1
00:00:00,000 --> 00:00:03,389
The rising star of MLOps.

2
00:00:03,770 --> 00:00:04,760
Is that how you feel about it?

3
00:00:06,010 --> 00:00:09,509
What do people get wrong about
MLOps the most, do you think?

4
00:00:09,580 --> 00:00:15,370
Many people feel that they need to
have all these shiny tools to do MLOps.

5
00:00:15,889 --> 00:00:17,350
And I don't think it's true.

6
00:00:17,500 --> 00:00:20,430
If you could go back the 10 years
that you mentioned and you were to

7
00:00:20,430 --> 00:00:23,980
choose whether to go into the same
career path, would you do it again?

8
00:00:24,310 --> 00:00:26,150
Yeah, I would definitely do it again.

9
00:00:31,072 --> 00:00:34,082
you've got a very nice
backdrop there with Iron Man

10
00:00:34,662 --> 00:00:35,582
It's all Lego.

11
00:00:35,582 --> 00:00:36,992
It's also flowers.

12
00:00:37,092 --> 00:00:37,992
Also Lego.

13
00:00:38,402 --> 00:00:44,612
Because I have two cats and if you have
any normal flowers or even dried flowers,

14
00:00:44,642 --> 00:00:48,892
they will try to eat them and they're
not interested in Lego flowers, luckily.

15
00:00:48,892 --> 00:00:52,032
So that's why only Lego
flowers at our house.

16
00:00:52,852 --> 00:00:56,532
And, I'm fan of Marvel and
that's why we also have company

17
00:00:56,532 --> 00:00:57,852
called Marvelous Mellops.

18
00:00:58,292 --> 00:01:02,732
Because Marvel and the Mellops,
these are two things I really like.

19
00:01:02,782 --> 00:01:03,892
That all makes sense.

20
00:01:04,402 --> 00:01:04,982
Yes.

21
00:01:04,992 --> 00:01:07,482
So I have a lot of Marvel stuff here.

22
00:01:07,492 --> 00:01:12,222
you can just see two, but I know
there are a lot in the room.

23
00:01:12,222 --> 00:01:16,722
Okay, Maria, so you've been described
to me by multiple people at this

24
00:01:16,722 --> 00:01:20,952
stage by the rising star of MLOps.

25
00:01:20,962 --> 00:01:22,392
Is that how you feel about it?

26
00:01:23,572 --> 00:01:28,272
I don't know in terms of following, maybe
on LinkedIn, it is definitely growing.

27
00:01:28,382 --> 00:01:32,647
a bit more than a year ago, I had,
A bit more than 1000 followers.

28
00:01:32,657 --> 00:01:34,767
And now I believe 46,000.

29
00:01:35,217 --> 00:01:38,027
And when this episode
comes out, I hope 50,000.

30
00:01:38,217 --> 00:01:39,617
That is my goal.

31
00:01:40,137 --> 00:01:41,467
and higher, obviously.

32
00:01:41,727 --> 00:01:46,817
I just felt that there is not so much
people writing about MLOps, talking about

33
00:01:46,907 --> 00:01:52,677
MLOps, and I've been doing it for many
years, so I felt like that's the right

34
00:01:52,717 --> 00:01:54,617
moment for me to start talking about it.

35
00:01:55,267 --> 00:01:58,137
I think I saw somewhere
that it's over a decade now.

36
00:01:58,137 --> 00:02:02,717
And You describe yourself as a
dinosaur of AI, which I don't think

37
00:02:02,717 --> 00:02:05,537
is entirely accurate, but why MLOps?

38
00:02:05,577 --> 00:02:09,137
Why was it interesting enough for
you to really double down on that?

39
00:02:10,122 --> 00:02:16,162
So my career started more than 10 years
ago as a data analyst, because there was

40
00:02:16,162 --> 00:02:19,272
not so much in data, available back then.

41
00:02:19,312 --> 00:02:22,262
data scientists didn't really
exist in the Netherlands at least.

42
00:02:22,662 --> 00:02:27,622
back then, and I started economics,
econometrics, and I never really liked

43
00:02:27,652 --> 00:02:32,372
economics per se, but I liked the
data part in it and we were using R.

44
00:02:33,052 --> 00:02:34,342
to do some modeling.

45
00:02:34,772 --> 00:02:40,552
And that's the part that I liked and data
analytics felt like the closest to it.

46
00:02:40,562 --> 00:02:41,992
So that's how I got started.

47
00:02:42,572 --> 00:02:46,542
And, even though I formerly was called
the data analyst, we were already

48
00:02:46,542 --> 00:02:53,227
building some churn model acquisition
models for telco in R and then I

49
00:02:53,227 --> 00:02:55,517
transitioned more into data science.

50
00:02:56,357 --> 00:03:00,637
So I started learning Python and
actually building first applications.

51
00:03:01,247 --> 00:03:03,857
And there was no one who could
put those in production, actually

52
00:03:04,277 --> 00:03:08,117
integrated with the real applications
that the customer facing.

53
00:03:08,697 --> 00:03:11,537
And, that's how it all
got started, actually.

54
00:03:11,622 --> 00:03:14,772
I teamed up with some software
engineers in the company that

55
00:03:14,902 --> 00:03:17,742
were helping me to implement that.

56
00:03:17,842 --> 00:03:23,032
and then I felt like that's a good moment
to actually start introducing it to a

57
00:03:23,042 --> 00:03:28,712
broader data science community to teach
them, some best DevOps practices, like

58
00:03:28,712 --> 00:03:34,122
using version control, having CI/CD
pipelines, registration in place, and

59
00:03:34,122 --> 00:03:36,402
that's what I started promoting back then.

60
00:03:36,422 --> 00:03:40,732
I felt that it was needed because it
took so much time for data scientists

61
00:03:40,732 --> 00:03:46,062
to have something that could actually
have impact, and not just some analysis.

62
00:03:46,132 --> 00:03:50,152
yeah, actually introducing CICD
pipelines, as I mentioned, and version

63
00:03:50,152 --> 00:03:54,052
control was already a big step,
because no one was using it back then.

64
00:03:54,102 --> 00:03:57,902
There were some shared drives where
people would put some shared code maybe

65
00:03:57,902 --> 00:04:04,242
and Jupyter notebooks and that those
were popular still of course very popular

66
00:04:04,952 --> 00:04:11,432
but I feel like now the community is
really becoming more professional and It

67
00:04:11,432 --> 00:04:13,752
will be even more needed in the future.

68
00:04:15,102 --> 00:04:20,847
So that's how it all started for me and
Since then, I've built multiple MLOps

69
00:04:20,887 --> 00:04:23,197
frameworks with different set of tools.

70
00:04:23,737 --> 00:04:28,397
It started all with some proprietary
software, then we moved to all

71
00:04:28,407 --> 00:04:32,747
open source with Kubernetes, and
then to AWS native infrastructure,

72
00:04:32,747 --> 00:04:34,977
and now I use Databricks a lot.

73
00:04:34,997 --> 00:04:37,837
So for the last three years,
I'm building it on Databricks.

74
00:04:38,837 --> 00:04:43,377
So would you say that this is what
you would describe as a hockey stick

75
00:04:43,377 --> 00:04:51,007
moment for, ML, this marriage of, DevOps
practices added to the previous stack of R

76
00:04:51,007 --> 00:04:53,377
and Jupyter notebooks and stuff like that?

77
00:04:53,797 --> 00:05:00,907
Is that really what enabled this growth
of MLOps and improvement in scale of how

78
00:05:00,907 --> 00:05:02,817
people deploy this machine learning stuff?

79
00:05:02,902 --> 00:05:05,842
Yeah, I think that was
indeed such a moment.

80
00:05:05,842 --> 00:05:11,572
And what I personally saw in my
career that around 2015 or so

81
00:05:11,572 --> 00:05:15,202
that a lot of companies started
investing into data science.

82
00:05:15,222 --> 00:05:17,132
Also, the company where
I worked back then.

83
00:05:17,722 --> 00:05:20,382
And after this investment was done.

84
00:05:20,382 --> 00:05:22,732
There was very low return
on that investment.

85
00:05:22,912 --> 00:05:27,692
And that's because of this,
separation between IT and, data

86
00:05:27,692 --> 00:05:31,492
science departments, which were
originally part of business teams.

87
00:05:31,942 --> 00:05:35,302
So to actually implement something,
you would have to create requests

88
00:05:35,302 --> 00:05:39,252
to IT teams, and it can take some
time before it gets implemented.

89
00:05:39,602 --> 00:05:43,547
So actually giving freedom and,
permissions to data science

90
00:05:43,547 --> 00:05:45,617
teams to implementing themselves.

91
00:05:46,037 --> 00:05:48,107
That's what makes a huge difference.

92
00:05:48,737 --> 00:05:51,607
And some companies are getting there, but.

93
00:05:52,127 --> 00:05:56,797
I feel like there is still a really
large amount of companies that

94
00:05:57,467 --> 00:06:00,197
don't use these best practices yet.

95
00:06:01,207 --> 00:06:05,037
I think the question that a lot of people
have is, and I'm asking that because I

96
00:06:05,077 --> 00:06:11,207
went on to marvelous MLOps.substack.com,
a little plug here, and, I found an

97
00:06:11,207 --> 00:06:16,877
interesting article, the ultimate must
haves and nice haves for MLOps and LLMOps.

98
00:06:17,457 --> 00:06:22,707
And maybe we talk a little bit more
about what MLOps versus LLMOps is, but,

99
00:06:22,757 --> 00:06:26,337
when I was looking at the illustration
at the top of that looks very similar

100
00:06:26,337 --> 00:06:31,987
to any modern deployment stack minus,
obviously the things that are a little

101
00:06:32,017 --> 00:06:36,107
bit more, specific, like the vector
databases, although, you could use

102
00:06:36,107 --> 00:06:37,457
that for other purposes as well.

103
00:06:38,397 --> 00:06:46,557
So why don't we just call it DevOps or,
why do we need a specific term for MLOps?

104
00:06:46,652 --> 00:06:50,302
I think MLOps is, an extension of DevOps.

105
00:06:50,302 --> 00:06:52,082
It's a bit more complicated than that.

106
00:06:52,502 --> 00:06:58,062
For DevOps, you don't have one
thing that is you have in MLOps.

107
00:06:58,082 --> 00:07:04,102
it's a model that can change when
the data changes, so even if your

108
00:07:04,102 --> 00:07:07,472
model hasn't changed the artifact
itself But the data distribution has

109
00:07:07,472 --> 00:07:10,252
changed and you get bad predictions.

110
00:07:10,572 --> 00:07:13,671
Code didn't change nothing changed,
but you get bad predictions that would

111
00:07:13,671 --> 00:07:19,697
never ever happen in standard DevOps,
unless maybe data schema changes, but

112
00:07:19,777 --> 00:07:24,707
if nothing changes, but just some data
distribution changes, things can fail.

113
00:07:25,357 --> 00:07:30,747
And that's a huge difference and
that's a lot of complexity to

114
00:07:30,747 --> 00:07:32,467
how you need to manage things.

115
00:07:33,267 --> 00:07:36,947
and that's, I think what a lot of
people that ask these questions

116
00:07:36,947 --> 00:07:43,332
usually don't realize that you need
to have a lot of things in place to

117
00:07:43,812 --> 00:07:46,142
make sure you can track it, properly.

118
00:07:46,702 --> 00:07:51,272
And you need to know, for any
deployment that you have, what code

119
00:07:51,272 --> 00:07:55,172
that deployment, came from, what
data that deployment came from, and

120
00:07:55,172 --> 00:07:57,162
also what model artifact was created.

121
00:07:57,942 --> 00:08:03,692
Need to have the track of all these
aspects in one place and be able to roll

122
00:08:03,712 --> 00:08:06,282
back if needed or modify things if needed.

123
00:08:06,362 --> 00:08:09,512
that's a much different process
than what you have for DevOps.

124
00:08:10,512 --> 00:08:15,022
I also noticed that you've got
on that list LLM monitoring.

125
00:08:16,002 --> 00:08:20,132
I wonder how LLM monitoring
works for anybody.

126
00:08:20,132 --> 00:08:22,922
And a lot of people listening to
this will be software engineers

127
00:08:22,922 --> 00:08:27,632
who are trying to peek into this
entire AI thing and see whether it's

128
00:08:27,632 --> 00:08:29,462
worth spending more time on that.

129
00:08:30,042 --> 00:08:32,382
How does LLM monitoring work?

130
00:08:32,382 --> 00:08:37,832
I think a LLM monitoring has a lot,
in common with NLP, model monitoring.

131
00:08:37,882 --> 00:08:41,082
When you basically have anything
text-based and you want to

132
00:08:41,082 --> 00:08:45,012
monitor the quality of it that's
something that you would have.

133
00:08:45,402 --> 00:08:50,382
Things like human in the loop that
is required usually for this kind of

134
00:08:50,382 --> 00:08:52,902
systems is part of this monitoring.

135
00:08:52,902 --> 00:08:57,112
You need to have some model
evaluation metrics in place as well.

136
00:08:57,182 --> 00:09:02,322
But those are usually not as
good as having some actual person

137
00:09:02,342 --> 00:09:04,102
checking and giving feedback.

138
00:09:04,842 --> 00:09:10,812
so that aspect, I think is what makes
it a little different, but there are

139
00:09:10,812 --> 00:09:15,062
a lot of other aspects that come to
LLM deployment comparing to, standard

140
00:09:15,262 --> 00:09:20,472
machine learning, basically everything
goes back to prompt engineering.

141
00:09:21,132 --> 00:09:24,572
And, you add a lot of different
things to the prompt and

142
00:09:24,732 --> 00:09:26,542
you need to optimize prompt.

143
00:09:26,552 --> 00:09:29,922
And there are different
optimizers that are available now.

144
00:09:30,382 --> 00:09:33,552
And one of the best ones
I think is adult flow.

145
00:09:33,672 --> 00:09:35,332
that's a product by Lee Yin.

146
00:09:36,002 --> 00:09:38,112
I really recommend you to talk to her.

147
00:09:38,112 --> 00:09:39,172
I think she's amazing.

148
00:09:39,532 --> 00:09:44,292
so she's building one of the best
optimizers for prompt, engineering

149
00:09:44,622 --> 00:09:51,112
out there and basically knowing what
you send to the LLMs, the whole prompt

150
00:09:51,472 --> 00:09:56,962
structure and what are the parts of
the prompt and making it, really clear.

151
00:09:57,162 --> 00:10:01,912
I think that's also a very important
part of this traceability for LLMs.

152
00:10:02,462 --> 00:10:07,102
So what do people get wrong about
MLOps the most do you think?

153
00:10:07,835 --> 00:10:12,155
I think that many people feel
that they need to have all

154
00:10:12,155 --> 00:10:15,225
these shiny tools, to do MLOps.

155
00:10:16,070 --> 00:10:17,550
And I don't think it's true.

156
00:10:18,240 --> 00:10:22,830
As you mentioned earlier, a lot
of components are similar to a

157
00:10:22,830 --> 00:10:25,950
standard software engineering,
the differences in principles.

158
00:10:26,945 --> 00:10:29,165
But you could still use the same tools.

159
00:10:29,195 --> 00:10:33,345
You don't need other tools for version
control or CI/CD or orchestration.

160
00:10:33,865 --> 00:10:38,905
you need some extra tools like,
model registries, p vector, databases

161
00:10:38,935 --> 00:10:44,235
and the whole LLM, stack, but
overall For the standard machine

162
00:10:44,235 --> 00:10:46,385
learning, you don't need other tools.

163
00:10:46,675 --> 00:10:51,195
I think the best would be just to go and
look, what do you already have within

164
00:10:51,195 --> 00:10:56,495
your organization, what tools are there
software engineers use and just use the

165
00:10:56,495 --> 00:11:02,805
same thing because the field evolves
very fast and the tools change very fast.

166
00:11:03,065 --> 00:11:06,885
And you would have to go through
migrations maybe every two, three years

167
00:11:06,915 --> 00:11:08,725
when there is a big change in the tool.

168
00:11:09,135 --> 00:11:13,515
And you want to limit that as much
as possible by limiting amount of

169
00:11:13,515 --> 00:11:18,375
tools and only use them if you really
need it and maybe build some features

170
00:11:18,455 --> 00:11:20,165
yourself if you really have to.

171
00:11:20,745 --> 00:11:23,895
but I believe you need to
stay pragmatic about it.

172
00:11:24,015 --> 00:11:26,255
And that's what a lot of people get from.

173
00:11:26,855 --> 00:11:30,215
So what would be your
advice to overcome that?

174
00:11:30,265 --> 00:11:33,555
take a look at what you already
have and just use what you have.

175
00:11:34,035 --> 00:11:38,485
And if you really miss a
feature, build it maybe first and

176
00:11:38,495 --> 00:11:41,315
otherwise look for alternatives.

177
00:11:41,995 --> 00:11:45,355
It depends also on what type
of organization you're part of.

178
00:11:45,395 --> 00:11:49,295
If you're part of large corporate
organization, getting any new vendor

179
00:11:49,325 --> 00:11:55,295
on board is a lot of work, and it's
often not really worth it in my opinion.

180
00:11:55,595 --> 00:12:01,145
it's a lot of bureaucracy involved,
so yeah, building your own thing

181
00:12:01,145 --> 00:12:06,635
maybe can be a better approach
just because of this complexity

182
00:12:06,685 --> 00:12:12,255
And what do you think is the next
thing in MLOps that needs solving?

183
00:12:12,905 --> 00:12:17,090
What would be the next breakthrough
moment, what would have to happen

184
00:12:17,090 --> 00:12:23,240
in your prediction to make the field
better, bigger, or easier, or, superior

185
00:12:23,260 --> 00:12:24,890
in any metric of your choosing?

186
00:12:24,890 --> 00:12:30,410
I think what needs to happen is
the professionalizing the data

187
00:12:30,410 --> 00:12:33,780
scientists, but there are different
types of data scientists, right?

188
00:12:34,150 --> 00:12:35,980
There are people that are doing research.

189
00:12:36,000 --> 00:12:42,185
I don't think they need to learn mlops,
but people that do data science for

190
00:12:42,185 --> 00:12:46,635
any organization where machine learning
models are professionally deployed.

191
00:12:46,635 --> 00:12:51,965
I think it makes sense to invest into
following the best practices and that

192
00:12:51,965 --> 00:12:57,155
there is no other team that takes
your code, rewrites it and deploys it.

193
00:12:57,155 --> 00:12:58,355
I don't believe in that.

194
00:12:58,880 --> 00:13:04,030
I really believe in data scientists
taking this responsibility and it's

195
00:13:04,050 --> 00:13:07,880
mainly people choice, not really
tooling choice that needs to happen, for

196
00:13:07,890 --> 00:13:10,530
things to become faster in the future.

197
00:13:11,650 --> 00:13:11,960
Yeah.

198
00:13:11,980 --> 00:13:13,110
All this humans.

199
00:13:13,220 --> 00:13:13,890
Oh, man.

200
00:13:15,010 --> 00:13:15,250
All right.

201
00:13:15,260 --> 00:13:18,710
Let's switch gears a little bit
to talk about Marvelous Ops.

202
00:13:18,780 --> 00:13:23,920
you mentioned at the beginning that the
name was inspired by Marvel and that you

203
00:13:24,160 --> 00:13:26,120
seem to enjoy that universe quite a lot.

204
00:13:26,640 --> 00:13:32,340
Tell us more about, how it started,
with whom you're doing it and what

205
00:13:32,340 --> 00:13:33,740
you're trying to achieve with it.

206
00:13:34,495 --> 00:13:40,165
it started quite some time ago, actually,
my colleague Basak, we worked together at

207
00:13:40,585 --> 00:13:42,545
Ahold Delhaize, where I currently work.

208
00:13:42,545 --> 00:13:44,005
She now works at booking.

209
00:13:44,055 --> 00:13:47,845
we worked together in the same
team and we were building MLOps

210
00:13:47,865 --> 00:13:49,265
framework with Databricks.

211
00:13:50,050 --> 00:13:55,150
And we felt what we are doing, you
can't really find anywhere written down,

212
00:13:55,260 --> 00:14:00,330
that a lot of different instructions
are lacking, certain principles

213
00:14:00,620 --> 00:14:02,330
and we decided to write about it.

214
00:14:02,420 --> 00:14:05,280
we actually wanted to have that
more as a part of corporate

215
00:14:05,280 --> 00:14:07,380
blog, but it wasn't allowed.

216
00:14:07,380 --> 00:14:12,980
So we just started our own blog and,
we've been writing there Since then,

217
00:14:13,110 --> 00:14:18,760
and we also have, LinkedIn page and I'm
also very active on LinkedIn to write

218
00:14:18,770 --> 00:14:23,020
about MLOps, to actually educate people
on Databricks specifically a lot of

219
00:14:23,150 --> 00:14:27,840
tutorials are notebook based and that's
something we are not advocating for.

220
00:14:27,840 --> 00:14:32,730
We advocate for tools that promote
better software engineering practices.

221
00:14:32,830 --> 00:14:36,750
we basically want to educate
people and now we go even further

222
00:14:36,750 --> 00:14:39,330
with that by providing a course.

223
00:14:39,350 --> 00:14:43,620
So we have an end to end MLOps
with Databricks course that

224
00:14:43,640 --> 00:14:46,650
actually takes the real use cases.

225
00:14:47,720 --> 00:14:53,120
And we show how to implement it end to
end, how to implement, best software

226
00:14:53,120 --> 00:14:57,850
engineering practices, how to go about
some tricky things, how to track your

227
00:14:57,850 --> 00:15:03,140
data, how to track your code, how to
track your models and how to, rollback

228
00:15:03,190 --> 00:15:06,610
as well, if needed, and what tools,
what features in Databricks you should

229
00:15:06,620 --> 00:15:08,990
use the best for certain scenarios.

230
00:15:08,990 --> 00:15:11,970
That sounds like you're actually going
pretty deep into the details, right?

231
00:15:12,020 --> 00:15:15,570
Are you going to teach people
how to use Git and, CI/CD?

232
00:15:15,580 --> 00:15:19,890
Is that all included or do you send them
back to see the basics somewhere else?

233
00:15:19,890 --> 00:15:22,600
And you focus more on the
AI related part of it?

234
00:15:22,807 --> 00:15:28,097
we expect that people actually know
what Git is and how to do basic CI/CD.

235
00:15:28,687 --> 00:15:34,657
so basically our target audience is not,
beginners, not People that just start

236
00:15:34,707 --> 00:15:38,477
with a machine learning that the people
that are already in the field for a

237
00:15:38,477 --> 00:15:43,377
bit doing data science, maybe machine
learning, engineering, and want to know

238
00:15:43,387 --> 00:15:45,767
how to do it better with Databricks.

239
00:15:45,922 --> 00:15:49,302
So we are going more deep into details.

240
00:15:49,462 --> 00:15:54,112
Let's say if you have a model, in
MLflow, how do you track things with it?

241
00:15:54,202 --> 00:15:56,122
How do you add certain tags to it?

242
00:15:56,122 --> 00:16:01,752
How you search for experiments, how you
need to register your model in Unity

243
00:16:01,752 --> 00:16:03,542
catalog, what is required for that.

244
00:16:03,612 --> 00:16:06,872
And when model is registered,
how you deploy your model

245
00:16:06,872 --> 00:16:08,632
using model serving endpoints.

246
00:16:09,252 --> 00:16:12,222
And what if you want to roll
back, how you would do that?

247
00:16:13,032 --> 00:16:14,412
A lot of different things.

248
00:16:14,442 --> 00:16:19,862
And we cover also Databricks asset
bundles as the way to deploy things and

249
00:16:20,492 --> 00:16:25,332
a lot of tricky things, like how to deal
with private packages, which I believe

250
00:16:25,372 --> 00:16:27,502
is a very annoying topic to deal with.

251
00:16:27,552 --> 00:16:30,522
a lot of details you can't find easily.

252
00:16:30,522 --> 00:16:31,892
When is the course coming out?

253
00:16:31,932 --> 00:16:38,232
It is ready for enrollment and, we
start the cohort on 14th of October.

254
00:16:38,262 --> 00:16:43,882
It's a seven week cohort where
we give feedback to people when

255
00:16:43,882 --> 00:16:47,382
they do assignments and they
are in touch with us constantly.

256
00:16:47,982 --> 00:16:51,852
to get information from us
if they have any questions.

257
00:16:51,862 --> 00:16:57,172
So that's actually an opportunity
to work with us, on a daily basis.

258
00:16:57,172 --> 00:16:59,352
And how do people find your course?

259
00:16:59,372 --> 00:17:00,602
Where do you go to sign up?

260
00:17:01,002 --> 00:17:02,062
It's on Maven.

261
00:17:02,102 --> 00:17:08,262
So I can give you a link, where you
go and enroll, but it's basically on

262
00:17:08,292 --> 00:17:12,982
Maven, if you just look for end to end,
MLOps with Databricks, you will find it.

263
00:17:13,607 --> 00:17:17,967
It is under engineering and data and
engineering, I believe, and it's also

264
00:17:18,117 --> 00:17:20,007
a top selling course at the moment.

265
00:17:20,935 --> 00:17:21,685
congratulations.

266
00:17:21,685 --> 00:17:24,675
I think a lot of people will be interested
in that and I hope it goes well.

267
00:17:24,675 --> 00:17:26,705
Is that the first cohort you're running?

268
00:17:27,400 --> 00:17:27,920
thank you.

269
00:17:27,920 --> 00:17:30,210
Yes, it is, it is our first cohort.

270
00:17:30,730 --> 00:17:34,320
It is actually coming from our
three year experience of working

271
00:17:34,340 --> 00:17:38,130
with Databricks and even broader
experience with MLOps in general.

272
00:17:38,640 --> 00:17:43,090
That's something that we did internally
and our struggles to do things with

273
00:17:43,090 --> 00:17:47,470
Databricks properly for all these years
are basically condensed in this course.

274
00:17:47,470 --> 00:17:47,860
Okay.

275
00:17:48,260 --> 00:17:52,260
So everybody listening to this,
please go and check out the course.

276
00:17:53,090 --> 00:17:58,510
And for you, Maria, I guess the one
last question that I really have is

277
00:17:58,930 --> 00:18:01,990
if you could go back the 10 years
that you mentioned, And you were to

278
00:18:01,990 --> 00:18:05,400
choose whether to go into the same
career path or a different one.

279
00:18:05,800 --> 00:18:06,860
Would you do it again?

280
00:18:07,500 --> 00:18:09,340
Yeah, I would definitely do it again.

281
00:18:09,420 --> 00:18:14,110
I think everything that happens
with you in life in general happens

282
00:18:14,110 --> 00:18:15,900
to you with the best possible way.

283
00:18:16,160 --> 00:18:18,910
it brought me to the person I
am now, so I wouldn't change

284
00:18:18,910 --> 00:18:20,660
anything in my life in general.

285
00:18:21,030 --> 00:18:22,550
and also not in my career path.

286
00:18:22,580 --> 00:18:27,340
And if you are not happy with your career
path, you can always change things, right?

287
00:18:27,365 --> 00:18:30,795
for anybody who is at the beginning
of their career and listening to this,

288
00:18:31,585 --> 00:18:33,345
what would be your word of advice?

289
00:18:33,345 --> 00:18:37,915
So I think in the beginning,
it's good to go broad.

290
00:18:38,435 --> 00:18:40,155
Try a lot of different things.

291
00:18:40,470 --> 00:18:44,310
if you like, data science,
go to try data science.

292
00:18:44,320 --> 00:18:48,150
If you like analytics, some people do
like analytics, engineering, building

293
00:18:48,150 --> 00:18:54,460
dashboard, go try do that work together
with software engineers, that are

294
00:18:54,560 --> 00:19:00,600
working on some features for the website,
for example, and learn what they do.

295
00:19:00,810 --> 00:19:05,720
Just understand as much as possible,
as broad as possible, and then find

296
00:19:05,730 --> 00:19:08,780
things that triggers you the most,
what you like the most, and just

297
00:19:08,810 --> 00:19:10,900
go deeper in that specific thing.

298
00:19:11,590 --> 00:19:14,740
I believe that's a good recipe
for achieving what you like.

299
00:19:14,740 --> 00:19:17,530
And if you end up not liking it,
you can always change, right?

300
00:19:18,130 --> 00:19:18,660
Absolutely.

301
00:19:19,370 --> 00:19:20,810
That sounds like very solid advice.

302
00:19:21,550 --> 00:19:23,110
Maria, thank you very much for your time.

303
00:19:23,400 --> 00:19:28,970
For anybody who wants to go and
follow Maria, Maria on LinkedIn,

304
00:19:29,170 --> 00:19:34,460
marvelous ML ops dot sub stack dot
com is where you find her writing.

305
00:19:35,130 --> 00:19:40,290
And for the course, you just heard how to
find it and to end MLOps with Databricks.

306
00:19:40,660 --> 00:19:43,170
Thank you so much for your time
and I'll see you next time.

307
00:19:43,595 --> 00:19:44,275
Thank you.

308
00:19:44,615 --> 00:19:45,315
Have a great day.

309
00:19:45,625 --> 00:19:46,115
Bye bye.