Latest / Agentic Conversations (formally mlops.community) / Non-traditional Career Paths in MLOps // Matthew Dombrowski // MLOps Podcast #144

Transcript

Non-traditional Career Paths in MLOps // Matthew Dombrowski // MLOps Podcast #144 · Agentic Conversations (formally mlops.community)

0:00
Yeah, my name's Matt Dombrowski everyone.
0:03
I'm a technical product manager and Amazon and ID, make coffee
0:07
these days as a pour-over party of a 60, sometimes in the summer
0:13
over, ice that usually a straight hot forever.
0:17
What's up, everybody? We are back for another ml Ops
0:19
Community podcast. I am Demetrios your host and
0:24
today I'm joined by none other than mr.
0:28
McHale Eric. Who oddly enough just came out
0:33
with a Rick and Morty GTP chat something explain yourself man?
0:37
What is this? That you just created?
0:40
Well I didn't realize I was gonna have the the mic to be
0:43
able to pitch here. But yeah basically worked on a
0:47
little project big fan of Rick and Morty, big fan of what's
0:51
going on with generative. A I thought I went to see if I
0:53
could Empower people or you know my other person I worked on this
0:56
with we could Empower people to make their own Rick and Morty
0:59
episodes. So we use the latest and
1:01
greatest in GPT technology plus stable, diffusion plus Dolly.
1:05
All that good stuff and you can have Rick and Morty.
1:08
Go on all kinds of crazy Adventures, So if you are
1:12
playing buzzword bingo, that was just about every word.
1:16
You could have expected in the to God in the cloud with a
1:22
little bit of crypto thrown in because we're making it over,
1:24
drop it at to F3 decentralized and bingo.
1:29
There we go. So everything.
1:31
You are democratizing Rick and Morty.
1:34
And that is what exactly is it? I didn't really see it enough.
1:38
I just saw your tweets about it. And I guess it's that use GTP to
1:44
write the script and then use stable, diffusion to create the
1:48
episode like the visuals. That's right, that's right.
1:50
So it's a basic a storyboard, you get the dialogue and the
1:54
plot points and then accompanying visuals.
1:56
As like the episodes of of the story that you've in some sense
2:00
created dude, how have you not gotten sued, you know, it that
2:08
very well might happen, I just feel.
2:09
Okay. So, here's the Thing is we were
2:11
trying to get. We were like when we were doing
2:13
the Twitter announcement, we were trying to act the Rick and
2:16
Morty official account because we were trying to get get a
2:18
little bit of publicity from them.
2:20
However, we never heard back from them.
2:23
So we kind of imagined that the first time we're going to hear
2:25
from Rick and Morty at all is going to be when they're
2:27
lawyers. Send us like a CSUN is.
2:29
Yeah. So hey anything any publicity is
2:33
good publicity, right? Isn't that the name of the game?
2:36
Yeah. Just I think the app right dude?
2:39
Seriously. Or just going well no, I won't
2:42
even say that here or not because I hold you in high
2:46
esteem and I will not put you on the same playing field as mr.
2:50
SPF but I will say that the director of Rick and Morty or
2:55
the creator of Rick and Morty. Super cool.
2:57
So I would think that he might be down with it.
3:00
I don't know. Oh, no.
3:03
Next, do community and just do all of the work that he's ever
3:07
done. The same creator of community
3:09
the show, not Community. Like in general, is the same guy
3:14
Dan something I think Daniels name that's it.
3:17
Dan Harmon so he's an incredible story teller and I'm glad that
3:21
you got GTP to follow his lead. Anyway, we're on here with Matt.
3:27
The dude is incredible. He's your fellow co-worker and
3:31
He also is, he's been doing some super cool stuff around project
3:37
management product management. He talked to us today
3:41
specifically about the different ways you can be in machine
3:46
learning, but not have to be the one that's actually writing the
3:50
code. And I find this fascinating
3:52
these days to because of chat GTP that came out and so many
3:57
people are crying that. The sky is falling and all of
4:01
the coding. Jobs are going to be obsolete
4:03
because this chat, GDP can do everything, but I look at it
4:08
very much, like, I look at Photoshop.
4:10
You aren't going to have this completely replace people's
4:14
jobs, but you're going to have it augment people's jobs.
4:18
And what, Matt talk to us today about was very much along the
4:22
lines of, hey, if you're not a coder, but you still have the
4:27
machine learning Acumen. What can you do?
4:30
Where are places that you can fit?
4:31
It into this fear. What were some takeaways that
4:34
you had? Yeah hey Matt Matt's awesome as
4:38
you suggested and alluded to he and I go back we go back to
4:41
Alexa days and way back. He was he was the guy right.
4:45
That when when we were trying to get anything running on on Alexa
4:48
systems we would talk to him. Basically him and his team.
4:51
So mad respect big fan of Matt a super smart guy.
4:55
I actually didn't know that his career progression really went
4:58
from early stage startups through big company Solutions,
5:03
architect through p.m. technical.
5:05
PM technical program manager roles, so he's really spanned.
5:08
So he's really seen, I think ML and data roles in a lot of
5:12
different context. I think we got to to see the
5:15
different insights that provides I mean, in a sense, they're so
5:17
very different, right? There's a lot of ambiguity,
5:19
which he talks about and different opportunities for
5:22
these different roles. At the same time, it was
5:24
interesting to also hear some of the commonalities, right?
5:26
How being a Solutions architect across different roles, you
5:29
know, in some sense there's a lot of very similar similar
5:31
responsibilities and then once your product manager, And a
5:35
program manager, how there's a lot of the same soft skills
5:38
around alignment that need to need to exist.
5:40
So that was really cool to see that in a sense, there's a lot
5:43
of diversity in the types of responsibilities.
5:46
You can have it at the same time.
5:48
There's commonalities, right? And finding ways to navigate
5:52
that is I think what leads to successful career in machine
5:55
learning. That's so good.
5:57
I loved what he talked about how to get stakeholder alignment
6:00
and, you know, from working at Amazon and I'm sure people
6:05
people find out real quick, when they start to work in big
6:07
companies, there's so many processes in place and you have
6:11
to get signed off from so many different people.
6:13
You want to figure out what's the best and fastest way that I
6:17
can do that. So, I can actually show that I'm
6:19
producing something of value, and he gave us a few nice,
6:23
little hacks. So check those out.
6:26
And let's get into this conversation.
6:35
All right, dude, it seems like we've got something in common,
6:39
you are a pizza and coffee nerd. I also consider myself a pizza
6:43
and coffee nerd. Give me the breakdown.
6:46
Yeah, so I hail from Connecticut, which is not
6:51
notable for may be many things, but one thing is notable for I
6:55
think is pizza. I'm from New Haven so, or like
6:58
suburbs of New Haven. And yeah, I mean, my position on
7:02
that is New Haven as the Undisputed Pizza.
7:05
Capital of the world. Wait, do you make your own?
7:07
I do. Yeah.
7:09
Yep. And what are you you make it New
7:11
Haven style. I do.
7:13
Oh yeah that's kind of my specialty.
7:16
LT is maybe even style. Every one of those uni portable
7:20
pizza ovens. I do I was going to ask.
7:23
If you got out of it, I feel like half the country bought
7:26
those ring. Yes, bread, making your
7:29
sourdough or something. So, yeah, I fell victim to that
7:33
trap to you. So that became that and coffee
7:35
became like my covid, like, rabbit, hole Obsession.
7:39
You originally mentioned to me that you had all kinds of ideas
7:42
around the greater field. Of machine learning and how
7:47
sometimes we can get stuck thinking that you need to be
7:51
either a data scientist or an ml engineer.
7:54
But there are so many other places that you can play in the
7:58
cycle of putting out machine learning products.
8:02
So why don't we start? There?
8:03
Can give us how you're seeing that.
8:06
Yeah, that's great point. I think it's true.
8:08
Like a lot of people get wrapped up into, you know, like what
8:13
algorithms do. I need to study data pipelines,
8:16
you know, like data infrastructure and like fast
8:20
forward and think about the technology and, you know, as a
8:24
product manager maybe I'm biased.
8:25
But I think You know, the totality of what goes into
8:30
building animal products, there's actually a lot that goes
8:33
into, you know, building and launching actually a successful
8:38
product not just a technology products and know any think
8:43
about not just the engineering or the science.
8:47
There's also a bunch of go-to-market type activities,
8:50
customer-facing activities. There's like internal project
8:54
management type stuff and I think like when you take that
8:58
all together, I think we talked about machine learning and like
9:03
the data category and usually, you know, we're talking to ml
9:08
Engineers or scientists or software engineers and really
9:12
kind of like pigeonhole those roles to be what they are
9:17
brothers 12 other. I think there's a lot of other
9:19
opportunities out there besides just those roles and especially
9:22
like big companies to learn kind of like roles or like
9:25
non-traditional parts of the Ml ecosystem early career?
9:30
Yeah like career paths in the ecosystem, begging people, add
9:33
value if you don't, maybe they don't want to write python for
9:36
the rest of your life. There's one thing that I think
9:41
it's great that we're already touching on.
9:44
Maybe these non-traditional career paths and ml which is
9:46
which is something I think we've mentioned discussing.
9:51
I'd be curious to hear maybe start off just to hear about
9:54
your progression through the MLS L fear from different roles
9:59
because, you know, when we were together I remember use these
10:02
like this. Like deep technical person
10:04
working on implementation, building out, core
10:06
infrastructure. And now you can transition to
10:08
p.m. but love to hear I'm sure the listeners are also would
10:11
love to hear just what has that non-traditional career path look
10:14
like for you. Yes boiler alert.
10:17
This is a reunions of the to AWS folks here I'm the only one
10:23
that's not got a WS in my blood Alan this call.
10:27
We're gonna start making some inside jokes and you just have
10:30
to phase out for that. Me and the listeners will go
10:33
grab some coffee or whatever. We do anyways, man.
10:37
Yeah, so I guess I started as a computer science major in
10:40
college and after college I think didn't love what I did in
10:44
school. Back in Moses 04, 2004 we did
10:49
you know we did convert science in college.
10:52
It was like Enterprise Java and I found it like very
10:55
soul-sucking and hated like Object-oriented design.
10:59
And like, just it felt like I didn't want to do that for a
11:01
living for the rest of my life. And one of the Air Force
11:04
actually, because I wanted to be a pilot when my eyes and I can
11:07
some, I issues didn't work out and so going there for us for a
11:10
while, launching Rockets, not super related.
11:13
But after I get out and a lot of in there, for security
11:16
experience of did some network security, pentesting type stuff
11:20
as a consultant and then moved to start up that my friend had
11:25
start was called localytics back then.
11:27
And it did basically mobile analytics and I started there as
11:32
a Solutions consultant actually. So this is kind of like an
11:35
onboarding role. Like working with customers to
11:37
help them get started with the product and become successful
11:40
and that's kind of like we can do some kind of like ding or
11:44
sound or that's one role we can talk about to you is like
11:47
Consulting and like implementation Consulting
11:50
onboarding, you think is like definite need for if you're
11:53
building kind of like ml products or kind of any products
11:56
ass products, there's probably someone that is there to help
11:59
like kind of provide White Glove service to help customers become
12:02
successful and that's what I did there.
12:04
Moved until a Solutions. Architect rule more in helping
12:08
customers if they have more hairy you like amputation issues
12:11
like custom implementation issues and did some kind of like
12:15
post-sale Consulting implementation work.
12:19
So, any like custom Professional Services, they would kind of
12:22
needs would help build those Integrations with customers and
12:27
did a lot of Work naturally with AWS in that role being a
12:30
start-up. They were pretty much all in on
12:32
AWS and moved to of course Amazon you need to be us as a
12:37
symbol, a Solutions architect and that in that role some
12:41
solutions architect is like a term.
12:42
That means a lot of different things, a lot of different
12:44
companies, but Amazon and AWS, at least it's kind of similar to
12:47
like other companies. I would roll called the sales
12:49
engineer, where you're working with, you're probably paired
12:53
with salesperson or like, team of sales people.
12:56
And you're providing Hiding, like technical guidance to
12:59
customers. Usually again, like in
13:02
consultation, with salesperson. So, a lot of roles have kind of
13:05
like pre sales consultant or like sales engineer.
13:08
If you will all that kind of depends on the products and how
13:11
they go to market. But like a pre sales consultant
13:14
that helps answer technical questions, and overcome
13:18
technical hurdles and sales process to more or less, get big
13:22
deals done. And then after sales done, maybe
13:24
the customer moves into like post-sale.
13:27
Hang in the work with that onboarding or like Solutions
13:30
consultant type person and get them like implemented in
13:33
on-boarded and but a 2B assets a little bit different because
13:37
yeah an AWS we don't necessarily have like deals or else.
13:41
I like SAS product. Very sign up or product, very by
13:45
like a big patch of software. It sailing so Solutions,
13:48
Architects or kind of like paired with a customer and maybe
13:51
it's one-on-one depends on the size of the customer, but like
13:54
some customers have big team of solutions.
13:57
Architects working with all their different words and like
13:59
different verticals. I like a different specialty
14:02
areas and then other likes water.
14:05
Customers might have one solution.
14:06
Architect that covers thing, many customers and like the
14:10
smaller startup segment. But yeah.
14:13
And then like Enterprise customers might be paired like
14:16
one-on-one with like an account manager or salesperson, the
14:18
cover, kind of like a patch of customers that do work with.
14:22
But yeah. And then, yeah, I guess from
14:26
Solutions are Move it into a product manager role moved from.
14:30
It'll be S to the Alexa team. The Amazon was in Boston, and
14:37
there was doing a similar role like Solutions.
14:39
Architect more for like internal customers or internal customers.
14:42
And Alexa hell was one of them were building building like
14:46
models with AWS and Alexa and like, for Alexa use cases as I
14:52
do. So up until this point, you had
14:56
dealt with Machine learning. It was when you went on to the
14:59
Alexa team that you started doing, which is leveled, mammal
15:04
staff. As we were doing some more of is
15:06
called machine learning. I guess we call that like
15:08
Predictive Analytics analytics and that was the code word.
15:12
Yeah, the startup. I was working at.
15:14
We did some of that was like we could tell maybe like when a
15:19
customer was going to the churn from a mobile app they're using.
15:22
Or when a customer would like click on a specific, push
15:25
notification. We do these kind of we call
15:27
them. Time Predictive Analytics.
15:28
So I started doing some of that before machine learning was
15:32
like, really a thing people talked about and then it would
15:34
be best if we did. I did do a lot of, like machine
15:37
learning, implications and projects with, like Financial
15:41
Services, customers, or like Insurance customers or like
15:44
biotechs that were doing, you know, like machine learning work
15:48
and using AWS compute Services is also before higher level
15:53
services, like Sage maker and other things.
15:55
Came out these By figuring out how to use AWS storage compute
16:01
services to other false potential for like machine
16:04
learning workloads basically. So you have to do a lot of that
16:07
enable you ask ml specialist there and then moved into Alexa
16:13
still is like a machine learning specialist Solutions architect.
16:16
So it does can get like pretty specialized in big companies.
16:19
I would say there's a lot of like X role for machine
16:22
learning. I would say to some of these
16:24
companies they have moved into. Alexa is like machine learning.
16:27
Specialist Solutions architect which is a lot of words.
16:30
But yeah. Basically like helping internal.
16:33
Alexa customers is the HRS. Are like automated speech
16:39
recognition team or like our natural language, understanding
16:43
team, like, all these different speech and language, modeling
16:46
teams would have a problems using, you know, like breaking
16:50
AWS and different ways, like, figuring out how to use
16:53
infrastructure and best ways for there like machine learning
16:56
workloads. So, Kind of helped internal
16:59
Alexa teams with that and then moved into the product roll.
17:03
Cuz ya feel like as a Solutions architect was great, enjoyed it
17:07
quite a bit, I had a lot of strong product opinions.
17:09
I want to have more people are kind of Upstream have more
17:11
ownership. We're like the engineering and
17:14
product development process and yeah I guess have more input
17:17
into what we're actually building and prioritizing versus
17:21
I feeling like I was implementing something.
17:24
Someone else, bill, I guess that makes sense.
17:26
Yeah, I I love 20 in a little bit more on transition from Deep
17:32
technical implementer to, to like kind of a p.m. role.
17:35
And one of things that I also love to touch on to get your
17:37
thoughts on here is Yeah, you PMS have been around for a
17:41
while. I mean PMS have been software
17:43
orgs for as old as probably software has been around.
17:46
And I do wonder, how does how is an MLP Em Right on a IPM
17:52
different and what does that what different skill set does
17:55
that require with the responsibilities that require?
17:58
If there are folks that are listening that are interested in
18:00
getting into that more ml style p.m. anger, may be transitioning
18:03
from regular p.m. rolls into ml PMS. What should they be
18:06
thinking about and preparing? Yeah, this is a phenomenal
18:10
question that I was unprepared for but yeah, so I would say
18:15
there's definitely a spectrum of p.m. roles you know?
18:18
And there's different archetypes I would say, you know, everyone
18:21
thinks about the classic like Steve Jobs, type Visionary p.m.
18:24
I guess where you know that these think Big Ideas throw it
18:30
over the wall for someone to get it done.
18:32
But I say I like the total other side of the spectrum, there's
18:36
like a lot of technical Manager roles and the ml type ml
18:42
adjacent p.m. roles or kind of like I call them like platform
18:46
p.m. roles where you're working on.
18:48
Oftentimes, I like big companies, been increasingly
18:51
smaller, and smaller companies are doing a lot of
18:53
infrastructure work and realizing like data is actually
18:56
a product even internally. So I finally, like increasing
19:00
there's a lot of platform p.m. type roles where even a smaller
19:03
companies, you're owning kind of, like internal, whether it's
19:06
like data products, maybe your Company does a lot of it work
19:11
and it's like Financial Services or something else.
19:14
And you have a lot of like specialized high quality or like
19:18
high-value data that your vending to other internal teams
19:21
or externally. And there's like PMS that make
19:24
sure, you know, the data, they're producing provides value
19:28
to your customers. And what is like, the roadmap
19:30
for the data products that you provide or there's traditional,
19:34
I guess, platform PM's who kind of own and Like the feature set
19:40
and roadmap for internal platforms.
19:43
And you've seen plenty of those, like, Michelangelo's is Uber,
19:47
whatever else there is. So you're owning basically like
19:50
internally. What are like the infrastructure
19:53
capabilities of that platform and your customers.
19:57
Usually like, internal, and maybe eventually external
20:01
customers that you provide like capabilities to other internal
20:05
teams to help other, internal teams, move fast.
20:08
After basically so I guess like core competencies, therefore
20:13
p.m. as you probably have to be much more technically inclined
20:17
than other types of p.m. role probably do main expertise to
20:22
you. I would say a lot of these rules
20:23
would be difficult to come into with out some ml cloud in a
20:31
domain expertise, I guess. I'll say the internal
20:33
stakeholder alignment, communication, usually companies
20:36
expect the PM's to be the best communicator.
20:38
Room. So that's why there's a lot of
20:40
that unsexy kind of internal Communications selling your
20:44
ideas, internally pitching for resources, that kind of thing
20:48
too, as well, that's an interesting one.
20:51
How you should be the best Communicator in the room.
20:54
So you have to be able to explain things that are a little
20:57
bit more technical in not just say dumb down but in a way that
21:03
everyone hopefully can understand quickly and easily.
21:06
So one thing that That you said there that I want to dive into a
21:11
little bit more, is you mentioned how data is a product?
21:16
And potentially you have PMS for data products and do just mean,
21:20
like, the way that you're giving other teams, the data is that
21:24
what I understood or is it like the data itself and is this,
21:30
because this is something that I know gets talked about a lot in
21:34
the, the data mesh Paradigm. And I Like, I've wrapped my head
21:39
around it, but I just want to make sure that when you say
21:42
that, are you saying it in that sense?
21:43
Or you saying in a different sense?
21:46
Yeah, I would say, all of the above, I guess, especially like
21:51
some of these large companies like Amazon or Fang, I guess,
21:56
companies have tons and tons of data needs, you know, pipeline,
22:01
hell everywhere. And there's a more data than
22:05
anyone knows what to do with and just like the The ability to
22:10
make sense of that data. And like, then that data out to
22:15
internal and external customers is a product role in itself.
22:19
Like what did a reading prioritize, vending how to weave
22:23
in that data out of the customer need to receive that data in
22:25
terms of like frequency velocity, like, call patterns.
22:31
And so I would say, yeah, I like the vending of data itself is
22:35
one thing, but also like offering data products directly.
22:38
Is also a thing that happens. A lot of companies now have like
22:44
data that they collect internally and massage, and Rich
22:48
somehow. And then, turn around and sell
22:49
that data as a subscription or as did product service, but also
22:54
in Rich data and provide data to internal customers as well as
22:58
like the entire purpose of that team.
23:01
So the others PM's and that's supposed to my name's Oliver
23:08
Shaw. Akka, and I'm head of machine,
23:09
learning engineering, origami energy here in the UK.
23:13
And I'm bothering you today to recommend, you subscribe to the
23:16
ml Ops committee podcast to keep up-to-date with a lot of
23:20
brilliant, goings-on, within the community, and get a glimpse of
23:24
what. Some of the greatest Minds
23:26
across the amount of space are up to one of things.
23:32
I still wanted to go back to is because when you're describing
23:35
your career trajectory you described going from More like
23:39
smaller kind of startup style but you mentioned working at a
23:44
friend's startup and I believe it was just like a smaller
23:46
organization and then obviously, moving into something like an
23:48
Amazon or an Alexa, which is kind of people, right?
23:52
And, and yet, having similar roles may be being both the
23:55
solutions architect in smaller organization, and also still
23:59
solution Solutions, architecting and Amazon style organization.
24:02
You tell us a little bit about like how that have that's
24:04
different and you know what, the pros and cons of being in water
24:08
The other and just what that's like.
24:12
It's good point. I definitely feel like everyone
24:15
should do a start-up once like in their twenties.
24:17
Maybe I feel like it's a great opportunity to kind of like you
24:22
get probably like a lot of responsibility, maybe even more
24:25
so than you should given maybe your age or experience or
24:29
whatever, melt might be a that credentials and especially like
24:33
fast growing startups, a lot of opportunity to advance quickly
24:37
and Clearly there's less like rigor or like career ladders.
24:43
Usually no career ladders whatsoever.
24:45
No single. There's a lot of ambiguity and
24:47
like with amputee I think comes a lot of opportunities so a lot
24:51
of the opportunity for advancement very quickly you
24:54
know if you do well but then it kind of depends on your style
24:57
and preferences. A lot of people also don't love
25:00
ambiguity and would love to see you know here's my career ladder
25:03
and here's my path and every three years I do well and get
25:07
promoted and get that like We percent less than inflation pay
25:10
back every year and if that's what you're into, you.
25:14
That's fine. I think that's great for the
25:16
people like me that have a family to support and I love
25:20
would want stability. So, I think depends on you just
25:23
phase of life here at and like, what you need in the moment and
25:27
you're looking for long term as well, but it's a like working
25:31
environment wise can be similar like, almost every company says,
25:36
or at least wants to move fast move quickly.
25:38
And Start up like so thank like culture-wise.
25:42
Different depends quite a bit of big companies like whose team
25:46
and like leadership and like, what is the org?
25:49
You can have as, you know, he'll and Amazon like a vastly
25:52
different different experience in one team Amazon versus
25:56
another. So, I think it's tough to
25:57
generalize to in some ways. When they said that I was really
26:03
cool point was just like with ambiguity comes opportunity.
26:06
I think that's a really, that's a really nice way to put that.
26:09
I feel like that's something that's especially true and
26:12
broadly true in general. But but especially in machine,
26:15
learning writing in such a fast moving space that were in where
26:19
things are constantly changing from Models.
26:21
First from technology, from Team organization, there's to all
26:25
these aspects. I'd love to huge could like,
26:27
tell us little bit more about, what does that mean?
26:30
We as we talked about like non-traditional career paths,
26:33
right? This this idea of a constantly
26:35
evolving landscape that is machine learning new.
26:38
What does that mean for someone who's maybe entering the space?
26:42
And, you know, maybe doesn't really want to be just pure
26:46
implementation person or is that and even p.m. is the right role
26:49
for me. Like how does that what can they
26:51
think or expect given this ambiguity?
26:53
Not just just kind of being beauty of the field in general,
26:56
Maybe Yeah, that's good point. And I would say things are
27:00
changing. Well, things have always been,
27:02
changing rapidly, but especially, you know, today in
27:06
2022, I think ml isn't just like a free lunch anymore.
27:11
You can't just like ml wash a project and get it funded
27:15
immediately anymore. So I think like leader and
27:19
companies are becoming a lot more Discerning about like the
27:22
projects they invested in and where where they want to make
27:26
targeted. Hi Ry.
27:28
Investments whereas before it was like, you know, we're just
27:31
going to hire in front projects just because we want to get good
27:35
talent in the door and I think the story has changed a lot last
27:39
few months but I think from a career perspective you know
27:42
things are not the best timing in the world right now.
27:46
I would say I don't think the fundamentals have changed a ton
27:50
though. You know keep learning and keep
27:52
if you're probably like listening to this podcast in
27:55
your own free time, you're probably the type of Person
27:58
that's like on the right track to being a costly like learning
28:01
involving state, but I would say definitely keep abreast of like
28:05
major changes. I would also say don't get like
28:10
too caught up in a lot of the distractions.
28:13
Like I'd like to take a stance. Maybe this isn't a great way to
28:16
do it but wait for things to mature a little bit before,
28:19
like, investing a ton of time or effort into them.
28:21
That's wise. I'm not I used to get caught up
28:25
in this lot, like, being Technical and like being young.
28:29
You want to dive deep into the latest sexy stuff, but give it
28:32
at least a few months or like years to bake before investing a
28:36
lot of your enzyme into it because, you know, Technologies
28:40
do is allow or not, get it option.
28:43
And there's a lot of stuff out there that is like, super,
28:46
interesting and compelling. And I think you should do like a
28:50
lot of that bleeding edge, like reading and kitty up to date.
28:53
But when you Like big Investments of your time.
28:56
Learning new technology that's where I think like it's
28:59
important to kind of vet it a little bit first and make sure
29:02
it makes sense for what your goals are I guess.
29:06
And so when you look at your day and maybe we can go over this as
29:13
a product manager and then also we can look back and when you
29:16
were a Solutions architect and all of these other roles that
29:21
touch machine learning, but aren't necessarily a
29:23
machine-learning engineer or software engineer, a data
29:26
scientist that type thing. But as you go through your day,
29:31
what are the main things that you're doing?
29:32
What are the different types of activities?
29:35
You're going through, is it? Gathering specs, take up, you
29:38
know, 50% of your work week and then pitching new ideas takes
29:43
up, 10%. And actually having to write
29:45
down what are they called? Prds those take up basically
29:49
110% of your day. What what does that actually
29:53
look like and what are some things that you do?
29:57
Yeah, I would say a good project requirements documents, I guess
30:01
is what you're referring to parties.
30:03
Amazon. You've probably also or maybe
30:05
you haven't heard of Press release FAQ process where we
30:10
kind of work backwards from the customer.
30:13
And right, kind of like press release or like launch
30:16
announcement for products that we want to launch and like, work
30:20
backwards to deliver what that is and actually, right after
30:24
this basically, like, Reviewing one of those with our leadership
30:29
team and that's kind of like at Amazon at least how we make
30:33
pitches for. Like investing in new ideas, is
30:36
you write like this? Document is a press release and
30:39
all these free to ask questions about why it's good business why
30:43
we should be investing in it. And I think like prepping for
30:46
that is not just obviously for editing document.
30:49
But also stakeholder alignment is something that I think I
30:53
spend a lot of time on is like Communicating with partner teams
30:58
communicating with other teams that might have like even
31:01
competing priorities or dependencies or against like
31:06
depending on your interest may be sexy, you're on sexy part of
31:09
the job. I like how do you align all of
31:11
the stakeholders to kind of like meet the objectives for your
31:16
business but also your customers as well.
31:18
So it's been a lot of time on that.
31:20
One thing I find it's like you kind of have to be intentional
31:24
about do is Talking to customers, which sounds weird or
31:28
obvious. But sometimes it's difficulty,
31:31
you know, as p.m. or even any role.
31:33
It's easy to get caught up in like, internal discussions or
31:37
like technology discussions or just like internal debates or
31:41
fighting fires or whatever and, like, slacking email.
31:43
But I find it's like really have to be intentional about talking
31:48
to customers and getting out there and talking to your users.
31:50
Whether that's for me in the Creator or Hands-On and work
31:55
with social Media influencers. So for me, it means right now,
31:59
reaching out to influencers and with, you know, mock-ups that I
32:03
have our ideas and that kind of vetting and validating those, or
32:06
kind of understanding, what their workflows are, what their
32:09
pain points are. And sometimes companies have
32:12
like Gatekeepers and therefore good reasons, like maybe to talk
32:16
to a customer, you have to go through the account manager, you
32:20
have to go through user research person in like, sometimes it's
32:25
difficult to get. Access to customers.
32:27
And I think it's important, though, to make it very
32:31
intentional on the part of your schedule.
32:33
Almost see, if I'm not talking to customers at least every
32:36
week. I'm probably not doing what I
32:37
should be doing, but focusing to internally navel-gazing is
32:42
product manager. If I'm, that's easy to do and
32:44
you have to be like really intentional to with the so it
32:49
comes full circle and want to talk about and I don't like no
32:53
like a huge map but like Developer evangelist developer
32:57
Advocate is another role that I think is seared.
33:00
Wrong one, it's like almost the holy grail for developers like
33:03
you get to write code sometimes when you feel like it but maybe
33:08
you don't have a specific deadline.
33:10
Well, of course you probably have like, content you put out
33:12
or like conferences. You're presenting a panic.
33:14
If you're young and want to travel a lot, you conference
33:17
circuit or own and manage community, and just like
33:19
Community Management aspects that contribute to open source
33:23
of now do a lot of public. And the public, I think there is
33:27
kind of like a social-media influencer for ML.
33:29
Type roll out there. Yes, exactly.
33:32
And the thing about the machine learning space and probably not
33:37
so much, the machine learning like deep model, building space,
33:40
but ml Ops and like machine learning engineering space.
33:45
I know so many companies that are dying for a good developer
33:50
Advocate, right? Like Devereaux is very hard to
33:54
come by in this sphere, because I think a lot of people have the
33:59
misconception that if you turn into a developer Advocate, or
34:04
you start doing Devereaux, and some way you're going to start
34:08
getting further and further away from the tech and I try and tell
34:12
people it couldn't be any further from the truth because
34:15
when you start doing that and you get to just play around
34:19
with, whatever floats your boat. And so, if you are interested in
34:24
a certain piece of tech, You can go build something with that,
34:28
and a lot of the companies don't necessarily even want to see,
34:34
you always putting out content with their tool because then
34:39
it's just monotonous and boring. Like, as long as you're putting
34:42
out content, that is educational or entertaining, or valuable.
34:47
In some way, these companies are going to be happy with you doing
34:52
that. So for all those out there that
34:54
are wondering, if they should go into Devereaux, I highly
34:59
encourage it right now, especially in the ml Ops or ml
35:03
engineering sphere. I know there's, there's not
35:06
enough of them out there, and I think that really moves the
35:10
space forward to the more people that you see talking about it.
35:14
The more people that you see, influencing on that realm, it
35:19
helps others understand and it's just getting more education out
35:23
there. So totally feel you on that.
35:25
I wanted to jump back though. One thing that you talked about
35:29
was stakeholder alignment, and stakeholder alignment feels very
35:35
much like playing The Game of, Thrones it away.
35:41
And so, what have you learned while doing that?
35:45
And how do you do it? What are some tricks that you
35:47
have up your sleeve? Yeah, stakeholder alignment is
35:52
can be a dirty phrase. I think, actually to go into
35:55
another role. You might have heard of like
35:58
technical program manager and like program manager type roles.
36:02
Like they're also kind of like, a p.m. adjacent whole.
36:05
I would say, I actually have no idea what that does.
36:08
What does that even what does that?
36:09
Yeah, it was like a Google others have similar roles.
36:15
Technical program manager is basically a lot of tech big Tech
36:19
organizations have big hairy like technical problems or
36:24
migrations or like massive technical efforts that are like
36:28
wide and far reaching through the business.
36:30
So they're probably massive like cross-functional efforts, I
36:34
don't know. Amazon had one a couple years
36:36
ago, I was like we need to move off of Oracle everywhere and it
36:40
was like deeply ingrained in almost every like service at
36:43
Amazon or like big GDP ours like another great canonical example,
36:48
of who is the GDP are person at your company.
36:51
That's like bringing the whole company forward towards this
36:53
objective and they're like lots of those type of projects
36:59
programs and there's like specific role on a lot of
37:04
companies. It's like program manager and
37:06
there's technical program manager is because for a lot of
37:09
these projects you would be pretty technical as well.
37:12
Their entire job is basically that like stakeholder alignment.
37:15
Communicating status updates, resolving dependencies resolving
37:20
escalating issues before they arise and like generally I don't
37:23
know if I can swear but like getting stuff done and I think
37:27
that is an also at there's a kind of a like a little life
37:31
hack or career hack here that like, I've seen a lot of paths
37:35
from that role into software, engineering type roles if
37:39
they've seen that a lot Amazon. Like, other companies is like, I
37:42
think a really great program, Manager is force multiplier and
37:47
makes a massive difference in a project like succeeding or
37:50
failing. And I think people notice that
37:52
very quickly like within a couple weeks of having a program
37:57
like a good program manager. I think it's very clear if
38:00
they're good or not good and I've seen a lot of people that
38:03
have there's almost like kind of a path from technical program
38:08
manager to software development manager, for better for worse.
38:12
I guess, Amazon says, kind of like Feeder into a lot of other
38:16
roles in a way as well really stepping stone I guess.
38:20
So I've seen that happen quite a bit as well.
38:23
Sophia, that's a good way. Like side step into an
38:26
engineering manager role but maybe not as indirect software
38:29
engineer if that makes sense. But actually, to answer the
38:31
question you asked like tips for stakeholder alignment, I would
38:34
say one people don't talk about enough is just like earning
38:38
trust. Like you're not going to force
38:40
alignment and, you know, I think you have to communicate both
38:44
like Good news and bad news you know like authentic way and it
38:49
burns a lot of trust and build relationships and Partnerships
38:51
that doesn't mean just like having meetings and talking
38:55
about dates and working backwards from those dates to
38:59
get a meeting like an objective accomplished.
39:01
But you know, just talking people as people and building
39:04
relationships and yeah, just generally earning trust in ways
39:07
outside of the normal, I guess. So that's just like pulling
39:10
peoples out or developing like, One-on-one relationships are
39:14
like making an effort to travel and be there in person or just
39:19
getting in front of people. I think it's important just like
39:21
building a human connection and not being a robot.
39:24
And I would say, also, I guess use the tools you have at your
39:28
disposal. A lot of people try to reinvent
39:31
the wheel get fancy, but I think great way to get alignment is
39:35
just like over communicate. So just something dumb like
39:39
sending a weekly status email, sometimes it's like something.
39:43
Which offer don't want to do. Like you probably your company
39:46
or organization probably has a ton of tools gets dropped down
39:50
whether it's I don't know, Wiki's or tickets or email or
39:53
whatever it is like, you don't have to get fancies really just
39:56
about communicating and even like ordering on
39:58
over-communicating. I think while people might not
40:01
respond directly just like seeing stata and letting people
40:04
know that you thought things through I think is like a big
40:08
way to give people comfort that you know what the hell you're
40:10
doing. I guess I have a little bit of a
40:13
A different question. So the researchers can fact
40:17
check me here. But if it's, you might be one of
40:20
the first guests, if not the first guest who because you had
40:24
time in the Air Force. So, military personnel, I don't
40:28
think regularly are esos animales podcast.
40:32
I mean, that's, that's cool. And I would love to, you know,
40:37
how is that if at all influence kind of some of the things
40:42
you've done in ml? In the course of your career.
40:44
I mean, any skills is there s only thing about non-traditional
40:48
career paths is there is there a path forward where the first
40:51
step is to spend some time and kind of a military division and
40:54
then go to ml? Is that something that we should
40:56
be recommending the people or is this, you know?
40:59
Did you feel like it was a boost or totally orthogonal and not
41:02
necessarily relevant or play out?
41:05
It's totally worth Ogle. Not super relevant but there's a
41:08
lot of soft skills and like leadership management skills, I
41:12
guess. Yes, you get as, you know like a
41:14
young officer in the military that like similar to some tech
41:19
rolls or like in a start-up, which is weird comparing
41:22
government to start up. But like similar to that
41:25
graduating college and like having much more responsibility
41:29
than you probably should, or would normally be afforded.
41:32
I think makes just kind of like put your feet to the fire I
41:36
guess and likes you figure things out quickly.
41:38
Just so you don't like embarrass yourself.
41:41
I guess. Like public humiliation is good.
41:43
Motivator. So like a lot of times as a
41:46
young lieutenant in the military, you're like
41:48
responsible for overseeing people, they're like twice your
41:51
age and I think that is a good again, motivator to like really
41:57
put yourself be better. And yeah there's a lot of ml
42:00
work happening in the intelligence community and
42:03
Military. I think that's a trend, you
42:05
know, like Air Force and other services are starting to like
42:09
work much more closely with contractors and even do a lot
42:12
more internal. Cyber and higher lot of
42:15
developers. I would say it's a good way to
42:16
get like a free college. Education is a good way to get
42:20
like free skills and if you're interested in oxidate, like
42:23
cyber security, type stuff, or even they're not, there's an ml
42:27
specialty that it is a good way to get like, free Technical
42:31
Training, free education, a lot of other benefits.
42:35
I would say in like a compressed amount of time, like for your
42:39
commitment or whatever, your commitment is and then like
42:42
Leverage that into Other opportunities I guess,
42:45
especially in the kind of constrained job market or other
42:48
opportunities might not be obvious, I guess.
42:51
Yeah, you get to travel a little bit.
42:53
I know that right next to my house in Germany.
42:58
There is a big us. I don't know if it's army base
43:02
its or airfare sits in Wiesbaden.
43:05
You may know if you ever went through, there might be a un
43:10
base. I have no clue because I'm not
43:12
to, in the know, in that type of thing.
43:15
But it is interesting to see, there's a ton of people that are
43:20
all around these bottom that are Americans and I found out some
43:24
pretty wild that. Nobody tells you, you've been
43:28
floating around the ml Ops fear for a while now in different
43:32
roles not necessarily doing the exact ml Labs or engineering
43:38
role but you see other ways that you can get into the space.
43:43
You've seen others that are in this space for you, what are
43:47
ways that you feel like are good ways in but What else do you
43:51
think like how would you recommend?
43:54
People get into ml Ops right now.
43:57
Yeah, that's fantastic question. I would say kind of like
44:03
Leverage what you have today and I just have a good introspective
44:10
understanding of what your strengths and the superpowers.
44:13
Are they a lot of people focus on what their weaknesses are
44:17
like I, you know, have Gap here. So I'm going to go waste a bunch
44:21
of time doing a certification to fill that Gap.
44:23
But like I would say, Most companies tend to reward
44:28
specialization and going deep in a subject area.
44:33
So it's hard to generalize advice but you know what?
44:35
Are you good at? What do you love doing?
44:37
Unlike double triple down on that?
44:39
And well, that's taking like technical skills and like
44:44
applying them to ml or ml Ops or data whatever it is.
44:50
I would say, I always think I've been rewarded by specializing,
44:55
so I'd say that helps a lot in terms of specifics to.
44:59
I think that's like, cliche to say maybe like but networking, I
45:04
think helps, I don't think there's a job.
45:05
I've gotten that, I didn't know someone at wasn't from my
45:09
network. So build those connections,
45:11
where you're at, which those for your next thing.
45:14
And also, just get involved where you're at to like a lot of
45:17
people focus on their job opportunities and what's in
45:21
their job description. But That's it.
45:24
Like most of the skills I've gained are from, you know, like
45:29
experimenting or playing around or like side projects that
45:32
weren't necessarily like part of my immediate job description but
45:36
probably like adjacent to what I was doing.
45:39
I found with improve of what I was doing with other
45:43
Technologies to try to like, try to like see what's out there and
45:46
see what you can apply to what you're doing in novel, but like
45:51
relevant and valuable ways. I guess we're just you go
45:53
projects that can apply Technologies.
45:56
You want to work on? Yeah.
45:57
So kind of like a mix of working in things where you're already
46:02
doing that. I'm really seeking out certain
46:04
projects in the space you want to be.
46:06
And I guess you said something crucial there which I completely
46:12
agree with about the networking piece and some people, I know.
46:16
But they're thinking like, yeah, it's easy to just say going
46:19
network. But then whenever I go and give
46:21
my business card out at events, Functions.
46:23
I don't make any friends. So one thing you can do, if you
46:27
would like to meet others that are in the ml up space or doing
46:31
machine learning stuff, we've got meetups that are happening
46:34
all over the world, and that is a really low barrier, to entry
46:39
way to just go and hang out with people that are doing this
46:42
stuff. And I love that.
46:44
We've got them in so many different cities, and I think
46:47
that's a easy way. You go to one of these, it
46:50
doesn't matter, even if you're alone and you just, you me, Eat
46:52
people. You can't help but me people.
46:55
So that's such a true statement like Network and that will help
47:01
you get your next job. Great, masterful segue, right
47:05
there. I had to shamelessly.
47:08
Plug that in. What did you expect?
47:10
I couldn't help it. Anyway, here we go.
47:13
That's all we've got, dude. This was awesome.
47:15
Matt, I really appreciate you coming on here and talking to us
47:18
about all of these different ways that you've seen machine
47:22
learning. Being done from other angles I
47:25
guess from different perspectives we could say.
47:28
So thanks for doing this. Awesome.
47:31
Yeah, thank you. I appreciate the thought that
47:33
appreciate you coming on and also like interested, only, guys
47:36
do for the community. Yeah, it's you guys out there.
47:40
Public sphere is Skyler. I lead machine learning and
47:44
health rhythms if you want to stay on top of everything
47:48
happening in ml. Ops, subscribe to this podcast
47:51
now. Now, now, now now now Now.