Latest / Agentic Conversations (formally mlops.community) / A Conversation with Seattle Data Guy // Benjamin Rogojan // MLOps Coffee Sessions #21

Transcript

A Conversation with Seattle Data Guy // Benjamin Rogojan // MLOps Coffee Sessions #21 · Agentic Conversations (formally mlops.community)

• 0:16
  Started back. This is gonna be a great chat.
• 0:19
  I'm excited to have this. Do it.
• 0:23
  Yeah. Domain Hamin date.
• 0:25
  A guy with us in the house. He is as we were just saying A
• 0:30
  prolific writer. He's got all kinds of good stuff
• 0:34
  on medium so if you are not following him on medium check it
• 0:37
  out right now or once we finished with this podcast and
• 0:44
  you getting out right now, everyone, stop what you're
• 0:46
  doing. Go right now if you are driving
• 0:49
  and you got to remember now and so yeah, Benjamin it is a
• 0:55
  pleasure to have you here with us.
• 0:56
  I'm really excited to talk to you about some of the learnings
• 0:59
  that you Found around ml Ops and, and data science, data
• 1:03
  engineering. I mean, you are a man of many
• 1:06
  hats. I could say, yeah, thank you
• 1:10
  both. Thank you.
• 1:11
  Demetrius and David for inviting me.
• 1:13
  Technically, David. You said my name?
• 1:15
  How it's how my parents had me, technically, incorrect Benny.
• 1:19
  I mean it's from Romania so it's like yeah it's actually right.
• 1:25
  Nice. Yeah, thank you guys.
• 1:29
  Well how about we start? By just telling us a little bit
• 1:31
  about yourself. What are you working at?
• 1:33
  What are you doing right now at the moment?
• 1:34
  Just tell us anything you want. What's up with you?
• 1:37
  Sure, sure. So, I mean, like great right
• 1:39
  now. The mom and again, like doing
• 1:41
  consulting plenty of writing and helping one company develops,
• 1:46
  some more of a like, categorization model for
• 1:48
  themselves. They do a lot of legal
• 1:52
  documentation work, where if they provide other companies
• 1:56
  with kind of condensed bits of information, about what's going
• 1:59
  on, is Far as, you know, laws and regulations.
• 2:03
  And so that you know, it can be very hard to kind of track all
• 2:06
  that information and what's going on.
• 2:08
  So they've got to help condense it down and clean it up and now
• 2:11
  they're just trying to improve kind of how to categorize it.
• 2:13
  So that way they can, you know, when person a and person B,
• 2:16
  you're interested in some topic, they can better, you know,
• 2:18
  provide them that information rather than them having to even
• 2:21
  sit through their site. Another company I'm working with
• 2:24
  is looking to kind of improve kind of how people rent buses.
• 2:30
  And I guess a lot of that is either done online through some
• 2:33
  Brokers or a little more like through kind of like the actual
• 2:39
  owners website. So they're trying to create
• 2:41
  something that's a little more, a little more friendly for
• 2:43
  everyone and and has everyone win and part of that also
• 2:46
  involves kind of like the pricing model.
• 2:49
  So like you know something that's very so much like Lyft or
• 2:51
  Uber. So those are some of the current
• 2:54
  big projects that I'm working on right now.
• 2:57
  Yeah. And then plenty of writing, I
• 2:59
  think I have this current goal for writing, where I'm doing
• 3:04
  from raw to machine learning. This kind of like the topic were
• 3:06
  from raw data to machine learning.
• 3:07
  So basically, I've just been trying to write from like etls
• 3:12
  data, warehouses elt. He's all the way to, you know,
• 3:15
  deploying a model, you know, either ml office or on a Lambda
• 3:18
  function or, you know, from scratch, like just like made me
• 3:21
  wait like, you know, and talk through the different kind of
• 3:23
  processes. And and so like and really going
• 3:25
  through each of those kind of steps in between to stab.
• 3:28
  Someone can think about, you know, that whole Data lifecycle
• 3:30
  data pipeline. Yeah, that's one thing that I
• 3:33
  definitely noticed in your writing is how you looked at the
• 3:38
  whole spectrum and it makes sense.
• 3:41
  Now, that you're saying that from raw to production and so
• 3:46
  what are some of these learnings key things that jump out at you?
• 3:49
  I know you've written a lot on many different topics and you've
• 3:55
  dove into the specifics. But what are some things that
• 3:59
  maybe BR are top of mind for you as you are doing this project.
• 4:05
  Sure. I think, I think one of the
• 4:06
  interesting things that I always remember is like back the first
• 4:09
  time I ever like try to do something even relatively close
• 4:13
  to add a science project. You know, you get you get to
• 4:16
  this point where you're like okay I've written this thing in
• 4:18
  our and now what like it does the thing that I wanted to do
• 4:23
  but but now what and you know I was like whatever 22 at the time
• 4:28
  so and like working full time but No one really had a that
• 4:31
  next step, you know, set up. And I, that's when I kind of
• 4:34
  realized like, oh, that's like there's a gap here where it's,
• 4:37
  like data scientists. Think about, like, oh well do
• 4:40
  this research and will, like analyze this thing but like,
• 4:42
  okay, but then, like how does it go from here to actually
• 4:45
  touching people and working with, you know, and interacting
• 4:48
  systems? And, and I realized that there's
• 4:50
  like, you know, there's there's that kind of Gap especially
• 4:53
  like, from like a, like, if you're fresh out of school, you
• 4:55
  don't really talk about that. You kind of learn how to make
• 4:57
  the thing work, but you don't learn How to play it.
• 5:01
  And there's some similarities with software engineering but I
• 5:03
  think software engineering has been around so long that like
• 5:06
  it's very, most companies have the process is clear here,
• 5:09
  right? Like great like it's like oh
• 5:11
  whether it's devops or what or if it's more Mania or more
• 5:14
  manual? It's very like out wide.
• 5:16
  It's like here's our deployment process that one through Thirty
• 5:20
  I didn't see that necessarily as much with like these nicest
• 5:22
  words like oh here's a deployment process.
• 5:24
  That was also like five years ago.
• 5:25
  I think you're we're definitely seeing the proliferation of
• 5:29
  tools and Expert people kind of think about that nowadays.
• 5:32
  So, that's the yeah. And it's funny you mention that
• 5:34
  though because there's something that's interesting that I
• 5:36
  thought you wrote about and it was like the importance of
• 5:39
  having a mentor. Hmm.
• 5:41
  And that makes total sense when you're looking at it that way.
• 5:44
  And like you said with software engineering, you can kind of
• 5:49
  it's already there. It's very much mature but here
• 5:53
  it's not so clear. And so when you're a data
• 5:55
  scientist and you're trying to go to that, next step is very
• 6:00
  nice to have someone at your job who has already done it and can
• 6:04
  show you. Okay.
• 6:05
  Here's how we do it. Here's and whether or not it is
• 6:08
  doing it manually or automating it just like being able to show
• 6:13
  you the trail that you need to walk on is super helpful.
• 6:17
  Yeah, no I totally, I definitely agree.
• 6:20
  I think in so many ways like whether your software engineer
• 6:22
  and or data scientist or you know whatever kind of technical
• 6:26
  role you have. I I was not fortunate enough.
• 6:30
  My first company to have a mentor, like my mentor,
• 6:32
  literally, it was like, he was there and then two weeks later,
• 6:36
  you like, peace out, I'm leaving.
• 6:38
  So it was like and then like they didn't hire new tech person
• 6:40
  and it was just like, oh yeah, you're the new guy.
• 6:42
  So. So it was like just learning all
• 6:45
  this stuff and I made a lot of mistakes.
• 6:48
  I made a lot of terrible code and I just remember going to the
• 6:52
  next company and like looking at their infrastructure looking at
• 6:55
  their code and being like okay so this is how I should have
• 6:57
  been doing it and Like realizing how valuable it would have been
• 7:01
  to have someone that like, you know, like kind of is there and
• 7:04
  like hits you in the head occasionally, like, you know,
• 7:06
  whether it's tough, love, or gentle whichever way the person
• 7:09
  likes to do it. Like, just some of this, like,
• 7:11
  this is a bad idea. Your code, written poorly, like
• 7:14
  calls you out, because you don't know all the time.
• 7:16
  I mean, some people do some people, you know, kind of
• 7:18
  school. I'm sure in our great coders,
• 7:20
  but I think most of us, don't like, we just were like well we
• 7:23
  made it work, so that's good. Learn by making mistakes, right
• 7:26
  by failing. A bunch of times and making big
• 7:29
  boo boos Production, you know, deleting stuff by accident.
• 7:33
  But yeah, that's how you learn, right?
• 7:34
  As they as easy. And I think it's a great point
• 7:36
  that you brought up like, yeah, like a lot of the stuff that you
• 7:39
  learned the school is theoretical.
• 7:40
  It's kind of abstracted away, right?
• 7:42
  But really, when you're working in a real-life system of
• 7:44
  production system, it's very complicated, right?
• 7:45
  There's a lot of moving Parts. There's a lot of particulars to
• 7:48
  The Domain and then it's just every day just like challenges
• 7:52
  that come up you know. And so like no one obviously can
• 7:54
  prepare you for that. It's just one of those things
• 7:55
  you have to just learn by getting in you know getting your
• 7:58
  hands dirty actually getting in the code, right.
• 8:00
  Ting, it fixing things, breaking things.
• 8:01
  Like, that's just how you learn this stuff.
• 8:03
  I mean, I wish it was like, yours, like a book that like
• 8:05
  taught you everything. This is what you need to know,
• 8:08
  to Be an Effective engineer, but there's really not much like
• 8:11
  that. I find it all spread out through
• 8:12
  a lot of different even disciplines, right?
• 8:14
  Like, there's a lot of influence not even from just engineering
• 8:16
  that helps me. But even you know, other random,
• 8:19
  you know, sort of disciplines. But yeah, yeah.
• 8:23
  And that's all along that point. I think it's interesting to
• 8:26
  think about this article that you wrote About like agile and
• 8:32
  data science, and what the benefits and what potentially
• 8:37
  the cons are for for like trying to work agile and in, in being a
• 8:46
  data scientist, can we jump into that?
• 8:48
  Because I think there's so many like nuggets of wisdom there.
• 8:53
  And first of all, like, maybe can you give us a little
• 8:55
  background on that blog post and how it came up, why you wanted
• 8:58
  to write it? Yeah.
• 9:00
  I mean, I think for me, it was, it was definitely kind of
• 9:03
  covering a concept that I'd heard a few times like, people
• 9:05
  talking about it and, you know, I think in the tech field we
• 9:10
  have this tendency to amalgamate everything like it's like oh
• 9:13
  this is All Tech. So it should all work the same
• 9:16
  right? Like because agile Works in
• 9:18
  coding at you working days, science the same way, you know.
• 9:22
  I think there is this tendency to just kind of like ml Ops even
• 9:25
  right. Like I agree with ML Ops concept
• 9:28
  but like you know we do have a tendency Just be like well we
• 9:30
  have devops so this is now Mo up and sometimes I think we do that
• 9:34
  to you too quickly where it's exactly as it really makes sense
• 9:38
  just to go back with Emma laughs.
• 9:40
  I do kind of see it as it making sense just because me I believe
• 9:43
  the option part is like once we have a discipline that's kind of
• 9:46
  maturing and we start like having more concrete processes
• 9:50
  when you start automating that process.
• 9:51
  That's where I kind of think, that's where it kind of fits
• 9:53
  that up thing. But as far as agile, yeah, no, I
• 9:56
  mean, that's kind of brought about was like, you know, think
• 9:59
  about like Oh, it seems like a lot of people are talking about
• 10:02
  this. Well, okay, it does.
• 10:04
  It fit sort of? Yeah, sure.
• 10:06
  They're, there are aspects of it that you can bring in to data
• 10:09
  science. But I think, you know, just
• 10:11
  sticking it on top of your, you know, data science workflow,
• 10:16
  doesn't necessarily always work, right?
• 10:18
  You know data science is very like, we're like or software is
• 10:22
  very results driven where it's like, very easy to be like, hey
• 10:24
  build this piece of code, you know, and it's do X, HV X Plus x
• 10:29
  equals. Y, that should be there.
• 10:30
  But right in, put me out punch, very, very clear.
• 10:32
  And I think sometimes a science isn't that way, right?
• 10:34
  Like when you look at like more, like, kind of true design for,
• 10:37
  like, it's more research based and like, kind of long-term
• 10:40
  project where there's like, likelihood of failure, it's
• 10:43
  like, you might not even bring like think about how much
• 10:45
  science doesn't bring anything on the other side, right?
• 10:47
  Looks like, you know? Yeah and that's so funny.
• 10:52
  You say that because somebody in the, in the slack Community was
• 10:56
  talking about that like they were asking how can Can I talk
• 11:01
  to my boss who's like angry at me because everything that I've
• 11:05
  been doing hasn't, like, there's been nothing that's come from it
• 11:09
  yet and he's like, how can I tell him that this is just
• 11:13
  research and you shouldn't be expecting everything that I do
• 11:18
  to come out with some, you know, groundbreaking model?
• 11:22
  Yeah, no. I think like you kind of have
• 11:25
  to, you know, maybe depending on what kind of research doing
• 11:29
  hopefully. I think your eventual hope is
• 11:31
  that you come up with that, you know, 10 million dollar savings,
• 11:35
  right? Like that.
• 11:35
  That's your Saving. Grace is like, I think, I think,
• 11:38
  if I recall even like Google or Facebook, like remember what it
• 11:42
  was, I think, I think I'm reading an article about Google
• 11:45
  or try to remember where they're basically like, oh, you're
• 11:48
  expected to do something, or create some sort of like, decent
• 11:51
  amount of either algorithm work or something or come up with
• 11:54
  some sort of research that saves the company x amount of money.
• 11:56
  Like every, like, one to two years or something, where it's
• 11:59
  like, then it Time. I'm trying to remember the
• 12:03
  breeding methods, like it made a lot of sense is it?
• 12:04
  Yeah. Interesting.
• 12:05
  Yeah. Like yeah, like yeah, you might
• 12:08
  not always come up with something and I think it's kind
• 12:11
  of silly to push for it. Yeah.
• 12:13
  But over the long term I was just gonna say, I think it's a
• 12:17
  really good point, because, yeah, there's a lot of failure
• 12:20
  failed experiments are not necessarily failures from a
• 12:23
  business perspective, right? You've learned something about
• 12:25
  the value of this data set, or maybe it's not valuable, you've
• 12:29
  eliminated. Mated some possibilities, you
• 12:31
  maybe try to frame it in a certain way and it's just not
• 12:33
  really helpful like sometimes you could build a model, right?
• 12:36
  For some problem, but maybe clustering or something, really
• 12:38
  simple, or some sort of heuristic is a little bit, you
• 12:41
  know, more. It's just as performant and it
• 12:43
  may be saves you a lot of trouble making you know, the
• 12:45
  super complicated thing work. So yeah, I think I've dealt with
• 12:48
  that myself, you know where you it's not a research is so it's
• 12:53
  like more open-ended than an engineering deliverable, right?
• 12:56
  And it's very difficult to say when something is like done.
• 13:00
  It's ready, you know. And it's like yeah, you could
• 13:02
  always like make it a little bit better or do another analysis to
• 13:06
  realize something that, you know, maybe you missed or
• 13:09
  there's also I feel like it's very difficult to know when
• 13:12
  something is done versus Like A system that you built.
• 13:14
  It's up and running it's available.
• 13:15
  I can make requests to. It's like that's like okay it
• 13:18
  works versus the ml model when it down production.
• 13:20
  You like sometimes you have to wait, right?
• 13:22
  You may have to wait a little while just to know if it
• 13:24
  actually is making predictions that make sense and that yeah, I
• 13:27
  think that's like you said, those are the differences that
• 13:29
  are Clean-cut and maybe an agile process doesn't really account
• 13:33
  for as well. But it doesn't mean it's
• 13:36
  pointless. I think that I've heard some
• 13:37
  people say it's like that, it doesn't work.
• 13:39
  I don't know if I agree with that because I do think it does
• 13:42
  have some value, just to organize work.
• 13:43
  It's like a useful way to learn how to work.
• 13:47
  Ya know. I think I think it's the way I
• 13:49
  kind of use it. It's like you can still break
• 13:51
  down work. Like I think you call that one
• 13:53
  point where it's like just just analyzing the data to see if
• 13:55
  it's worthwhile like that can be a bit of work.
• 13:58
  That's like yeah like that could take like a week.
• 14:00
  I can see that taking and like you should take a week like, I
• 14:03
  think, like somebody like, oh, well, do it in like a day.
• 14:05
  It's like, should you take the day to like, look at that.
• 14:07
  Data is like could take some time, like, think about it
• 14:10
  because, you know, taking that week now.
• 14:12
  Maybe say, you know, saves you six weeks of model development
• 14:16
  that you realize is trash later on, right?
• 14:18
  Like your point again. Younger version of me would
• 14:22
  like, just someone give you data.
• 14:23
  It's like, okay, this data must be good and like you just start
• 14:26
  building a model and like, later on to be like, this doesn't this
• 14:28
  data is garbage like He spent four weeks like building
• 14:31
  something that like you know, as based on stuff that's either
• 14:34
  manually input. So it's not great or or
• 14:37
  something like that. So you know, spending that time
• 14:40
  on that first part I think is valuable.
• 14:42
  So yeah, I think if there's this volume kind of breaking it up
• 14:45
  like breaking up work, because again, it just helps you
• 14:48
  approach things and know when things are done.
• 14:51
  I like to say like when boxes are opened and closed out, you
• 14:53
  know, when you all your be nice, if you open some boxes and then
• 14:55
  close them and put them away and you're done and yeah, it it also
• 14:59
  helps. Like providing communication,
• 15:01
  right? Like oh, where are you on this
• 15:02
  project? Like, which part is delaying it
• 15:04
  and I think it just lets you kind of say like, oh it's you
• 15:07
  know, the I'm stuck on, you know, exploratory data analysis.
• 15:10
  I don't think it's where you be stuff but like, you know, some,
• 15:13
  you know, some clear step in the process so they can someone can
• 15:16
  kind of get like, okay, I'm extends now I know.
• 15:18
  Yeah. And that's something that you
• 15:21
  also wrote about that. I thought was super important
• 15:24
  and I wanted to bring up was the idea of defining your data
• 15:28
  science process. And the need for that and how
• 15:33
  different it is like you were talking about, between like,
• 15:37
  software, engineering and data science, because it is, it's not
• 15:40
  so clear cut. You can't just say, like, okay,
• 15:43
  I'm gonna do this and then expect for that to be done in
• 15:47
  two weeks, because maybe you have, you do this, but then you
• 15:50
  have to go back around and go to go back to the drawing board,
• 15:53
  right? So maybe we can dive into that a
• 15:57
  little bit more, and hear your thoughts on on, Idea of defining
• 16:01
  the data science process and what you've seen that works when
• 16:05
  you have to define it. Yeah no I think David kind of
• 16:09
  brought this up earlier where it's like you know, you can your
• 16:11
  kind of never done are you Lee with research right?
• 16:13
  Like that's one of the things I think with like analysis or
• 16:16
  research is like you can always go down another Rabbit Hole.
• 16:18
  So like as far as like defining a day's time to process that's
• 16:21
  kind of like one of the reasons I like wrote about that concept.
• 16:24
  It's like if you don't have something that's kind of at
• 16:27
  least again whether it's like 100% clear cloud or just
• 16:30
  guidelines. I'm more of like a guideline
• 16:32
  kind of persons. Like, you don't have to like
• 16:33
  follow, exactly. But just have a general process.
• 16:37
  It just lets, you know, when things are done, you know, it's
• 16:39
  like okay, we've looked at this data, we've kind of done some
• 16:42
  general analysis. We've, you know, we've gained
• 16:44
  some interesting insights. It might not be anything that's
• 16:46
  a worthwhile, but it might be interesting and worth writing
• 16:48
  about. So we wrote like a quick one
• 16:50
  pager like a quick conclusion, you know, save your query
• 16:53
  Xavier, whatever kind of code and and kind of bummed it all up
• 16:57
  and now you kind of have a deliverable so at least you have
• 16:59
  something, you can feel like it's done.
• 17:02
  And so, you know, and you've gone through that whole process
• 17:04
  and again like it lets you communicate very easily.
• 17:07
  Ali to your manager like hey, where are you in this process?
• 17:09
  And and you can say, yeah, we're again, we're looking at the
• 17:12
  data, we're kind of had a weird kind of currently kind of
• 17:14
  writing up some graphs and charts.
• 17:15
  I'm writing a report and we'll present it or something.
• 17:18
  We'll talk through it just to something again to kind of like
• 17:21
  make sure your manager and anyone who else is like
• 17:24
  interested in the project has some sort of something at the
• 17:26
  end that that's tangible thing. Again with data scientist we
• 17:29
  don't always have something that's very tangible and I think
• 17:31
  there's this value I think in just having something that's
• 17:34
  like okay this is the work that I did both for your Own
• 17:37
  betterment and also your like mental health of the why it's
• 17:40
  like what I do for the last two weeks.
• 17:42
  I was just like, yeah, that's an r and or and that counts as
• 17:46
  work. And I think that that's one of
• 17:48
  the challenges, like, for doing research is that, like,
• 17:50
  especially when you think about your career, your, you know,
• 17:53
  talking about what you've done, what you've accomplished at
• 17:54
  your, at your role in your current role.
• 17:56
  It's like, yeah, I've done a lot of data science, but what the
• 17:58
  hell does that even mean? Like right?
• 18:00
  Well, did a bunch of experiments had to write some code to build
• 18:03
  this, you know, to see if it works.
• 18:05
  But yeah, like, you know, really trying to think Think through
• 18:07
  what exactly you did is not. Always so clear, cut, one thing
• 18:10
  that I found, I saw that I thought was really cool.
• 18:12
  We haven't used it. Ourselves is, I don't know if
• 18:14
  you heard of it, the Microsoft data science process.
• 18:17
  So it's like this, basically the guidelines around writing up the
• 18:21
  things that you did. So let's say you had to use this
• 18:24
  piece of infrastructure, reach out.
• 18:26
  You get get data from this data set.
• 18:29
  You would just document, like, literally that process like I
• 18:31
  did this, I use this data said it had this many columns is
• 18:34
  features in. And when I, when I shared it
• 18:36
  with some Some of the data scientist, it seemed a little
• 18:38
  bit like, you know, verbose like you have to do all this writing,
• 18:41
  but actually think it's really useful for your own, like, I
• 18:44
  guess like understanding of what you actually did and to be able
• 18:47
  to share that. And communicate the value, what
• 18:49
  you did, even if it was like you said something that was a
• 18:52
  failure. Like the model actually isn't
• 18:54
  performant, but we know hey, a baseline is actually better
• 18:56
  than, you know, a fancy model that too at that.
• 18:59
  Sometimes I think can be communicated writing about it
• 19:02
  right versus just, you know, assuming that, you know, if you
• 19:05
  didn't productionize it, It's not valuable which is, you know
• 19:08
  I feel a lot of the times of what people feel like if it's
• 19:10
  not on production. If it's on a software product
• 19:12
  it's not really doing anything but there's a lot of value in
• 19:15
  the Messes in the in the failures that I think that
• 19:18
  writing about it or finding some way to reproduce the results in
• 19:22
  some way could be really helpful.
• 19:24
  Yeah. And you know, that's another
• 19:27
  thing, another benefit of that like, writing it down and having
• 19:30
  it. I know, I can imagine that
• 19:32
  people gave you push back on it because they're probably like to
• 19:36
  do so much extra work. Like, I don't have enough work
• 19:38
  already, but it gives you a little bit more job security.
• 19:43
  We could say, because how many times have I heard people say,
• 19:48
  oh, the higher-ups have no idea what the data scientists are
• 19:52
  doing, and they just think That they're paying him, a bunch of
• 19:55
  money to sit around and twiddle their thumbs.
• 19:58
  And so this like will help them understand what is happening,
• 20:03
  even if they're not getting models into production.
• 20:07
  Yeah, and I think that's important and I think it's about
• 20:09
  again it's about creating that that layer of communication.
• 20:11
  I mean that's that's why I ended up picking up, picking up the
• 20:14
  kind of idea of like having a process was like I was working
• 20:17
  with a team and like my like the manager was just like, look,
• 20:21
  this is what I want. Like, at the very least, Just
• 20:24
  like in your Excel just like put the question that you're trying
• 20:27
  to answer, put your conclusion and then like, put your charts
• 20:30
  below that you like you're supporting it.
• 20:31
  And they she basically went to everyone's like, I'm like, I
• 20:34
  just need this just to like and then we'll save it and we'll
• 20:37
  have it in the future. They just like, yeah, it makes
• 20:39
  us that like yeah, we yeah, duh, again something you don't think
• 20:42
  about in school because like no one tells you how to do like
• 20:45
  data analyst process. Like I remember, I would just
• 20:47
  like I first started, like I'd write queries and then I'll be
• 20:49
  like, I do something else is gonna be someone would ask for
• 20:52
  something done again and If you like cute.
• 20:54
  Where's that query? Wait, where do I put like I have
• 20:57
  that query somewhere. Totally.
• 20:59
  Yeah, like and on what yeah, yeah, yeah, exactly or
• 21:03
  something. Where it's like someone asks to
• 21:04
  my question is like you could have just given them them that
• 21:06
  report or it's just those things where it's like having that
• 21:10
  ability to always look back. There's a lot of value there and
• 21:16
  I think a lot of people just like, they're like, oh, I just
• 21:19
  like analyze this data and write up a PowerPoint and then be done
• 21:22
  with it and lose all that information that I did and all
• 21:26
  the work that I did. No see.
• 21:27
  I bet you probably have to deal with this a lot, right?
• 21:29
  When you're working with clients, like invoice, like,
• 21:31
  writing up an invoice. Maybe right like, hey, I like I
• 21:34
  did, like, you have to account for like the analysis where they
• 21:37
  did and maybe there's not like a finished notebook, but maybe you
• 21:40
  spent all day, writing a bunch of code that actually doesn't
• 21:42
  get production eyes, but it's work, right?
• 21:45
  How do you navigate those challenges?
• 21:46
  Like how do you what? Are there any tools in
• 21:49
  particular? Any processes that help you
• 21:51
  quantify the value of your work to your clients?
• 21:54
  Sure, I mean II think for me one it's usually their end result
• 22:00
  like for example with the payment optimization like,
• 22:05
  right? Like we're like oh look you
• 22:06
  know, based on last year versus this year you know you can able
• 22:10
  to increase your sales or I like average sales cost by like
• 22:14
  whatever nine years. 10% is like yeah, there's proof right there.
• 22:19
  I think that's a big one in general.
• 22:21
  When you're, if you put something in production.
• 22:22
  I guess you also asked if it's not in production I think.
• 22:25
  Yeah, there's like yeah, I'll generally end up always putting
• 22:29
  together some sort of like research or, you know, just the
• 22:32
  one pager to Patriot. Just says, hey, here's what did
• 22:34
  here's my conclusions from this in like, you know, the research
• 22:37
  that we did hear some things, you might want to look up later,
• 22:40
  like, you know, this was just kind of an initial analysis if
• 22:42
  you want to, like, look up these points later and I think that's
• 22:44
  like one point. The process is like, you'll
• 22:46
  always have any questions, write those down, kind of be like,
• 22:49
  okay, you're going to forget those questions, you know, later
• 22:51
  on the future. So, write down those questions
• 22:53
  based off. You know, you've got it's like,
• 22:54
  hey, based off of, you know what, we figured out here, you
• 22:57
  could look in more and also we find Value, you know, in these
• 22:59
  directions and write those down. So then if if that client either
• 23:03
  wants to do more in the future or whatever, they've got it and
• 23:08
  they know where to look or they know where to send someone else
• 23:10
  to look whether it's you or you know, a different client.
• 23:15
  Or consultant. I think it's always the things
• 23:17
  like you should always make sure if you're hit by a bus right
• 23:21
  like that someone can pick up your work.
• 23:23
  I think that's the big thing that I always try.
• 23:25
  Regardless if I'm Consulting you're working from for
• 23:28
  full-time, kind of gig. It's like you know some should
• 23:30
  always feel pick up your work and okay this is good, you know.
• 23:33
  Nice nice. So I forgot to mention to you
• 23:37
  all then been coming. I know you.
• 23:40
  You don't know this but we tend to wear Shirts on this show, I
• 23:47
  just got a new one from my wife ice for Christmas early
• 23:52
  Christmas. So I wanted to show it off real
• 23:54
  fast at checkout for a pineapple is with some glasses.
• 23:59
  It's like a teal and it's just great swag.
• 24:06
  Thanks for that, thanks for that.
• 24:07
  Anyways, back to her say so when I was going to ask but Benjamin
• 24:11
  so what one other thing that I want to just kind of talk about
• 24:14
  more. So you mentioned, Like being
• 24:16
  able to show the Epp that the impact of your work by saying,
• 24:18
  okay, there was an increase in X or decrease in y or whatever.
• 24:23
  So, what tools and there's some one thing we talk to me, step
• 24:26
  back a second. So we use my, we talk about
• 24:27
  monitoring a lot, right? Be able to observe the
• 24:30
  performance of a model and production, you know, thinking
• 24:33
  about concept rif data Drift. But even something even, like,
• 24:37
  simpler than that. What about, you know, just
• 24:39
  showing that or how do you relate it to, like the business
• 24:42
  metric? So like, let's talk a little bit
• 24:44
  about how you can. Show the value of your work, the
• 24:49
  impact of your work using monitoring and observing like
• 24:52
  absurd observation tools, Arps observability.
• 24:55
  There you go, observability tools, sorry about that.
• 24:57
  Yeah, and I think there's like, you know, there's definitely a
• 25:00
  lot of, there's a lot of fancy things you can do.
• 25:03
  There's a lot of not as fancy things you can do.
• 25:05
  And especially when I work with like smaller companies, which I
• 25:07
  think has been a majority of my clients, you know, just even the
• 25:10
  simple Tableau dashboard where it's like, you know, you're
• 25:12
  tracking the right metrics. I think it's about picking the
• 25:14
  metrics you want to. That are you think are related.
• 25:17
  So again like average sale price like okay that makes sense.
• 25:20
  Like you know if you're if you're doing a price
• 25:22
  optimization tool, you want to see that average sale price go
• 25:25
  up probably. So so that's or maybe it's like
• 25:29
  you're trying to turn around maybe like what am I saying?
• 25:33
  The amount that you actually close a deal or something, like,
• 25:35
  pray for something, where if you're doing, like a lot of
• 25:38
  sales. So you want to pick a metric
• 25:39
  that I think makes sense. And I think you see this even
• 25:41
  like with large companies, right?
• 25:42
  Like you know, Facebook Twitter, whatever social media company,
• 25:45
  you'll see like engagement right or YouTube right like engagement
• 25:48
  is their big metric. I think that a lot of these
• 25:50
  companies focus on and as they're doing, you know,
• 25:54
  whatever some small change in the model, or some small change
• 25:56
  website, they make that change and then they see if engagement
• 25:59
  is impacted. And so really, I think, A lot of
• 26:03
  it can be as simple as as like, make sure you pick metrics that,
• 26:06
  you know, can be impacted by these things and then track it
• 26:10
  again. You don't necessarily need a
• 26:11
  crazy observability tool. I know there are a lot of great
• 26:14
  tools. I know.
• 26:14
  Like, I think, I think I talked to you guys or at least
• 26:17
  Demetrius about this, like, why Labs that a I think is doing a
• 26:20
  lot of stuff. As far as you know,
• 26:23
  observability goes. I also realized how hard that
• 26:25
  word is to say, when I talked to Andy day and I was like, oh my
• 26:28
  gosh, I can't say observability. I've never had to say this like
• 26:31
  out loud. Make sure you tell her no.
• 26:34
  It's like it's I'm still speaking and it's I haven't
• 26:37
  finished the word so so I get that.
• 26:40
  So there are a lot of tools and I think especially like to talk
• 26:43
  about day drift and and just logging and track the right
• 26:47
  metrics. I think they're definitely there
• 26:48
  for more complex models but I think in general, like once you
• 26:52
  put a simple model Aeneas, like make sure you have metrics that
• 26:55
  are that you can either support or believe are like connected to
• 26:59
  this. Yeah, this is the model that
• 27:02
  your building. So I think that's definitely is
• 27:06
  you need to pick them first. I think it's the one thing that
• 27:07
  that people also talk about is like don't pick metrics after
• 27:11
  doing it, right? Okay, these veterans went up,
• 27:14
  you know, you kind of to be like, okay, what are we, what
• 27:16
  are we watching for and pick those up in the first place and
• 27:19
  then then watch them versus the other way again?
• 27:23
  I think I think sometimes we like doing that right.
• 27:25
  Like it's like oh, we're gonna pick these metrics?
• 27:26
  Now that we're done. Yeah.
• 27:28
  You see what moved and what actually looks better and you
• 27:32
  go? Oh well it must Going to
• 27:33
  success. Look at all these metrics that
• 27:35
  went up. Yeah.
• 27:37
  But it could have been luck. Yeah, exactly exactly.
• 27:41
  One thing I wanted to ask you about, was this idea of
• 27:43
  communication, because I know you've blogged about it quite a
• 27:46
  bit and the importance of communication, how communication
• 27:51
  can enable collaboration with data scientists?
• 27:54
  And the like, I think it was a program that you have about a
• 28:02
  data leader workshop. And so it's basically like
• 28:06
  amplifying the abilities for data scientists to clearly
• 28:11
  communicate. And so how, how can we do that?
• 28:14
  How can we be better communicators?
• 28:16
  Yeah, I mean I think there's a there's a couple ways, right?
• 28:19
  Like one is just having that ability to be more transparent
• 28:23
  in your work. And then also, just, I think,
• 28:26
  one of the things that I heard about a lot of, like, kind of
• 28:29
  see a leadership that I would talk to you on the problems,
• 28:32
  they would have is like He said, we'd hire these really smart a
• 28:34
  scientist and they would like right up these ports, but they
• 28:37
  would never either have a conclusion or they would never
• 28:39
  like kind of like distill it all down into very concise, you
• 28:44
  know, like what why is this important kind of thing.
• 28:46
  And so I think that's one place I saw that like, could be
• 28:50
  improved a lot with some people like, yeah.
• 28:52
  It's like just have kind of a perspective or point of view or
• 28:56
  conclusion it at the end of all this research, right?
• 28:58
  Like you did all this stuff you gave this your manager, all
• 29:00
  these numbers but you didn't really say what they mean.
• 29:02
  Mint and so I think there's this kind of that point that's really
• 29:07
  important. And then I think General also
• 29:11
  you know just from a collaboration standpoint you
• 29:13
  know being able to say what you're doing and be able to
• 29:16
  being able to kind of say the value.
• 29:17
  I think it's very important, I think some sometimes, you know
• 29:22
  and I have the same problem. I think sometimes us is like a
• 29:25
  scientist and Engineers software Engineers.
• 29:26
  We get very tied up in the the how I think.
• 29:30
  I think a lot of us get tied up in the house and things going to
• 29:32
  get done. Not necessarily the value of it
• 29:34
  like the business value of it. And so like trying to step away,
• 29:37
  not just from, you know, saying like oh I'm going to build this
• 29:40
  model. It's going to give you like 90%
• 29:42
  accuracy on this, you know, whatever.
• 29:44
  And it's like great. But like how does that impact my
• 29:46
  business again like kind of going back to like you know
• 29:48
  whether it's that bottom line that sales and average sales
• 29:51
  price and making sure you kind of think through that as well.
• 29:53
  So you can talk to the to your business owner and be like, hey
• 29:56
  if we do this well you know, we could, you know, improve X or Y,
• 29:59
  you know, reduce costly or something, where it's like, it's
• 30:02
  very clear. And you can very clearly tie it
• 30:04
  to the business. And so part of communication is
• 30:06
  more like empathetically understanding what the business
• 30:08
  owner needs and I think like you know I think that's always like
• 30:11
  that whatever listening versus Talking kind of concept where
• 30:14
  it's like you have to just as much listen and understand
• 30:17
  someone else and understand what they want and not just like,
• 30:20
  kind of what you want, what you want to do.
• 30:22
  So I think I think that's kind of the aspect.
• 30:23
  I usually try focusing on in hearing towards, as far as like,
• 30:27
  you know, improving communication especially from
• 30:28
  from technical people. And, again, I don't prefer it
• 30:31
  from it just as much as I think any.
• 30:32
  Technical person does, maybe not to the same degree.
• 30:35
  Some people I'm always impressed on their technical knowledge and
• 30:39
  I'm just like, man, I don't have that.
• 30:40
  But then like, you'll hear them try to communicate.
• 30:41
  It's like oh, all you talk about?
• 30:44
  Oh yeah. Like, were you statement?
• 30:45
  Whatever CPU, like, any like great.
• 30:48
  How does that help us? Like from the business?
• 30:50
  I can like, what you realize that the business side mostly
• 30:53
  what they care about is like the bottom line, right?
• 30:55
  Like give it, how is this going to let us improve, you know,
• 30:57
  customer satisfaction or whatever, they're looking for
• 31:00
  the dollars and cents and trying to That story from their
• 31:03
  perspective just as much as for my hours.
• 31:05
  Have you seen when you go into these companies like the
• 31:10
  importance of the different roles?
• 31:12
  And I know somebody that I was talking to was telling me how
• 31:17
  they feel like they're a translator first and it is
• 31:21
  exactly what you're saying. Like, the business side wants to
• 31:24
  know one thing but then the technical side wants to talk
• 31:27
  about another thing and you need somebody in there to translate,
• 31:30
  but have you seen other Roles that really helped like the gel
• 31:36
  of the team. Yeah, I think generally when it
• 31:40
  comes to will try to do, they'll have like either something like
• 31:42
  business analyst or some similar role where they basically
• 31:47
  they're maybe not technically as Savvy as like an engineer would
• 31:50
  be, you know, there may be sequel Savvy or they may be you
• 31:53
  know very and analytically Savvy but they're also very Savvy on
• 31:57
  the business side and so they usually try to play that that
• 31:59
  middle role. Of course then you are you end
• 32:02
  up with other problems where it's like you know, the business
• 32:04
  analyst, want some data. But the data Engineers, you
• 32:06
  know, bog down Find millions of other data request and then
• 32:11
  that's the whole other set of issues.
• 32:12
  But yeah, I think, as far as, you know, rolls go, I think
• 32:16
  that's generally why companies like to have, like some sort of
• 32:17
  business analyst, or someone that's like that middle ground
• 32:20
  between, you know, Tech and business because yeah.
• 32:22
  There's this weird, this weird Gap and it's like, it's like, I
• 32:26
  remember talking about it in college.
• 32:28
  Like they literally, they had like, the two parabolas, right?
• 32:31
  Like they were like Tech business, they're not like, they
• 32:35
  don't think the same. We don't use the same, you know.
• 32:38
  Larry. No.
• 32:38
  Do we need someone kind of Middle Ground to like the two
• 32:41
  branches and so so yeah. It's a very common thing.
• 32:44
  It's interesting. I was weird how much it is that
• 32:47
  way? Yeah, so yeah, I totally I think
• 32:52
  in general though. I'm sure be a or something.
• 32:55
  No, she says, I've seen the value of that directly.
• 32:57
  So I've done some contract work. We're adding a business out,
• 33:02
  analyst actually added a lot of value because you had, you know,
• 33:05
  someone more senior in the company doing.
• 33:08
  A lot of things, you know, like, having to keep up with the
• 33:10
  developers. All the other contract workers
• 33:12
  ability to really small company and it's just a lot of
• 33:15
  inefficiency. So sometimes makes much more
• 33:17
  sense to have someone like just, you know, focus on that one roll
• 33:20
  and it does allow the opportunity to Branch out into
• 33:23
  other things, right? Like you get a good amount of
• 33:24
  experience working with SQL, you're analyzing data bases, you
• 33:27
  know, doing all that eventually, maybe pick up a new skill.
• 33:29
  Now, you can model it now, I do think there's some benefits like
• 33:33
  differentiate the roles because it can be really difficult when
• 33:35
  everyone's kind of doing everything.
• 33:38
  And I've seen that you know like the challenges of that in a
• 33:40
  small company like the startup that's working with a lot of
• 33:43
  contractors is tough, you know. Anyways this has been so
• 33:47
  awesome. I feel like we wrote down a lot
• 33:49
  of good thing. I've been taking notes as to
• 33:50
  some of the things you've been saying but the last thing we
• 33:53
  like to do want to close out with this.
• 33:54
  So we like to leave away with some sort of, like, advice on
• 33:57
  on, you know what best practices you have learned, what things to
• 34:00
  help her for you. But before we can get into that,
• 34:02
  one of the things I'm curious about because you write about a
• 34:04
  lot of different topics is, what are your interests?
• 34:06
  And like what something right. Now that you're really
• 34:08
  passionate about at the moment, is it a technology, a new
• 34:11
  framework. Maybe you're looking at old
• 34:13
  stuff to refresh your memory, what's the thing that you're
• 34:15
  interested in right now? Yeah.
• 34:18
  So I think as far as like, say so interest go.
• 34:25
  Sorry. I'm like suddenly just stop
• 34:27
  thinking, oh, no good data. So the cloud you'll talk about
• 34:32
  the cloud, right? You have a lot of papers on
• 34:35
  infrastructure ETL, I saw a lot of stuff on each cell so that I
• 34:38
  bet you have a lot of experience doing a bunch of that so is it
• 34:41
  usually related to like your day-to-day work or do you try to
• 34:43
  maybe? Is there something that's new
• 34:45
  that that you're learning at the moment?
• 34:47
  Yeah. So it's a combination of both,
• 34:49
  right? Like so I think as far as like
• 34:51
  what I like, or what I'm interested in, I'm very much
• 34:53
  interested in trying to help a lot of again.
• 34:55
  I think small companies get themselves to a point where they
• 34:57
  can take advantage of a lot of this stuff that exists that is
• 35:02
  now affordable. I think, you know, it's like,
• 35:04
  you know, before, you know, having a full-time data,
• 35:07
  scientists having Time. Do you engineer managing, you
• 35:09
  know, data warehouse or an ETL would be super expensive, right?
• 35:13
  Like, especially nowadays, when you know, the average software
• 35:15
  engineering data engineer cost, somewhere in the range of 120 to
• 35:18
  150 K, you know, a piece a year, you know, and that's not even
• 35:22
  cost, you know, putting in AWS charges and all of that, you
• 35:25
  know, there's something, you know, you're spending three
• 35:27
  hundred thousand dollars on, you know, just trying to answer some
• 35:30
  question that you don't know will bring you value or not.
• 35:32
  And so I think I've been very focused and very interested in
• 35:35
  like ways to help small companies like really approach
• 35:38
  these problems without having to take on that burden of cost.
• 35:40
  You know, whether it's using tools similar to like 5 train or
• 35:43
  Stitch because you know you can have this locodoco dtl kind of
• 35:46
  approach where, you know you you can take a lot of the advantages
• 35:50
  of having something that's like an ETL or elt without having to
• 35:53
  do a lot of that heavy lifting of having to code every single
• 35:55
  connector and so can't questions, you know your.
• 35:59
  You can still answer them without having to hire a
• 36:02
  full-time data engineer. Maybe you just need a higher,
• 36:04
  you know, again, someone like me or someone else you have to
• 36:06
  consult and to bring a Data in to help you out.
• 36:09
  Just to answer those questions. Hi, Herb and higher Benjamin
• 36:13
  guys are listening to. You gotta have.
• 36:17
  I just I mean that like regardless they can be any like
• 36:20
  you know, I think I think I'm seeing definitely a lot more
• 36:23
  people looking into that Consulting kind of like
• 36:26
  specially in this space. No, you right?
• 36:28
  Yeah. It makes sense to for a lot of
• 36:29
  things especially if you're just getting started and trying to
• 36:32
  like, you know, oh, you know, I think some people they rush into
• 36:35
  applying ml into their, maybe their app or their business, And
• 36:38
  it's like, sometimes those two things don't mix, easy have to
• 36:40
  do some basic things like, you know, organize your data and
• 36:43
  have a good sense of what you're trying to do.
• 36:45
  But yeah, if you're any worse yet, I just, just a thought on
• 36:48
  that. No, yeah, no.
• 36:49
  I told I totally agree. So, yeah, I think that's where
• 36:51
  I'm like, really interested in. Like, you'll see, even some of
• 36:52
  the topics I've been writing on recently.
• 36:54
  I've been like, you know, how you can do stuff for a small
• 36:56
  business just to, like, start taking account of, like, what do
• 36:59
  you have, right? Like, because I'll get a lot of
• 37:01
  small businesses businesses are using things like, Salesforce,
• 37:04
  or Saint are some tools that like they have data connectors,
• 37:06
  right? Like these are things that like
• 37:08
  Didn't exist, 20 years ago or something similar words?
• 37:11
  Like yeah, you could connect to your like old Erp system, so it
• 37:15
  was really hard, right? Like you wouldn't even want to
• 37:17
  if you're a small company, but now it's like super easy, you
• 37:21
  know, and even more. So, again, with these low code,
• 37:25
  no code, kind of tools, you can do it.
• 37:28
  Technically, you know, your business owner, could do it.
• 37:30
  Should you probably not? I say the same way as like,
• 37:32
  would you do your own accounting or your own legal work?
• 37:34
  Probably not, probably just hires it, just saves you time.
• 37:38
  Could you probably do it? Maybe.
• 37:39
  Yeah it's not that you couldn't, it's just like sure.
• 37:41
  I can also make my own clothes I guess.
• 37:44
  But that might be really tricky to sit there and just like space
• 37:47
  because our foes. Did you guys check out this
• 37:49
  shirt yet? Look at him.
• 37:54
  He's honest. Yes, this is on fire, man.
• 37:57
  So I wanted to ask about when you go into a company and you're
• 38:04
  looking to start before you even build anything out, I You're
• 38:08
  trying to answer these key questions.
• 38:10
  Do you have a process about how you go about figuring out if
• 38:14
  even data science or m/l is right for this company and how
• 38:19
  you can figure that out? Yeah.
• 38:21
  I mean like I think this like, kind of like a almost like a
• 38:23
  date of maturity, kind of like, set of questions like, you ask
• 38:27
  again, like, what kind of data do you have?
• 38:29
  I think it's like question one, right?
• 38:30
  Like, do you even have data? Like I think I once had someone
• 38:33
  be like I have this list with emails on it and like okay I
• 38:36
  can't do much like what What do you want me to do with that?
• 38:39
  Like, I like you. Do you have dates when you got
• 38:41
  those emails. So I couldn't even tell you like
• 38:43
  you're you know increasing or decreasing you know some sort of
• 38:45
  weird time series model like at least I could tell you like how
• 38:48
  many people come on average and having you can expect in the
• 38:51
  future. But yeah, like like just
• 38:53
  checking out like your data, it's like kind of doing it in
• 38:55
  the inventory. Kind of getting an idea of what
• 38:58
  the person even wants like, what is their end goal.
• 39:00
  I think it's always a big thing. It's like, trying to understand
• 39:03
  like what kind of initiative or project, or they actually even
• 39:05
  try to take on. Like, all right, are they trying
• 39:07
  to reduce costs Kris whatever. Because if you don't understand
• 39:10
  that, like, you know, it's that's what's going to drive
• 39:12
  whatever you're doing. So trying to understand their
• 39:15
  business goals. Because again, like you said,
• 39:17
  you might not even need a fancy model.
• 39:18
  And like, if you understand their business goal, it's like,
• 39:20
  well, we could probably do that with some basic, you know, your
• 39:23
  kpi or, you know, some some sort of basic analytics, I can answer
• 39:26
  that question. You don't need fancy ml ml model
• 39:31
  so I can try and understand their goal and where they're at
• 39:34
  so many questions. As far as kind of thinking that
• 39:37
  I was like, go. Al's data sources.
• 39:40
  I haven't processed somewhere written out but I that's why I
• 39:42
  have it written out have to link to that thing because that
• 39:46
  sounds really interesting is that means I don't do you have a
• 39:49
  Blog specifically about that? Don't ya?
• 39:51
  I do like and I can send it to you guys later so I sweet.
• 39:54
  Yeah. So I have a I have that process
• 39:58
  at least kind of right now just to kind of like help guide
• 40:01
  people to be like okay like where are we you know, I think
• 40:04
  that's very important just to understand where you are.
• 40:08
  Kind of data data data analytics and data, kind of strategy
• 40:12
  maturity in general. Yeah, so it's like also
• 40:15
  understand what tools you have. Oh, also understanding like what
• 40:18
  your team currently has as far as like skills, right?
• 40:21
  Like if your team can't maintain this thing, it doesn't matter,
• 40:24
  right? Like you also have to kind of
• 40:25
  gear your Solutions towards the team, right?
• 40:28
  Like you're not going to build a python solution for a team
• 40:31
  that's building in Java or or even more.
• 40:33
  So you're not going to build a python solution for team.
• 40:35
  That's a bunch of analyst, right?
• 40:36
  Like it's like maybe but they support it, right?
• 40:38
  Like so you guys have seen about like support I think I see that
• 40:41
  a lot wouldn't like even like data engineering.
• 40:43
  I'll see like analysts who want to build like the engineer
• 40:45
  pipelines and it's like yeah. But then you have to maintain
• 40:47
  them like so when it breaks that means you need to come fix this
• 40:51
  like and do you want to do that? Like, I think a lot of people
• 40:55
  don't understand that, a lot of engineering and maintenance like
• 40:58
  things break, easy to migrate things change.
• 41:00
  And that's what we spend. 70% of our time doing it just like, oh,
• 41:04
  something broke. We better go fix egg diversion
• 41:06
  pretty much. This is this API just change.
• 41:09
  We gotta go, you know, retains the connections.
• 41:11
  Great you know. So yeah, so yeah.
• 41:15
  Go ahead and start to meet you as hear anything else you want
• 41:18
  to add? I don't want to cut off the
• 41:19
  Jimmy Choo stock. I was just writing that down.
• 41:22
  I wanted to make that a little quota, 70% of engineering is
• 41:25
  maintenance things, break all the time, all the time.
• 41:29
  Awesome. All right.
• 41:30
  So the last thing. All right, this is what we like
• 41:32
  to do. We want to.
• 41:33
  If there is any nugget of wisdom, it could be something
• 41:35
  that's all your mind at the moment.
• 41:37
  Something Like maybe something you were recently dealing with
• 41:40
  what's like one nugget of wisdom that you like to share with the
• 41:42
  community. Again, it could be something
• 41:44
  really small. Something big anything, anything
• 41:46
  that you wish, I don't know. Looks like went out like the
• 41:53
  most recent challenge. So it's probably around the
• 41:55
  data. Let's say something you're on
• 41:56
  the data you mentioned earlier. Doing like I'm you mentioned so
• 42:00
  many good things. We're going to have to linked to
• 42:02
  a lot of this of the YouTube im gonna teach em.
• 42:05
  But yeah anything anything that's like relevant.
• 42:07
  I think the biggest thing that I've recently, kind of, kind of
• 42:11
  work been kind of my internal Mantra more because like I had a
• 42:14
  recent project that there was just a few hiccups and it was
• 42:18
  the old kind of tied to a combination of factors in, like
• 42:21
  one of those factors being that, like, there was multiple kind of
• 42:24
  companies working together to try to bring something about and
• 42:28
  I think, you know, the concept of like taking ownership
• 42:30
  regardless of the, you know, who's responsible for what is
• 42:35
  really important. It's partly communication.
• 42:37
  But also partly like, you know, you kind of need to be willing
• 42:40
  sometimes to be very direct with people and I think, you know, we
• 42:44
  can all work on that and I think there were there like, looking
• 42:46
  back at that. Like yeah, it wasn't like the
• 42:48
  smoothest project that I think I've ever done and, you know, I
• 42:50
  think their products but it definitely looks smoother.
• 42:52
  So, it's like, one of those things where it's like, well,
• 42:53
  what could have made it smoother and like, looking back, it's
• 42:55
  like, okay, if I had been more direct and taking more ownership
• 42:58
  of things, you know, it would have made more sense like
• 43:00
  regardless of if it was mine or not just taking ownership of the
• 43:03
  process. Like I think that's one thing
• 43:04
  where it's like, you know, sometimes the process itself
• 43:06
  doesn't have anyone. You know, because there's like
• 43:08
  three or four people involved in three or four different
• 43:10
  companies it's like well who should be in charge.
• 43:13
  It's you know it's like it someone needs to have like that
• 43:16
  ownership of that process because otherwise you just have
• 43:19
  you just eventually see things kind of crumble because like no
• 43:21
  one's taking can't you like having accountability for
• 43:23
  something and I think I think that was something that I really
• 43:26
  picked up or it's like yeah you know regardless of the situation
• 43:30
  you kind of got to take like some ownership of something so
• 43:33
  that you know things are really going.
• 43:35
  Otherwise, you know, if you're constantly waiting for someone
• 43:37
  else to do it, No one's going to do it.
• 43:39
  Yeah, I think it's so valuable that the one thing that I
• 43:42
  thought was really about like practical, like, application of
• 43:45
  that is that you mentioned doing a write up with in, and out
• 43:47
  after an analysis like just share in summary in the, in a
• 43:51
  concise way that someone non-technical can understand the
• 43:54
  value of this work. I think that that is something
• 43:56
  or some effort there in translating, what you have done,
• 43:59
  and communicate it in a way that makes sense to someone.
• 44:01
  But that also allows you to take more ownership of the process,
• 44:04
  right? Like don't just expect things to
• 44:06
  automatically work and dearie. Reducible and they have all
• 44:08
  these great things that you actually have to do that.
• 44:10
  And I think that you've laid out many concrete examples of what
• 44:13
  that looks like. So that's super valuable.
• 44:15
  Guys, anyone listening out there?
• 44:16
  I really appreciate that. That's something.
• 44:17
  I'm Duffy going to take myself to be more direct is, yeah, a
• 44:20
  lot of times, like it just easier to kind of just like pass
• 44:24
  that responsibility to someone else maybe someone on a
• 44:26
  different team, but it makes things more complicated in the
• 44:29
  end. So I think that's a good piece
• 44:30
  of advice. Yeah, passed the buck.
• 44:33
  Yeah, that that's what we got. Awesome, man.
• 44:37
  This has been so cool. I want to thank you Benjamin for
• 44:41
  coming on here and mr. Data, mr.
• 44:44
  Seattle data guy. I love that name.
• 44:48
  That name is awesome. Yeah, I really like that name.
• 44:50
  Had it, by the way, how did you come up with that?
• 44:51
  Was that just like something you like, I'm in Seattle.
• 44:54
  I like data. Oh my God.
• 44:56
  I think he's got he's got a ring to it, right?
• 44:59
  Like it's like the lizard guy or something like that.
• 45:07
  You call for your child, third birthday.
• 45:11
  That brings lizard-like. Like it's guys got that ring to
• 45:15
  it. I think people kind of get it.
• 45:17
  I totally get it. Yeah I think that was like with
• 45:19
  a guy make sense is amazing. Yeah, it's a really cool way to
• 45:26
  remember it and it makes it easy like if I'm on medium and I'm
• 45:32
  like, who is that guy that I wanted to follow?
• 45:36
  Yeah. Oh yeah.
• 45:39
  That's right. We're going to link to a lot of
• 45:47
  the resources that we've been mentioning and talking about
• 45:49
  some really good articles or blog articles basically.
• 45:53
  And yeah, if there's anything, if you guys have any questions
• 45:56
  about, you know, some of the things we talked about, please
• 45:58
  reach out to me Demetrius or Benjamin himself and we really
• 46:01
  appreciate the time that you made speaking with us Benjamin.
• 46:04
  It was their Benjamin. If anyone wants to get in touch
• 46:06
  what's the best way to You yep? Yep, yeah, I mean you can you
• 46:10
  can pay me on LinkedIn for one, you know, Benjamin Roshan.
• 46:14
  I'm that's another very unique things.
• 46:16
  Like, if you Google me, I think I show up very quickly or you
• 46:18
  can go to the website and go to our contact page.
• 46:20
  So the Seattle day, the guy.com just reach out to us, feel free
• 46:24
  to contact that. You check the email, you know, I
• 46:27
  try to answer questions even if they're not necessarily about
• 46:30
  Consulting. I'm trying to also, you know,
• 46:32
  just be, you know, help other people who are probably, you
• 46:35
  know, starting their Journey, whether it's in data science,
• 46:37
  It's indeed engineering, whether it's in Consulting.
• 46:40
  I try to be, you know, pretty open as far as the entry
• 46:43
  questions and how I can get more information like that.
• 46:46
  Sometimes that's like what you'll see as far as blog post
• 46:48
  go. I think like one of the more
• 46:50
  popular ones was like a day of science interview guide where
• 46:52
  was just like a Google sheet that I put together.
• 46:54
  It was like Hey when I was interviewing, this is what I
• 46:57
  use. Like you know this is a great
• 46:59
  way of just tracking what you've already studied and what you
• 47:01
  should study and breaking it down because again there's so
• 47:03
  much information out there. It's like just have a concise
• 47:06
  clear way that You can track yourself so that, you know what
• 47:09
  you've done and know what you haven't done.
• 47:10
  Awesome. So cool.
• 47:12
  Well, thanks again. And to everyone out there.
• 47:16
  We will see you next time in our coffee session.
• 47:19
  Thank you both. Thank you, David.
• 47:21
  Thank you dimitrios. Cheers you.
• 47:23
  All right?