know people who are likely to hire you, that's going to be the best launchpad that you can have. If you already have people who you've worked with who would love to bring you in and have you staff up a team for them, that's a huge that's a huge help. But like I said, right now, it is really I've watched. Five people who have tried it over the last two years and didn't make it. And it's not because they weren't smart people or good data scientists or couldn't find the people that they needed to stop projects. It's that if you don't understand the process of getting a contract, of becoming a vendor, of even figuring out who to talk to, to start the conversation, where to go to find contract opportunities, because it's not like they're broadcast out there. It's if you don't have a network where you can start pounding through the network to get clients, it's really, really tough. But one thing that I've been telling people is about micro engagements, and I've had some success with this, and I could see a lot of other people having success where you offer your services for an hour, 2 hours, 3 hours at a time.

Speaker2: [01:29:50] Somebody just goes online, clicks a link on like a Squarespace. Like that's what I got out of Squarespace. And they book 2 hours with you, one hour with you, 3 hours with you. [01:30:00] And it's low risk for them because it doesn't cost as much as booking you for like 160 hours or 40 hours. You can deliver value very quickly. You can almost it's almost like you're selling yourself, but also solving some of their problems pretty quickly. You can do small projects. You can do evaluation of their current projects. You know, things like code reviews, model reviews. You can help them with ML ops types problems, types of problems. You can help them scope. I mean, there's a lot of things you can do in two or 3 hours that would demonstrate value and you can build up a client list. You can also get some referrals. This way you can get some people who will recommend you so that you have some sort of reputation online, and you can do all that while you're working. So it's not something like you have to quit your day job in order to start doing this. That's the one way I've seen that. I think you can overcome some of the major problems, but at some point in your very near future, you're going to run into competitors who are just monsters, you know? And it's it. That's where it gets.

Harpreet: [01:31:06] Hard.

Speaker2: [01:31:07] You know, if you don't, I'm lucky I have ten years. You know, you don't normally if you don't have a ten year old business and that long of a reputation in an industry, a big player shows up and competes with you for the same client. And you're I mean, there's just no way you can't compete. So, you know, there's definitely ways to do it. But I've got to say, be really cautious. Don't don't put yourself in a position where you may lose your life savings.

Harpreet: [01:31:33] Or.

Speaker2: [01:31:34] May not be able to pay the bills or something like that. Because I've seen a couple of people do that. And it's it's hard. It really is.

Harpreet: [01:31:43] Mark, want to hear from you on this because I know you kind of venture off in that same kind of kind of domain.

Speaker3: [01:31:49] Yeah. So, I mean, I think I talked to Ben like probably like six months ago. And the micro engagements are exactly.

Harpreet: [01:31:55] What I'm doing right now.

Speaker3: [01:31:57] I have my day job, which I love, and I'm learning a lot. [01:32:00] But also, you know, I'm an entrepreneur at heart and I've tried before. And I was like, you know what? I keep on trying to go too big. Let me do something small. I really especially last time, I just figured out there's so many gaps I had as want to fill that in. And so consulting and these kind of side micro engagements allow me the flexibility to pick and choose which gaps I want to fill in for for the time being. And I have like certain goals for a quarter. And more importantly is that I'm not trying to grow to some massive scalable consulting business because that's just not the business model. That just brings me joy. Some people are like, Yeah, that's why I love like one of my mentors who does this for like PR consulting for school districts. And his dream is just growing that scaling that consulting business. Right? That sounds miserable to me. And so I won't be doing that. But by doing all these small micro engagements, I'm learning all these different use cases, especially when you're more focused on the strategy side of things and talking through their problems. I'm learning about all these different openings in the market because I love building product and I want to build a scalable product. So for me, doing this consulting on the side is one that brings me some extra cash that can save and flip into building my own product. But also I'm talking to a bunch of people and learning about their pain points and just collecting all the information as it was like user interviews.

Speaker3: [01:33:28] So when I do, when I am ready to build a product, I'm like, well, I've worked for like 1000 people over the past three years and putting a random number out there. And, you know, this is the pattern I'm seeing of like a huge pain point that everyone kept on bringing up. And I feel like I already have 1000 potential customers because I've already talked to this or pain point. And so that's the strategy I'm kind of kind of using. And then like the next question is how do I bring people to bring into micro engagements? This is where LinkedIn comes in. My content is my marketing. I spend a lot [01:34:00] of time on it, not because it feels good to have views and likes. Like those things don't matter. And what matters is the connections I make and the reach I'm able to get to find those special connections because all my engagements have been inbound. They come to me and my LinkedIn DMS, but I'm putting in 5 hours a week minimum creating content. So it's not like a free lunch kind of thing. And so by having that LinkedIn content, it can be YouTube, it can be whatever it does. A couple of things. One, it essentially drives the top of my sales funnel. People are aware of me, they see my background, they either DM me or they go to my website and my website. I have that links to the square where you sign up, boom, sign up for for service. You know, sometimes I don't even know the person to sign up.

Speaker3: [01:34:54] I'm like, I guess I'm doing work now. Great. And we figure it out from there. The other aspect of it, going back to you want to build product and one of the challenges of being a founder when you're trying to get investment is being a first time founder. It's high risk for investors. They're like, You don't know what you're doing, which is true, and there's no track record. So you have a couple of options. You bring in someone who has had a successful exit before, which I've done before, and that that worked out pretty well. Or you de-risk yourself by building products maybe on a smaller scale or having a personal brand that there's a lot of trust built within you, which I'm currently doing through my LinkedIn content, is building that personal brand and awareness and and quote unquote thought leadership. So that way, like when they search me up and do their due diligence, they're like, Oh my God, who the hell is a smart guy? Like, you just stumble into our office. Of course we'll give them money. And so that's kind of the thought process of that. And so I really love Ben's advice of like starting small, doing these micro engagements and build slowly, because then it gives you options. [01:36:00] Like, I might change my mind and next year I'd be like, Actually, product sucks. I want to scale of consulting thing and because I started slowly and have these building blocks in the foundation, it's easier to jump into that and just like going to.

Harpreet: [01:36:14] You know, it's actually different for then we go to Mexico.

Speaker4: [01:36:19] You know, because because I'm just sort of crying. So my consulting thing is like about six months old. So it's still like starting out. But what I've learned or I'm trying to figure out is what madness is. And I've done a lot of things in 15 years. So when I was starting out, I was like, I can do this, I can do this, I can do this, I can do this. But then I'm now I'm like, Who's paying me the most? And then what? You know, and I have some possible real clients, but they just it's a really long sales journey to get to them. When is looking at you. Now you get it. But I'm also trying to position myself in a way that how can I not be replaceable easily? And what is what are those one or two that I know that I'm good at? So my specialty is digital and marketing analytics. And within marketing analytics, there's a lot. But now I want to focus in on a couple of things. But what I've learned in this process is that there are all these vendors who do these things that scam basically, they basically scammed them in these. So when you know what I'm talking about, you know, hey, by my model, I'll give you this product. It costs you half a million dollars, blah, blah, blah. It's nothing for them. But the problem is they're scammed into this. And three years later, they realize they've [01:38:00] spent $500 million in the product and there is no consultant to guide them along the way. And I kind of I'm trying to figure out what Manisha is. The problem is there aren't many people who do that. So they don't know. How to do that. So I think the part that I'm trying to get to is that I think this first couple of years are really going to you know, I think I figured out my niche somewhat, but I do not have a client yet to say I got it. So that's kind of the challenge. So when don't worry, I'm not competing with you. I do very different things.

Harpreet: [01:38:44] Can I jump in anything?

Speaker3: [01:38:48] Sorry. Unless you have something about the niche market goes and shut up.

Speaker5: [01:38:51] No, I was literally going to say I yield my position to Mark. I'll come after.

Speaker3: [01:38:57] I was going to say something that I noticed that Ben does that worked well on me and hence got me through his sales funnel is that he uses his content to educate potential salespeople. He does it really well. And so before you even like the first conversation about like, can I potentially work with you? I've already been educated about his process through his content, so I would just encourage you to go to Ben's LinkedIn and just read through all his posts and realize he's playing chess while we're all playing checkers.

Harpreet: [01:39:34] Og in the game. Up at. Just go for it.

Speaker5: [01:39:42] Yeah. One thing, I guess for people who are considering consulting, two things I would offer up out there. Number one, my general advice to people is don't jump into anything unless you know you can pay the bills like the [01:40:00] living bills. So it's really kind of good. And I think honestly, like the trap is like and I think, Vin, like you mentioned this was it discounted consulting? That is like one trap that people can get into where like if you're working full time and you do projects on the side for startups and you don't charge for it and all that because you do have like a living job, you have a job that pays a living wages and all that. Or it was a Ben Yeah, it was Ben. So I think that's one, one potential con of like having a full time job while doing consulting on the side. But at the same time I feel like that helps ride out some of the disappointments because like I've had, for example, like in the last couple months, I've had some clients not for consulting but for other like paid gigs like cancel. And I'm like, Whoa, if I was really depending on rent for this, I would be horribly disappointed. Instead, I'm just kind of annoyed and maybe I'll just mute the next email from them. So. So that is a thing. So I think like Mark strategy is awesome. I think the other part too, and this was kind of eye opening. So there's this one guy who is like a YouTuber blogger, whatever.

Speaker5: [01:41:15] He he does some wild stuff. He's constantly the one as a as an X Facebook employee and as a millionaire. My wife left me and took the kids and ran and I got fired from Facebook and all that. He's got some wild, wild stuff out there. But the thing that was really interesting that was almost worth the 20 hours of watching his stuff of nonsense was when he said, a lot of times people assume that you have to be an entrepreneur or you have to be a9a fiver. And he's like, that's a very binary mentality. And also to like people go through different stages, like in different parts of their life. So for example, when you're young, it might make sense to. Go very risky or whatever. Know if you have family, maybe [01:42:00] not whatever. But you can always like flip back and forth between being an entrepreneur and being a full time employee or even being both. Your experiences working for a company will help inform the decisions you make for consulting, and all the people in problems you solve in consulting can help kind of inform projects for a company. So that's the whole thing. That's the only thing I would put out there is that sometimes you don't have to like go full bore and quit your job to go do your own consulting firm. You can do the consulting well, keep a job, you know, and also like always make sure that you're paying your living expenses for sure.

Harpreet: [01:42:42] Go back to the.

Speaker2: [01:42:45] Yeah. Just real quick on the niche part. And when you said that you're trying to talk people into understanding that they need something that they pretty much don't understand they need yet. What you want to do is that can really mess you up because you'll take a ton of phone calls and a ton of sales calls, and at the end of the sales call you'll get a yeah, yeah, totally, totally. You'll never hear from them again. I spent my first year like 2012. That was 2012 for me was everybody said, yeah, no, that sounds amazing. And never I got one, two clients the entire year and I did a ton of sales calls. And so if you filter out, you'll save so much of your time just really filtering people by how bad are they hurting? And the the people. Once you find your niche, you'll realize that people are just coming to you with this horrible pain point that they've seen you solve once before. And it's really just that if you can solve it for one customer, one case study, and you can explain it in such a way that the company goes, Oh yeah, that's [01:44:00] this thing.

Speaker2: [01:44:00] We've had it forever. You mean you actually have a solution for it? You'll find companies that have almost given up because you're they've never met someone that could solve that problem. So they just gave up and thought it was a problem. They're going to have to just deal with, like a cost to doing business almost. And so if you can if you can find some really just find the people who hurt bad and they'll be the ones who resonate. Your message, you'll just hit home. And this is what happened with me with my first client was I found somebody who was in a lot of pain and all of a sudden they said, Wait a minute, so you can wait. How would you do that? And you'll hear, I'm serious. It was this moment across the table. Wait, how would you do that? And that's kind of when that clicked for me that I should have been filtering better because I was wasting my time with just a lot of people that would never. Yeah, they got it. But it didn't hurt enough yet, right?

Speaker4: [01:44:58] Only for those people.

Speaker3: [01:45:02] I'll go ahead. I just had a follow up question for for Vin on that. Go ahead.

Speaker4: [01:45:07] No. Go ahead. Sorry, guys. I have to go anyway. Okay.

Speaker3: [01:45:11] I was going to ask then because you said she was talking about getting like qualified sales leads. Is how are you filtering out for that? Is it is it through just not taking meetings or are you like saying like actually want a sales call? You're going to I'm going to charge for this. And that way you only get people who are serious, like, you know, how are you creating kind of like that? Like, you can you can get a whole bunch of leads, like in, like through the content. But like again, a lot of there's a lot of leads I get that are just like, I want pay for time to be a student for you. And I'm like, that's, that's like a one off thing that's not going to be like a lasting engagement, right? How do you how are you like getting those qualified leads to, like, figure out that system? Now your source. Maybe tell me I had to pay for that.

Speaker2: [01:45:59] No. The first [01:46:00] thing you do is know who you have to talk to in the business. Because a lot of people will talk to like a director. A director can't sign my check. So don't talk to a director. You that's the that's your first piece is qualified because you're going to get a lot of those people who are at a level where they feel the pain because they're on the ground floor. But they aren't someone who can sign the check. And so you can you will spend an hour convincing that person who will then have no idea how to convince the person that they have to convince. Who then has to convince another person. And then maybe you get brought in and you aren't involved in any of those conversations. And I didn't realize that for the longest time that you have to have the right person at the right level. So it's whatever group that you're targeting and then whatever level that can sign the check. The second thing I do is I charge a stupid amount of money. I mean, it has to be to the point where if you're talking to the right person, they're the only ones who have ever paid for anything at that level. If you're if you're not charging enough, you'll have somebody with the signing authority to bring you in who doesn't really understand the problem very well and who doesn't have to buy in. And that's going to lead to like you're gone in a month because the person paying for you is paying a lot, but they don't really care. And so you're giving them results, but they don't really care. And so that's another piece of it. So it's at the right level. You have to be charging at the point of the value that you're returning.

Speaker2: [01:47:43] So it not only makes sense, but it's also going to put you at the level that you need to be at and you're not going to have a lot of leads. That's another reason why you charge a lot. You provide a ton of value you don't want because you're one person. How many people [01:48:00] can you actually service? How many clients a year can you actually take on? And so that's another reason to charge significantly because again, like I said, it's to the point, it's to the pain. My clients come to me because they are hurting and that's when they're willing to do all of the things that are necessary in order to do this correctly and to make nine figures from data science and machine learning initiatives. Because if you're not going for that, bring someone else in. I mean, that's what Accenture and Deloitte and all those other companies are for. If you want to go smaller projects, smaller return on investments, do the incremental stuff, the prototypes, this isn't going to be a core capability. I mean, you hear what I'm saying. Like, I've gotten this target market and niche to the point where I know exactly who I need to talk to, who I'll refer people to almost immediately, how to deflect people that are going to waste my time. Because every hour I spend on a sales call that doesn't result in a sale, it's not or I don't get paid. And so it's really just eliminating. And you do that through every mechanism you have available to you. And that's that's the important piece, is when you're starting to talk about getting qualified leads. It's not about finding.

Speaker3: [01:49:22] Who.

Speaker2: [01:49:23] You know. It's not about finding a lead. It's about getting rid of all of the leads. That, number one, I'm not going to be able to provide enough value to in the first place. I say that to a lot of startups. It's like, I can't, you know, I would love to, but I'm going to rip you off because you're a startup. I can't give you enough ROI. I just I can't. So you start eliminating people and you use paid engagements. If you want to talk to me, I'm more than happy to give you a free half hour where we talk about what I do. But once your half hour's done, that next meeting's paid. Here's the link. And [01:50:00] you're going to find people who are very serious versus people who are just kind of casually exploring the problem.

Speaker3: [01:50:09] Super insightful because I for product at least that makes sense to me. Like you have your users but then you have your purchaser and that's a whole different sales cycle. And I think an example like the it was like in health tech like your users are physicians, but the purchaser is the CFO. He's a completely different person with backgrounds and stuff like that.

Speaker2: [01:50:28] But realize sales cycle too, when you're a small business individual, you can't wait six months, you know, and the vendor, the vendor registration process for some of these companies is it's like a month and a half, you know, and don't. So if you're if you're a single person working, you can't spend a month and a half getting registered as a vendor. You have to go with companies that have a fast track program, and most of them do now because they're used to working with individuals, one person consulting shops, five people consulting shops. And so they'll get you registered as a vendor in two or three days. And they're used to paying with net 15, net 30 terms, not net 90, you know. So there's all of these little stupid things that you have to take into account when you filter out who it is that you can take as a client. Because, I mean, everybody wants to take on the biggest clients on Earth, but like trying to get registered as a vendor for Walmart or for BMW, I could be like a three month process. You know, and like I said, they're streamlining it now. It's getting faster. It's getting better now. But these companies are not in a hurry unless you have somebody at the C-suite or at the EVP level going, know this person needs to be a vendor next week because they're starting in a month. So I don't care what your process is. Your process is different today, you know, and you can have that sort of fast [01:52:00] track happen. But if you're not talking to somebody at the right level, you can find yourself just. Yeah, and like Mexico says, cover charge. Yep, I charge a cover. Unless you're unless you're in the VIP line. No cover charge. Sorry.

Harpreet: [01:52:17] That's. It's like a master class, right? Thank you so much. This question kind of go back on me because I think I'm really sort of venturing into the type of stuff that you do with that, chunks of time and consulting, but not necessarily for machine learning solutions and stuff like that. My niche is now like, okay, I'm not even talking about breaking into data science anymore. Like, I'm just not interested in that. I'm not interested in talking about career advice for individual contributors that are under two years in the industry, not interested in that. So I'm kind of redefining my niche as mostly talking about developer relations with the primary audience, who are founders of a series of startup companies that are developing open source or open core products for machine learning tooling, specifically in ML ops or ML ops adjacent spaces. I feel like that's something that I could be amazing at that nobody's really doing that. I'm the person for that. What are your thoughts on defining a niche like that?

Speaker2: [01:53:26] That's an emerging creator role. And I just want to say to to Mexico real quick to hers. I figured this out. You have different rates. So if for my normal clients, it's net 15 or net 30. But those are people who have an established payor relationship with me. Those are people who have paid in the past. I'm good with giving them net 30 and they get one rate from me. Net 60 is 150% of net 30. Net 90 is 250% of net 30. [01:54:00] And it's amazing how fast they go. Yeah. No, no, no. We can actually do net 30. Yeah, we can do that. And even for a large projects, they can put stuff in escrow where you have access to it and the approval process is way more streamlined. So there's those tricks and putting your rate into the contract where here's the rate for this term, here's the rate for these terms, and if you're going to make me wait three months, you're going to be paying me over double, which I'm good with. I mean, that's, you know, that's usury, but it's a good interest rate.

Speaker2: [01:54:37] I'm okay with that. But talking more about your niche, you're in a creator and a community creator niche. And I know this is the artists of data science, not metaverse, but you're wandering into metaverse territory because the one thing that every investor who has been able to quantify how do you value a metaverse has always talked about community. That's every single metaverse. It's always about the community. It's always about the network effects. It's always I mean, and so every company right now that's wandering towards, yeah, exactly what Mark is talking about, everyone that's wandering towards that area for new for new means of monetization is beginning to look at companies like, you know, any company with Uber. Uber is an example. They've got a community and they're monetizing it in the absolute worst possible way. Like their business model is so busted. When you could look at a web3 concept as a centralized community based concept, you could do so much with that. And so when you talk about where your niche is, if you can talk about.

Harpreet: [01:55:55] That.

Speaker2: [01:55:56] And start saying you're not just [01:56:00] building for today, you're not just building a community, you're not just creating relationships. You are doing something bigger, longer.

Harpreet: [01:56:11] Term.

Speaker2: [01:56:13] And you're going to hit and this is anyone who's establishing their own category, anyone who's really going into a space that's blue ocean, you have to connect it to a need that companies understand. You have to connect it to something that they they realize is probably free cash flow in the future, but they have no idea how to get there. And if you're telling them so, this is the first.

Harpreet: [01:56:36] Floor.

Speaker2: [01:56:37] Of what you're trying to build. And I'm going to come in here and I'm going to build the first floor and what you're going to be able to build on that, it's up to you, but I'm going to get you in there and don't use my message. I'm just kind of making something up. But you get where I'm going. You see how I'm connecting it to a need that they understand. I'm talking about a technology that's cool and that's interesting and that everyone's, from a business standpoint, wanting to be attached to but doesn't really understand what step one is. And that's what I did with data science back in 2012 is I said, okay, so this is step.

Harpreet: [01:57:12] One.

Speaker2: [01:57:13] Because there was a three slide presentation. Slide one was big data, slide two was just empty with a question mark, slide three was profit. And so I came in and like my strategy was I am slide two. I'm going to teach you that. And you could do something very similar.

Harpreet: [01:57:32] Thank you so much, Mark. Go for it.

Speaker3: [01:57:35] Yeah. I was going to say I posted a link to a video in the chat. The 816 start of school for, what, three business models? That's the video that made me jump in and be like I Web3 is that metaverse entities like once I saw that video just clicked like the business opportunity with that. And so then, then I had a great strategic long term vision thing. I'm not going to go into that like you [01:58:00] guys not covered, but I'm also thinking like, what can you do today before you set up that long term vision? And something I'm noticed from people who from I kind of like microservices is a is a group that was not expecting at all our ML Ops and data solutions sales executives. Many times they keep on coming to me and in my in my LinkedIn chat and they're like, Hey, you're a data scientist. Y'all are weird. I have no idea how to sell to you. Can I talk to you? And I'd be like, Yeah, of course. Here's my link. And they're the quickest people to pay. They're like, Oh, oh, and just and just clicking it. And my hypothesis is, one they don't know. They don't know who to sell to. They know how to sell. They don't know like us really? Well, there's too much jargon.

Speaker3: [01:58:50] It's like data scientist was even mean. Like, I don't think any of us can define what data scientist. Because just so ambiguous. Right. But also. What the what we can provide to help them sell easily equates to large profits. So the value we bring is absolutely bonkers. If I can just spend an hour with you and help you sell ten times more, I provide you so much value. If anything, I'm charging way too little, which I realized already. Hence why I increased the rights shout out Ben for pushing me to increase prices. And so the reason why I bring this up is that you're in developer relations and I've talked to you before, like developer relations content side, but you're also in marketing, we're also doing sales. And so I feel personally like this is a market that I want to tap into. I just like having it focusing on having two entities. For better or for worse. But if I were to do a course and actually sit down and readjust my strategy, I would build a whole course targeted to salespeople trying to sell more OP solutions, and [02:00:00] I feel like that would be leveraged easily into these larger kind of initiatives.

Harpreet: [02:00:06] Excellent. Excellent tips as well as the inside as well. That great, great question from whoever was asked that question, because he kicked off one hell of a discussion. Did that did that right, guys? Let's go ahead and wrap it up. Thank you so much for hanging out another one of these two hour, half hours. I spent a lot of these guys. Thanks so much for being here. Be sure to tune in to the podcast if you have not already. And if you are listening and you made the part and you enjoying the conversation, don't forget that there is a link in the show notes to support the podcast, so please make sure you support the podcast. Shout out to everyone who has done so so far. Appreciate you guys. We're back next week. And in two weeks, two Fridays from now will be the two year party for the Art of Science. Because you've been around for two years, I don't know what the hell is going to happen, who's going to be here, but it will be a good event. It's a happy hour, but another one is normal things. So, of course, when we get to that, you guys take care of the evening, be in touch. Remember, we got one life on this planet. When I tell you something. Really?