2020-05-29-carlos-mercado.mp3

Carlos Mercado: [00:00:00] You can't do it. Start at the bottom with me. I was making no money.

Carlos Mercado: [00:00:04] I remember I was living in like a 300 square foot tiny place in Incheon, Korea. And I was like, I need to get into grad school. I need to go back to, like, my life in economics. And like three or four years, like, huge turnaround. Be patient like interest compounds. So it's OK if it takes like a year to get where you want to have, like here long visions, like people, they overestimate, what they can do in a week. But they underestimate what they can do in a year. It's very, very true.

Harpreet Sahota: [00:00:46] What's up, everyone? Welcome to another episode of the Artists and Data Science. Be sure to follow the show on Instagram, @theartistsofdata Science and on Twitter at @ArtistsofData. I'll be sharing awesome tips and wisdom on Data science, as well as clips from the show. Join the Free Open Mastermind Slack channel by going to bitly.com/artistsofdatascience. I'll keep you updated on biweekly open office hours. I'll be hosting for the community. I'm your host Harpreet Sahota. Let's ride this beat out into another awesome episode. And don't forget to subscribe, right, and review the show.

Harpreet Sahota: [00:01:36] Our guest today is a data scientist, economist and urban studies enthusiast. He's a renegade that love to you shiny to automate business processes, create interactive maps and leverage public data to model urban health phenomenon. He's earned a bachelor's degree in applied economics from the University of Central Florida, a master's in Applied Economics from John Hopkins University. Throughout his career, he's had a diverse range of experience, including time as a freight broker, a year long stint teaching English in Korea and working as a Data science freelancer. He's currently a senior data scientist at an up and coming global consulting firm working in the public sector on the Data science consulting side and specializes in the intersections of our shiny GISS machine learning, NLP and data visualization. In his current role, he's focused on product management and leads the technical sales team within the Artificial Intelligence and Intelligence Automation Group. So please help me. Welcoming our guest today, an innovator who is always striving to improve processes, improve efficiency to each project that he's a part of. Carlos Mercato, Carlos, thank you so much for taking time to get to be here today. Man, I really, really appreciate it.

Carlos Mercado: [00:02:42] Thanks. And thanks for adding in that recommendation from an old boss of mine into that intro. Very nice to me. I may have missed working for in some ways, trying to poach her. We'll find out if that works.

Harpreet Sahota: [00:02:52] Let's talk about how you first heard of Data science and what drew you to the field.

Carlos Mercado: [00:02:57] Yes, I graduated econ and I ended up in sales for a brokerage. I saw a little bit of sequel and mostly dislike maps like price modelling forecast for freight brokerage stuff. And I like this school. It's not really economics or whatever, I got in a teaching, I was a nonprofit before that, I was in retail management before that. So, I mean, I was working all through college. Ten years I've been doing something and like, I just started drifting further and further away from economics. And I just like woke up and I'm like, hey, I need to get back to what I was trying to do. Like, I was trying to do economics and math and, like, forecasting and stuff. And now I'm teaching, which is fine. But it's just it's not what I wanted.So I was Googling like, what are the cool jobs to do for people with economics degrees, and then I saw all the articles, number one job of the century data scientist, this is like 2016. And I was like, oh, like I thought I was late. Now I'm thinking that I was like, right on time. Yes, I saw the titles and I was like, okay, well what do they need to know? And I remember at the time I was looking at the rankings and I'm like, OK, R is above Python, but R is growing at a slower rate than Python. And they estimate that in one year Python will beat R. And I was like, OK, well R is one of the leading language now. And I worry about what happens in the future later. And then now, I mean I taught myself are there like the Johns Hopkins or Coursera specialization? And I went to grad school and did a lot of R there. My first job automated a bunch of stuff and R. So I got to Data science because data science was a good idea. And R, there's the language slant. And I came from a state of background. So R was super easy relative to like a real programming language.

Harpreet Sahota: [00:04:32] So it's pretty interesting just to draw parallels to my own journey. I actually also study economics as an undergraduate first undergraduate courses that I went to and went on to teach as well. And I was a math teacher at a public charter school and this was around 2008. And at that time, when I googled top careers, top career that came up was actuary. And so that was what led me into this field of data science. Eventually sorted all the way back then with the actuarial exam P. But, you know, taken into consideration now the journey you've had over the last few years breaking into Data science.

Harpreet Sahota: [00:05:12] Where do you see the field headed in the next two to five years?

Carlos Mercado: [00:05:16] Yeah, so I might be a contrarian man. I see, I don't think Data science is that new. I mean, when you say actuary, my guess what they used to be called before they were called something else, like we've been doing big data and fancy statistics since the 1950s. I mean like physics, geography has big data problems for decades. You know, I used to be called expert systems. So I mean, if you're looking to the history or the field, you see that it's not just the titles changing the works still the same, you know, converting Data and the money. So when I see the field the next two or five years, I don't see it that much different. I mean, you know, it's cloud virtual machines sure. That's just computers in Virginia instead of a computer in front of you. Like the parallels are there. So that, like, nothing much is going to change. Like, don't get distracted by the new titling and new ways of doing stuff. And I don't get distracted by the coolest new neural network games whatever. Like, the fundamentals are still good. So, I think the field's going to stretch out. I mean, I think we're going to see speech become more and more important. Then we're going to see speech replace a lot of the dash fording technologies that we're using. Natural language processing is going to get bigger, I think, computer. Going to get bigger. They're going to find new ways of using algorithms from one field into another. But I want to remind people I think the fundamentals are going to stay because they have not left.

Harpreet Sahota: [00:06:42] You mentioned that the field, in your view, is pretty much more or less going to be similar or the same as how it is now. So whether it's two to five years in the future or now, what are the qualities as separate good data scientists from a great data scientist?

Carlos Mercado: [00:06:57] I think the great ones will listen to the advice from other people that I am repeating advice. It's not my place. The fundamentals are the difference. The fundamental difference. So that's the fundamentals of product management. The fundamentals of communication, of negotiation, of, you know, understanding how machines work, how computers work, going back to the original languages. I mean, one of the best Data side, as I've seen on LinkedIn, is always tell him to learn like Haskell or something. And I'm like, how am I going to learn these like 50 year old languages? Well, you look at the fundamentals and they you know, it's like these are the languages you should study. If you want to learn functional programming, you should go to the basics and statistics. You should understand, you know, how general linear models work as opposed to studying how neural networks work. So the fundamentals are the differentiator. I don't know if it gives them book recommendations or if they'll see the video. But I got this book off LinkedIn Kevin Gray posted it Essential Mathematics for Political and Social Research. This is like the second Bible to me in terms of like statistics and stuff like that. The first one being, of course, the elements of statistical learning, the actual statistics Bible. But if you have a social science background and you're like, oh, man, I didn't do engineering, can I do Data science? I didn't do computer science or math can I do data science? Yes, you can. And like there are books for you that are written for you that will get you those fundamentals. And this is a great one. And I keep it literally next to me all the time as I am working.

Harpreet Sahota: [00:08:31] I'll definitely add that one to the show notes. Great recommendation. So you mentioned NLP, audio Data. What do you think is and be interesting about those next two to five years?

Carlos Mercado: [00:08:40] I mean, I think what people don't. I think what people naturally do is they try to, like, focus on one thing at a time. And the big picture is like how or how is our culture changing in the Data generation process? And what are the next problems? So when I said some of the fundamentals, it's because the next the next problems will generate Data the way that all Data is generated and stored now. So this audio Data is not going to be stored in MP3. You're going to be translating it into other formats. You'll be doing feature engineering to bring it back to the fundamentals? So the answer then I'll be thing. You know, we got to think about what's changing today. What's changing today is that people are getting really used to speed, to getting used to telling their phone what to do, telling their car what to do, telling their little, you know, robot assistant in their house what to do. And that robot assistant translating that audio into machine understandable, you know, keywords and stuff. Then it will do a command and then it'll get immediate feedback. So it's a constantly learning. The NLP audio and speech and the fact that people will be using speech to get things done. It's the next frontier. And like I said before, there's already company. I know Klick is doing this. I imagine Microsoft to be, I assume, and a thought spies is out there like NLP and speech will be how we dashboard in the next few years. So if you're looking to find a niche like NLP and audio is going to be a niche for Data science especially.

Harpreet Sahota: [00:10:03] It's right you said that. So I think a lot of data scientist, they do spend a lot of their time studying data structures, algorithms, coding, not enough time and economics. I don't think. 

Carlos Mercado: [00:10:12] No! Not enough time in economics, economics is so Cool.

Harpreet Sahota: [00:10:15] So give us a description of what economics is kind of from your point of view. And if you can make it accessible for data scientists out there.

Carlos Mercado: [00:10:23] So I think it's funny, if you know, they say like you can't explain it to your grandmother, you're over complicating. People tend to hear economics and their brain goes to like money, finance and like trades. And like that's not really economics. Economics is understanding, like how you structure decisions in the context of, like, business goals. So it's like some cost and opportunity cost and marginal versus average cost and fixed costs and variable cost. And these words sound money related, but they're not money related. There's just some value related. You can apply these concepts to time and you can apply them to funny things that automation like numbers of clicks and, you know, the business processes. I mean, I think of it it's just like an applied decision. Science, more so like a social science to me personally.

Harpreet Sahota: [00:11:12] So got big data. We've got the era of big data, era of machine learning. How does the application of economics differ now than it did before? And maybe, you know, adding to that. 

Carlos Mercado: [00:11:24] Big thing right now is I was just talking with this on the Internet about Frequentists versus Bayesian in like probability theory in how people think about these problems. I'll explain them to each other. I think the big benefits of big data for economics is that it will let you do better and easier like causal research. So in a context of like medicine, for example, and I don't have a whole thing about Hip-Hop, I know that just plenty of data scientist something to your anti hip. And you'll find their posts on that. But let's just let's just keep going. You know, imagine that you have, you know, millions of people taking a medication and you have all this data on these people. Well, now you're in a territory where you could start doing but causal research because you have the same individuals who have better health outcomes over time. You have medications over time. And it's no longer a cross-sectional, like non causal thing because most of our Data is cross-sectional. And what I mean by that is like, OK, well, I have this point in time and in this point time is this Data and as different individuals. But I'm going to run regression and I'm gonna look at the marginal stuff and I'm going to say, hey, this relationship significant. That's a highly causal research. Causal research requires like no different than difference, regression discontinuity with all that stuff requires that you're holding something's constant at the individual level. And I think if we think of big data and economics in terms of like longitudinal studies and natural experiments, I think that's the big improvement, is that longitudinal high information at the individual level allows for natural experiments to like really good causal research. So, yeah, I mean, I think medicine is the easiest example because you think of like, okay, well, we'll have these people who take this medication over time. And I'm not comparing them to other people who are older and across sexual and compare them to themselves. And I can sort of actually like making really solid arguments that these medicines are causing changes on our health. So I think to summarize, long term individual level longitudinal causal research is the big impact of economics and big data like that.

Harpreet Sahota: [00:13:31] And that's very, very insightful. Definitely give me a lot to think about. I'm sure listeners have a lot to think about as well.

Harpreet Sahota: [00:13:45] Are you an aspiring Data scientist struggling to break into the field within check-Out dsdj.co/artists to reserve your spot for a free informational webinar on how you can raking the field? This would be filled with amazing tips that are specifically designed to help you land your first job. Check it out dsdj.co/artists

Harpreet Sahota: [00:14:11] Let's talk about Bitcoin as it relates to Data science and machine learning how those two are going to play in the near future.

Carlos Mercado: [00:14:18] Let's not - let's backtrack from Bitcoin and talk about block chain, what is block chain and let us do block chain, public ledgers. This is like a self verifying system, right? So if you want to make a change on a data set, it has to be like approved by, like, the crowd almost. And this is a rough this is very rough explanation of what's happening. But like you could imagine, like, OK, you know, again, about medical context, like, oh, hey, I'm going to submit this person's record to this system. And you're saying like, hey, this record is this at this time? Well, if you're trying to submit this block to the ledger and then all these other people are checking their system. So it's like, oh, hey, your block cannot be added to the single source of truth, the ledger, because it's not passing these checks that are in my block chain. And I'm like mixing concepts here. But so, like, I want to submit person A is an updated record. Hey, you know, you can't do that because I don't even have the record a person coming to see you. So there's no way that I can allow you to sign the record to person A, given that I haven't fully mapped that person has been to see you. Which doesn't fully map to the fact that the person who hasn't taken the medication even know updated their prescription for this medication in the last 30 days. So your record can't be valid and it's just as natural self-correcting mechanisms to like have a single source of truth. That's like super, super validated. And I'd be mixing a lot of concepts to simplify there. But I think it's super cool. I mean, I think it's super cool. The applications in terms of like self validating systems to some extent and having a single source of truth and inversion controlling data, we don't have a person controlling data enough to really know what version controlling our code. That's, I think, another thing. So let me add that to the two to five years early. Controlling your data will be way bigger in the next two to five years.

Harpreet Sahota: [00:16:08] Yeah, definitely, man. That's one thing I don't think gets discussed enough. I don't even see a lot of posts on LinkedIn about that at all. And I think that's something a very important piece of conversation is version controlling Data, especially when you have got models in production. Right. And you're naturally going to have Data drift occur. So if you don't have versioned history of your Data, you can't.

Carlos Mercado: [00:16:33] Yeah. I mean, if you think in terms of model trips and like Black Swan events, I mean, there were models in production back in 2005 and 2007 that were feeling great. They were like, we're recovering from 2001, recovering from 9/11. These models are all bul Bul bul, we're going to make tons of money in the next five years. And then Lehman Brothers falls and the Black Swan event suddenly. Can you use those models anymore as your model going to correct for this massive shift in your Data? What does that actually mean for your production systems? And I'm not sure that we're using the lessons from Data science in that time period for our expectations for Data science after COVID. I don't want to talk about that too much, but I think we're not really using history enough to inform our expectations for the next two to five years as you recover from that. I'm not sure what we should do about that. Just something I'm noting.

Harpreet Sahota: [00:17:24] So changing gears a little bit, I know that you've got significant expertise and experience using GISS. So first, can you define that for us? What is it? How does it relate to Data science and maybe within the context of urban economics and how it's being applied to urban economics,

Carlos Mercado: [00:17:42] Such as geographic information systems, if everything related to the creation collection, storage, retrieval, analysis and presentation of data that is fundamentally spatial. That's a rough quote from my class that I learned in grad school because I was like, oh yeah, GISS, it's like machine learning. It's like no. GISS says the entire day it's every single step is included in GISS. And I was like, okay. Very cool. Extremely bright. So as it relates to urban economics, there are fundamental spatial relationships for a lot of reasons that I think of something called Tubulars First Law. It's like things near each other are more similar than things far away. And it's like, oh, holy crap, that's so obvious. Like it's just that it's obvious statement. But it was just such a big deal because so much of our data we just throw away the spatial part. We're just like, well, you know, it's collected from where it's collected. But these are the features that I want and these are the features I engineered. And it doesn't matter what you run on. Like, you start looking at residuals and it's residuals are correlated and you're like, well, how do these correlate residuals, no big deal. And it's because you're omitting stuff and really, really often gives you emitting spatial like Data emitting spatial information. So in economics, you know, you can talk about a bunch of stuff. You talk about health disparities and how you don't understand differences and health outcomes across minority groups based on where they live. But you might not be incorporating the fact that where they live doesn't just impact their access to hospitals. It also affects their access to food and the quality of their food. So GISS just becomes a mechanism for assigning food swamp ratings, food desert ratings. You know, health disparity ratings to geographic area that let you start doing interventions. I mean, that's like the social determinants of health and urban economics and, you know, health equity. There's a bunch of, like buzzwords the intersect on that. But yet, GISS is a big deal. And there's a lot of things to learn. I think if you if you have no clue what to study in GISS, I definitely recommend you at least Google this. MAUP, the Mappable aerial unit problem, I think is a common sense for the idea that aggregating up from individuals to geographic units is often arbitrary and it's extremely dangerous. But it's also mandatory because you have to do it because you have individual level data and talk about us an area. So I'm getting a little into the weeds, but if you look at that phrase, it will change how you look at maps for the rest of your life.

Harpreet Sahota: [00:20:09] Very interesting. So in the sense that it's dangerous, is that because people can move from one geographic location to the other? So.

Carlos Mercado: [00:20:16] No, it's the aggregation, so I guess my point is, OK. So think about some income, right? Like, oh, you know, we have people in York City and San Francisco. They make tons of money. So you start looking at, you know, income maps across the country and you're like, OK, well, I'm going to start doing, you know, pay equity stuff or maybe I want to start investing and, you know, a new employment programs. I'm not going to do it in New York City. Insurance that's good to. They make tons of money, but that's a culture at the city level. Go down to a level. Go to the county level. Oh, this is actually the county is too big that, go to the census tract level. OK, nice. Are you saying that there's differences in income across census tracts? So now you're a more granular unit and you have more information since strikes still have problems because you're still mixing people up and census tracts. You go to the census, Booger. You go to the census block. There's always a problem when you're aggregating individual stuff up to an area level. And when we don't think that we forget that, it leads us to, you know, not taking factors into account. We don't take it into account. You know, gentrification or we don't take into account like food swamps in one area. One thing that we found in a paper that I'm hoping to make external in the next few months is, you know, we look at looking at the importance of using census block groups whenever possible instead of census tracks, because census tracks are not everyone does their analysis. It's very granular, 70000 of them, census block groups are much smaller, there's two hundred twenty thousand of them. And even within a single census tracks and like very small town Alabama, this is a random example. There is a railroad splitting the census tract and the census block group on the north side of the railroad is like significantly wealthier, significantly less minorities, significantly less isolated. If you define that as like living alone, a household size of one in the census block group on the south side of the railroad tracks, literally the wrong side of the tracks. But the census tract level blurs that. So that's the mythical area union problem in terms of like we are anything more aggregating up, you're losing information and if you don't respect that, you will leave people behind.

Harpreet Sahota: [00:22:26] Very, very important insight, thanks for sharing that. Do you have any other resources or articles that are kind of covering that topic that our readers can go check out if they want to learn more?

Carlos Mercado: [00:22:37] So I did this a lot. If I want to learn something, I just started Wikipedia and just started reading the Wikipedia page. And any people feel like the mother of this, the father of that read that personal Wikipedia page. So for urban planning, gosh, where do I have a Jane Jacobs? You know, the mother living, planning the death and life of great American cities I have it on paperback. I haven't finished the book, it's a thousand pages. But just like read the history of the field and read the history of the individuals who are making huge impacts on that field. So that's my first step. I don't go straight to a blog about the field like learning history, because without the history, you'll be knocked out of context. And then you're gonna end up on an open office asking people what articles should I read? Only shouldn't read anything. Don't read any articles. Are the history first. Well, you know what interests you.

Harpreet Sahota: [00:23:24] Yeah, definitely learned the vocabulary of the field that you're going into. So then you can develop mental models for these terms and issues that people in that space who are working with. There's been through some of your LinkedIn posts here. And I saw a post about the lessons you learned from chasing a job with AirBnB, let's talk about that from your perspective. What do you think you did wrong and how did you approach the process or how would you approach the process if you were to go through it again?

Carlos Mercado: [00:23:52] Yes, sure. So I might carry out this really quick. I was not in consulting at the time. Consulting is very cool. So I'm not sure what to do. Definitely check out consulting firms, they're very interesting work. But to answer your question; So what happened was I was using the apps and I was like, oh, that's cool, you know you like go live in someone's house for a few days and, you know, you meet strangers and they turn in like light friends. And that's cool. So they're looking at the company, looking at their data science stuff. And I was like, wow, a lot of data science postings and the Data science job advertisement, the job descriptions are like really easy to read and they feel very possible. So if you're also if you're writing job descriptions. Look at what other major companies are doing, I really applaud the way that they write their job descriptions. So are the descriptions. And I felt like, oh, this is very doable and it feels very possible. And the more I read, the more I was interested in their new kinds of problems, because they weren't just invested in just finding people to host homes, they're also interested in being a full experience applications. I read this blog post about their launch experiences and things like that. I mean, I was like, wow, there's a lot of stuff that's going on here. There's a lot of Data is a lot like urban design Data here, but economics here. So I just thought it was a vehicle company I work for. So what I did was I added everyone I could find. I worked there. I would read other blog posts, I read other media releases, press releases. I would think about their problems. I would look at the app and think about like the user experience of the app, here's the thing about my experiences in the app all this stuff I built about this knowledge about them. And then I applied and I got rejected. I didn't even get rejection. I just never got anything because I wasn't qualified and I wasn't doing it right. I was adding people, but I wasn't talking to them or engaging with them, I was reading stuff, but I wasn't including that in my cover letter. I wasn't like, you know, showing my relevance as I was targeting them, I just had this magic idea that, well, if I know it, they'll know that I know it, and your resume is a sales document. So if you don't include it in your cell, they're not going to know to buy. So the big problem and I think the thing I would do now is I would spend a few months finding the people who are most active on LinkedIn, who work in that company, engaging with their stuff, messaging them, having a casual relationship with them, learning about their day to day, verifying that it's actually a place that I'd fit in. I didn't know that I wasn't checking the culture or anything like that. Yeah. And that building those relationships and then lightly poking for referrals, I just saying, hey, you know, I saw this recent post, job posting. I know we've been talking on LinkedIn for a few months. You look at my resume and tell me you think I'd be a good fit for the firm. Like lately, ask for some feedback and stuff. And you do that to 5 to 10 people who work there. One of them's going to say, actually, no, I think you be a good fit. Let me put you in touch with someone. And that's how you find a director or an H.R. rep and you start getting the fast, you know, the fast track to the job, because if you're on the fast track, you're probably not on any track, it feels mean. But like, you can't be one of 500 applicants. And it's just it's impossible. It's just so.

Harpreet Sahota: [00:26:58] And also there's great tips and great advice. So I want to dig into another post you did on LinkedIn about what you've learned about Data scientists working for a psychiatrist at her nonprofit school.

Carlos Mercado: [00:27:11] Oh, yeah. So that was that was crazy. So I was I was at a nonprofit and it was like a selling thing. I was a management intern and well, you know, part of our thing was to communicate and like sell these projects that we do to convince companies to invest in these employee experiences that also get people fed that are food insecure. And I called Calder and talked to her and she was like, of course, I sounded a great experience for my students. And I was like, wild. So then I met her, she was really nice, I learned a lot, she did a little project. And then years later, I was at a freight brokerage, I was thinking of moving back to Orlando. Okay, let me just send her a blind email asking her if she'd be if she is interested in having, like, an economics grad teach economics or something. And she's like, yeah, of course, I would love to have, like, talk to you more about this. And we talked and I join her school as like, you know, a teacher. I was doing all kinds of stuff like math teacher, history teacher, PE teacher. It was a very small school of 30 kids across like sixth grades. So it's very fun. And the things I learned were, you know, the reason that people went to her was partly because, you know, her school was really cool or those because like her as a personality, like she was highly invested in her students. As someone with expertise in child psychiatry. And I was there in those meetings with her, you know, where appropriate on all of them with, you know, the parents and the therapist and psychotherapist and her being a child psychiatrist and just listening to how you talk situations that are really difficult, really, really difficult situations like your kid who's having problems that are not problems that regular kids have about their own personal self-esteem, their psychology, their confidence, their interpersonal relationships, you know, obedience includes really serious psychological problems. And just the empathy learn and like the leadership you learn. And like the little things like re-framing and paradox, which I've posted about, I definitely will look into just like how do you turn a really serious, difficult, painful situation, into something that can be like a fun challenge that you believe you can conquer. And I was like, I'll never forget, just like all those hours and those around with her, just like learning that stuff and like reality. And they have been so important because the most important part of data science, besides knowing math, is being able to communicate to business people and making sure that they understand, like you can trust me with this problem. And I will get you recommendations that will make you look like an all star and also make the company an all star. Think it was a long answer. And also, I'm on the front end of Data science of the the back and some very people first centric. I can understand how some data scientist don't eat any of that advice. 

Harpreet Sahota: [00:30:03] I think all data scientists need that people first advice. I think there's something that is definitely lacking from a lot of data scientists. But I wanted to ask you, you mentioned there's something about re-framing how do I take a negative and view it as a opportunity for do something positive. 

Carlos Mercado: [00:30:20] Let me summarize, reframing paradox, if you like.

Harpreet Sahota: [00:30:23] Yeah. Is a reframing paradox. And let's try to put that in the lens of let's say there's somebody right now who's applied for job after job after job, COVID 19 is happening and they're not getting any responses back. How could they use that re-framing paradox in this situation?

Carlos Mercado: [00:30:39] Sure. So let's do for reframe first. Reframe is the idea that the information is just a reminder to yourself that the information you have is essentially negligible in volume. And what I mean by that is like all the information you're getting, like some of this job and applies replies, I have my resume. I don't like train skills. It's impossible. Like no one's getting hard right now. Those are like that's data that you're collecting as your individual lived experience. But you can rephrase, you can put that individual experience in the context of every one. So don't just think about it from the person looking for jobs. Think about it from the company standpoint. Think about in terms like, okay, like we have a lot of uncertainty. Google, I just I just saw an article is canceling 2000 temp and contract employees like as of a few days ago, something. So it's like these companies are facing mass amounts of uncertainty. Also, these H.R. managers, they're getting inundated with applications because these people just need a job, any job, and they're desperate. So I'm not saying that this sort of magically work for anything like thinking of this will magically get you a job, but it will make sure that you're in context of the situation in a way that other people might not be. And you can include that in your cover letter. You can include that in a conversation that you're having, that you understand that data science right now doesn't feel like something that you should be investing in. If you're a company that's very early in their maturity curve. So the reframe there. So I get off track really easily. The reframe is remember all the perspectives for the people that you're dealing with, the company, the hiring manager, the recruiter or the other applicants. Because if you don't do that, you're going to do the same thing over and over and over again and wonder why our results don't change. Paradox is the idea that sometimes your action, the opposite of your action, will be better than your action. So people tend to have this bias that, OK, well, I want to make I want to do an action. So here are my possible actions. And they always do this for some reason. Their possible actions are all really similar. And that doesn't actually make sense. Like, that's really like one action with five different flavors. Like, if you have five possible actions and they're all we're really, really similar, that's bad. Paradoxes idea is like what would the opposite of these actions be? Is it possible that the opposite of this action is better than this action? And I think on my LinkedIn example, what I said was so that I think an example is, okay, hey, we want to automate this pipeline so that this new forecast gets updated to our sales team. Okay, so let's talk about what that would look like. And then the data scientists lead does OK? Hey, you know, our level effort on this is like 400 hours, which is equivalent to some 100, you know, some tens of thousands of dollars on staff time. You look at the equity, the situation, and there's only like $50000 an equity rally for the sales team there. So what's the idea, right, that your mindset is OK well, can I make us a weaker version? Can I make a passive version? Can I make a you know, a version that maybe drags out a little longer and only uses the leftover time of my staff instead of making the primary project? Those are all really similar actions in that you're all you're still trying to build them what they asked for. The paradox is, if it is not valuable, tell them and maybe do something else that's completely opposite. Maybe instead of automating some pipeline, you can just, you know, replace the entire pipeline or remove it. And maybe you're inundated with KPIs and they need less KPIs. Actually, they don't need a 25th KP AI. What's the paradox here? It's like instead of adding new features, can you remove. So that's that's the idea with paradoxes like what's the exact opposite? The possible the exact opposite as more equity.

Harpreet Sahota: [00:34:36] Does that kind of tie in to working as a consultant as well? I feel like as a consultant you have to wear very different hats all the time, right? Many hats all the time. Can you kind of walk us through the fundamentals of consulting for Data science?

Carlos Mercado: [00:34:48] Sure. I don't think it's anything special about Data Science Consulting, Relative to other Consulting. I think it's just one flavor of consulting. I mean, I think even our Data science team, I think we think we're consultants first, which skews us because, you know, we're not doing this internal big data science of like, you know, here's our like here's our revenues and here's our costs. And let's, like, optimize this thing where we're not internal retail data scientists. We're consultant, data scientist. So what that means is we're constantly dealing with new clients and constantly dealing with new problems. So we have to, like, really refine our problems statement generation process. And something that I'm doing is setting a sales leader of our group is, you know, reading more about product management and reading more about like job to be done theory. It's I mean, you can Google there's a great book. It's free online. When coffee and kale compete, it's on base camp. I recommend that book. And one of the things that we're learning through our mock demos and our book sessions, we literally have like a four hour job, which is like really cool. So it's really been helpful when these are learning like your customers don't know what they want. They only know that they need to make progress. So if someone tells you, hey, you know, we need you to build this for us. Don't take that at face value. You to take music, step back and say, OK, well, you clearly come to this conclusion because you've considered other things and other things that positive and negative, both the negatives outweighed them. What other solutions have you considered? And tell me why you didn't choose those. Because it's possible. I know something that you don't know that would lead us to picking a different solution. That's one reason. Another reason might be the classic example in the book is, you know, there's a drone manufacturer and they're saying, you know, these people don't want drills, they want holes so that they can hang their picture frames. And it's like actually take another step back. They don't want holes. They want to hang a picture. So it's not just the hole. It's actually not the drill. Their progress is having more in their home. So when you can actually get the progress that they want and you don't get distracted by these tasks, that they may be ignorant leaving that not not really naively think we'll solve that problem. You can actually re-frame paradox. Give them another solution that fits their moldings tiredly. And one example, this is actually published, we got published in the American Journal of Epidemiology, along with one of our clients for this. I mean, they had this idea that they're going to have this excess script and they're going to training webinars around how to, you know, learn SAS and use SAS in order to implement this algorithm they developed. And the paradox for us was well, you don't really want that. You don't want to do training sessions. You don't want to teach people proprietary software. You want to give them an algorithm that will improve public health. What's the paradox here is let's not that's not train anything on all the time and use this software at all. Let's get your algorithm into open source instead of learning to change the code. Let's talk about an application that has defined parameters so that people have a Web based, no user friendly, user centric, designed interface that will allow them to perform this algorithm without code. And that was a huge re-frame for them because it really took out that really put them in the context of, you know, these people are tiny health department, they probably can't afford proprietary software. So we're like losing people before they're even joining. That's my job theory. Super helpful. I'm thinking about what solutions they've tried and didn't consider making sure that you're not taking their tasks at face value. Those are all super important.

Harpreet Sahota: [00:38:31] Very, very fascinating. I specially like that point about the problem statement generation. Having to look at it from a different angle pretty much.

Harpreet Sahota: [00:38:40] What's up, artists? Be sure to join the free open slack community on bitly.com/artistsofdatascience. It's a great environment for us to talk all things Data science, to learn together, to grow together. And I'll also keep you updated on the open biweekly office hours that I'll be hosting for our community. Check out the show on Instagram @Theartisticdatascience. Follow us on Twitter at @ArtistsOfData. Look forward to seeing you all there.

Harpreet Sahota: [00:39:09] How does this differ from working in a regular organization?

Carlos Mercado: [00:39:12] I barely know. The only time that I worked on internal data science was for startups, and those always didn't really go very well and they're always really small. So, I mean, I read about it. There's a really cool book called The Unicorn Project. It's realistic fiction. It's about this engineer, she's a senior engineer. And she gets put on like a product. No one wants to be on, and she found that all this terrible stuff in terms of like I.T. and dev environments and like no inability to work at a fraction of all the stuff. And she's she's grinding out how to, like, turn this project around 180 and make it like the star of the company that brings in tons of profit. And through that book, I learned a lot about like what is an internal Data science group look like, internal I.T. function look like. So she's like delegating engineering tasks, delegating Data science tasks, delegating analyst tasks. And like through that, I was like learning like, OK, like these people are building systems of applications. And it's very different when I do. I don't build systems applications. I build micro applications like you do, like one specific task extremely fast and well, so that problem is solved and you can move on to another problem. And I think for internal data scientists, they work on a giant problem and a giant problem has a bunch of moving pieces. That's a very different ecosystem, to my understanding. I recommend the book.

Harpreet Sahota: [00:40:34] Excellent book, excellent book. I highly recommend that. I don't recommend Phoenix projects as much. But you could project this great.

Carlos Mercado: [00:40:42] So the difference is we recommend Unicorn Project for developers and then Phoenix Project for Managers. Yes that's we do it.

Harpreet Sahota: [00:40:48] Yeah, I found Phoenix Project to be two DevOps-y for me. But Unicorn project cause I just I guess I understood it more.

Carlos Mercado: [00:40:57] I mean I struggled through, Unicorn project I was like, what are some of these words like. I've never been on the back end developer. I don't know what these are.

Harpreet Sahota: [00:41:04] So do you have any tips for people who are trying to break into Data science by doing freelance work?

Carlos Mercado: [00:41:09] I think for freelance is hard because it's not going to pay the bills. It's just going to make you meet a lot of friends and hear a lot of cool problems. Our recommendation for breaking in is like, don't be afraid to start at the bottom. Like, I started at a firm as the lowest paid, like lowest ranked associate. My job was to copy paste out of word into notepad and remove formatting. And yes, I timed it. It was faster to do it this way because at the last program and then paste it into like Adobe Captivator or something, which I hated if you tried to right click paste without formatting. I did like eight hours a day. Seriously, I was just like copy paste in a notepad, formatting, copy paste, that's terrible. So what I did was I was okay if I could automate this. Would that be OK? And they're like, I don't care what you do, just get it done. So I downloaded R kind of our shining. I didn't know what I was doing. I just have to be away because I have to copy paste another eight hours a day. Like I'm going to quit this job and I'll lose my apartment. So I learned about, like, you know, library officer and how to like you can read in documents and I can break up into paragraphs and look at their formatting and their styling. And I was like, OK, let's re-engineer this whole process. What would it look like to have a perfect word document that I could convert into a PowerPoint, that I could have that PowerPoint already stylized and prepped and templated, so they can import it into the software for e-learning courses. Going to trial and error and I figure out how to do it. And suddenly we were pumping out like iterations, like we went from one review to like five review phases. We were going like a million miles an hour in terms of like the content or spacing, we did 23 online courses that are published now on the Internet containing medical education credits in like less than six months, including like I was using I was using like a Google text to speech audio, like robot audio sounds, super humanoid, all because I started at the bottom and they had low expectations. So it was really easy to, like, impress. And then I asked to move to the data analytics scene because clearly such a skill set of mind, like within months, it's like 10 months within a year. They are like telling me that I should move teams. And I was like, dude, like I would have never gone into that job without, like, already being inside here. I ended up switching firms, but still, like starting at the bottom. And there's nothing wrong with that. Nothing wrong with the Data analyst title. There's nothing wrong with any of these titles like Just get In.

Harpreet Sahota: [00:43:32] That's some straight hustle and grind. And the best advice that anybody can hear who's trying to break into Data science is to ignore the job title or just take the job. Even if it's Data adjacent, do the work, crush it, and then learn from who else is in your organization and then just make that jump, that's nice.

Carlos Mercado: [00:43:51] Low expectations are great. Like, I love low expectations. It just makes it so easy. It's like a little sad that they will be racist, but like it will get so impressive to me, just the fact that I speak English. My name is Carlos and I'm just like, this is such a low bar. Like this is like an incredibly low bar to have for me. But I'm crushing this bar because English is my native language. I don't mind low expectations. And I think if you put yourself in those spots and you're okay with those and then you just completely crush like you said, it's will be awesome. And I know there are perils electric. You know, an education there and stuff like that. But for adults, like, there's nothing wrong with it.

Harpreet Sahota: [00:44:31] Yeah. Low expectations are good if you know that you can crush them. But if you use them as a excuse to just keep your productivity, your output and your just mentality level back. That's definitely an issue.

Carlos Mercado: [00:44:44] Like I said, I mean, it was a 90 something percent of the people who asked me for a resume interview while reading my article, and I gave it to them. I don't know. I know they don't read it because I look at my view, counselling don't go out. So, yeah, so I'm like People are lazy. I mean, if you're lazy, like do something else because there's too many people trying to get that title like. It is hard to stay in Data science sometimes just because there are so many people gunning for your spot.

Harpreet Sahota: [00:45:15] Yeah, I mean, these resume. What are some some of the biggest areas for improvement that Data scientists can make on the resume?

Carlos Mercado: [00:45:24] Sure. Easy ones. It's a sales document. So use it to sell. I don't want it's not a menu. It's not restaurant menu, why are you having 50 skills? And like, this is like it's a constant. It's like, yeah, here are my like eight programming languages. Here's five to ten packages pro programming language. Here's six different database architecture stuff. Here's three different cloud, whatever. And I'm just like this is a restaurant menus, that build your own data. Scientists right now, like, tell me what you're really good at. And make it really easy for someone to say this guy would fit on my team or this girl on my team, that's step one. Step two is Polish like, why is your linked in name impossible to type like Bobby Smith and three eight nine four three, come on. Just Google how to clean that up. Why is your GitHub repo homework to IPY and B your clearers, you know how to open a notebook. Is that an expectations you have for them that they're gonna open up your notebook? Because I open up your notebook. I know what I'm going to see when I see a bunch of Data cleaning mixed in with Data of is mixing what modelling. And it's gonna be ugly. I'll be like, you don't know how to split your code in a script. You don't know how to write a function. I would I hired like you're just not making it easy for yourself. Like your GitHub should feel like a blog post. It's usually about those who shouldn't get a pages. And disposal people shouldn't get hub free pose both to get to pages that feels like a Web site. Like it's just these things have so much equity. People are just so short term biased that they don't want to put in 10 hours to get a job that pays 20 grand more. And it's like, are you serious? You won't put in 10 hours, 20 grand in like year one equity. They're just not thinking of it. Right. And then also emphasizing the wrong thing. I mean, honestly, I am impressed that people who travel across for an education. I'm extremely impressed that you just uproot your whole life, move to another country, go to graduate school in your second and third language. Very impressed with that stuff. That's not gonna get you a job like you have to put your experience first and you have to under emphasized your education because you need to look like a pro life. If your education is up top, you're telling me that the signal is higher, because I went to a good school and the signal needs to be hiring me because I can make an impact. I think last one that so many. I mean, how much more time do you have? I can give you three more. Maybe give me two more. If time permits.

Harpreet Sahota: [00:47:47] Well, I think an important point you made there is about a restaurant menu of your resume. You should have one simple rule for your resum. If I put this on my resume, and people ask me questions about this thing from left, right, center, upside down, inside out, and I cannot answer questions on this thing. Maybe I shouldn't bullshit myself and clevers in interviewing me by having it on there, because at that point, say, what fuck you doing.

Carlos Mercado: [00:48:12] I did it too. My first resume were like, oh yeah, I touch Google Text to speech GCP. I'm like, no, I don't know Google cloud. One of my about only take that, I use one API that's on Google Cloud. So I mean, that part of it is just honesty with yourself and part of it is making it simple. Two more pieces of advice are make a summary statement like be specific. Like I want to use Python in my next data science shop, that's nonspecific. Like pick an industry they have experience and domain expertise is really good. I'm, you know, a better one, for example might be. It's an article I posted to better. One would be, you know, I'm a Python specialist with a strong background in finance, especially interested in integrating like natural language processing and machine learning to forecast financial markets. And that's like really, really specific name drops, like two key skills. It has a cool twist than NLP because NLP is under use in finance. That's changing now, that's been changing actually. People have been using automated like web crawlers to understand like stock events for a while. I should say I'm so summary. And the last one is just like narrative based. Like if I see another resume that says I increased the accuracy of our database by 15% using imputation. And just like, OK, one, I don't even know if you know that imputations dangerous based on this story. And two, I don't know if that accuracy actually makes a business impact because he didn't connect it to money or percentages or time saved or anything else. So, like, I have to ignore that bullet point because it's bullet point isn't connected to generating value. And I can't assume that's slightly accurate more accurate or imputation actually leads to value.

Harpreet Sahota: [00:49:57] And the other thing Git-hubs, it's like, do you see cookie cutter Data science like Git hubs, repository structure and organize your code accordingly?

Carlos Mercado: [00:50:05] So much should get over. We are just like blank IPYNB. And the reason he says, I did an analysis on Data inside my Python notebook included. I'm just, you're really going to make someone do. They've gone like three clicks. They went from your resumé to the Git hub, they went from you Git hub up to a specific repo and now you're telling them that you're not going to tell them the nice experiment Data science design and link result of your analysis in a blog format on their read mate. You're gonna make them open your code. Like, come on, no one wants to stare at code. They want to like, see the story. And the code can be like the fourth thing they check.

Harpreet Sahota: [00:50:43] Yeah, well, you have to make it easy for people to see the value that you contribute. And if your first impression is I mean, people just click on 18 different links before they get to the punch line. That's a good impression. Right? Make it easy for people to see the value you can contribute by just having a very well thought out read me, which is kind of like an executive summary of sorts.

Carlos Mercado: [00:51:01] Yeah, I think we would just forget that there's humans on the other end of this. They're just like they're pitching to a robot. So they're like dumping keywords and they're dumping texts and they're dumping like numbers. Even if you give a chance. Right. Let's say that your resumé has like 80$ HST pass rate. That's all relevant if it has a zero percent human pass rate. Much rather only get through like 2% of the ATS and then perhaps the humans by like 80% plus. And also, I'd actually try to score on a test ATS by doing the fast track, as mentioned before.

Harpreet Sahota: [00:51:33] So let's talk about this and tie this in to the importance of building your personal brand as a data scientist. How can go about building your personal brand for yourself?

Carlos Mercado: [00:51:40] Personal branding is important for two reasons. The first one is these platforms will multiply what you give. So Twitter is a great social media source for data scientists. A lot, of course, designers are on there. The Medium is great for posting articles. I'm not sure what the community is like. I haven't used it that way. Reddit community is really nice, LinkedIn is really cool. The reason that you build a personal brand is not for other people, it's for yourself. And that's because what you give gets multiplied. Like, I've only been, like, actually trying on LinkedIn for a few months. And this is my second podcast. Like, tons of people are following me. I'm just like ranting. I'm not really structuring. 99% of my LinkedIn is just like on the phone while walking around, like feeding my cats and stuff. I'm like all I'm saying right about this random idea while I'm like sitting outside enjoying the weather for a beer, drinking coffee, I don't drink coffee, drinking water. And it works. I mean, like it. I get so much more than I put out. And you can't do, so that's the first reason. Second reason is it just puts you on people's radar. And that's good for job security. And it's good for just like your own growth. Like, I have met so many people on LinkedIn that I then go to a conference and I know that they're there. And it's like, oh, we're not actually friends. This is really cool. And I guess that, like you, whatever you give, you get like so much more back.

Harpreet Sahota: [00:53:05] What are the qualities that you are looking for when you're hiring a data scientist for your team?

Carlos Mercado: [00:53:12] If I were hiring data scientist, I would keep it very simple depending on the level. Can you write reproducible code? Do you understand like object classes? Do you have, like, fundamentals of statistical programming? Can you debug that like commenting in an outline? Do you know that notebooks are not everything that you're allowed to not use Notebooks? Do you understand how the cell of data scientists and the consulting side so they know our consultants first? Do you understand, like a lot of this stuff is philosophical, that these are not like easily as a cure problem and go No Network. That's a really bad sign. And that's your first instinct. So these are like the kind of flies and basics. I posted more on that post in terms of like kind of six things that I hope you have three of. And I don't even have all six things that well, there's just a six things that I think are important.

Harpreet Sahota: [00:54:06] This last question here before we jump into the lightning round. What's the one thing you want people to learn from your story?

Carlos Mercado: [00:54:11] You can't do it? Start at the bottom like me, I was making no money. I remember I was living in like a 300 square foot tiny place in Inchon, Korea. And I was like, I need to get into grad school. I need to go back to like my life in economics. And like three or four years, like, huge turnaround. So that's the main thing I want people to get. And also be patient like, I don't know, interest compounds are compounds. So it's OK if it takes like a year to get where you want. Just have a tap like long visions like people, what they overestimate, what they can do in a week. But they underestimate what they can do in a year. It's very very true.

Harpreet Sahota: [00:54:49] So jump into lightning round here. What's your Data science superpower?

Carlos Mercado: [00:54:53] R Shiny.

Harpreet Sahota: [00:54:53] What's an academic topic outside of data science that you think every data scientist should spend some time researching on?

Carlos Mercado: [00:55:00] Cell psychology.

Harpreet Sahota: [00:55:00] So what's the number one book, fiction or nonfiction or if you were to drop both. that's OK. That you recommend our audience read your most impactful takeaway for me.

Carlos Mercado: [00:55:10] Yes, I think swipe to unlock when I say it's an easy read. It's getting more popular. And I think it gives you gives you a really nice overview of what the average person might not know about technology. And you're going to talk to a lot of people who don't like study technology that much. So I think that looks like really cool. It really gives you like a nice level set of what a miscellaneous, like business person might not know that much about.

Harpreet Sahota: [00:55:35] So if we could somehow get a magical telephone that allows you to contact 20 year old Carlos, what would you tell him? First, give us some some context 20 year old Carlos. So where was he? What was he doing? Would you what would you tell him?

Carlos Mercado: [00:55:47] 20 year old Carlos was entering his senior year undergrad. I would tell him to take school a little more seriously, but also to start studying R more earlier and start reading on the side, more like I loved economics and I just hated economics classes and I felt bad. And the session I felt that I should it's been like, okay, they're not teaching me what I want to know because these classes that I would love aren't available. So just don't I don't feel so bad. And I read economics on the side make you feel better about not liking economics classes.

Harpreet Sahota: [00:56:24] You can have a billboard put anywhere in the world. Where would you put it? What would it say and why?

Carlos Mercado: [00:56:31] In Jacksonville, Florida, at the beach, I was like, enjoy the weather. I don't know if I'd just say, like, enjoy life a little bit more like appreciate the little things. Like I really miss Jacksonville Beach and like the day that I was just like my whole life. And every summer I was just again, I go to the beach, I play beach volleyball and I swim. And I do like 10 hours  every day. And now looking back, I was like I took that for granted. That was so fun and easy. So, yeah, it's appreciate little things. I put it back in that sentimental spot.

Harpreet Sahota: [00:57:02] There you go, man. What's the best advice you ever received?

Carlos Mercado: [00:57:05] Someone once told me what I was talking about, about something like, oh, that's not really relevant to me. And I was like, that's rude. That's such a rude thing to say. But also, it saves so much time. I get to save so much time for them to just acknowledge that what I was telling them was not relevant to them and that we should not continue spending time on it. And I was like that really woke me up to the sunk cost fallacy being so real because I'll get like 20 minutes into a meeting or a conversation and I'll realize that this is so off track. We're so far away from our goal right now. But we're a 20 minutes into this like this, you know. Right turn Segway. Let's just see where it goes. Now, just say, actually, we need to scrap all of this. Like we need to start this all the way and get back to where we need to be. Like the sunk cost fallacy is so dangerous. And that woke me up to that. And I've been using that phrase more and it feels good, but it gets the job done.

Harpreet Sahota: [00:57:58] What motivates you?

[00:58:00] Money, money is the unit of exchange. So, like, I actually feel this way, like money will solve so many problems. And if money gets problems that it won't solve, it will solve so many other problems that you get to focus on the problems money can't solve. Money will just be able to substitute so much headache. You can do it all your life like important stuff. So like anyone who tells you that money is an important, it's like trying to take your money.

Harpreet Sahota: [00:58:29] What song is giving you life? What song do you have on repeat?

Carlos Mercado: [00:58:33] Have you seen that new man that goes around that you go on your Spotify and you see you actually type like on repeat and tells you what you've been listening to, like crazy.

Harpreet Sahota: [00:58:42] I had no idea that was a thing on Spotify. 

[00:58:45] Yeah Spotify is fancy. So it's Sugar by Brock Hampton. I would also say that 20 minutes by a little woozy for I don't know what he's saying. That song that I'm like not really paying attention, but the beat just seemed like I'm so ready to party on that song. Also, Roses by Saint John.

Harpreet Sahota: [00:59:01] So, Carlos, how do people connect with you? Where can they find you? 

Carlos Mercado: [00:59:04] On LinkedIn. Very easy to add on LinkedIn.

Harpreet Sahota: [00:59:08] Right on, man. Carlos, thank you so much for taking time out of schedule to be on the show today. I really, really appreciate having you on. Thanks.