1
00:00:00,250 --> 00:00:02,058 
In this episode of Data Driven,

2
00:00:02,154 --> 00:00:03,886 
frank and Andy get back to the

3
00:00:03,908 --> 00:00:05,406 
data engineering side of the

4
00:00:05,428 --> 00:00:07,274 
equation by speaking with Sakit

5
00:00:07,322 --> 00:00:09,802 
Saurab, cofounder of Nexler.

6
00:00:09,946 --> 00:00:12,222 
Nexler specializes in tools for

7
00:00:12,276 --> 00:00:13,834 
automating data engineering

8
00:00:13,882 --> 00:00:16,240 
processes. Now onto the show.

9
00:00:20,770 --> 00:00:22,302 
Hello and welcome to Data

10
00:00:22,356 --> 00:00:24,206 
Driven, the podcast where we

11
00:00:24,228 --> 00:00:25,446 
explore the emergency fields of

12
00:00:25,468 --> 00:00:27,110 
data science, machine learning,

13
00:00:27,180 --> 00:00:28,630 
and of course, the ever present

14
00:00:28,700 --> 00:00:32,930 
data engineering. This is season

15
00:00:33,010 --> 00:00:35,094 
seven that we're now in and we

16
00:00:35,132 --> 00:00:38,178 
are welcoming. Andy is shaking

17
00:00:38,194 --> 00:00:39,414 
his head. If you're not watching

18
00:00:39,452 --> 00:00:41,142 
the video, it is hard to believe

19
00:00:41,196 --> 00:00:42,362 
that we hit season seven

20
00:00:42,416 --> 00:00:44,506 
seasons. But by the time this is

21
00:00:44,528 --> 00:00:45,754 
launched you probably have heard

22
00:00:45,792 --> 00:00:47,738 
our one or two shows where we

23
00:00:47,744 --> 00:00:49,754 
did kind of delve in deep. So

24
00:00:49,792 --> 00:00:50,906 
you're probably tired of hearing

25
00:00:50,938 --> 00:00:57,054 
us bang on about that. I really

26
00:00:57,092 --> 00:00:58,334 
like kind of kicking off this

27
00:00:58,372 --> 00:01:01,402 
first guest interview for season

28
00:01:01,466 --> 00:01:07,442 
seven with a Saket Sarab who

29
00:01:07,576 --> 00:01:09,726 
runs, who is co founder and CEO

30
00:01:09,758 --> 00:01:11,246 
of a company called Nexla, whose

31
00:01:11,278 --> 00:01:13,650 
tagline is automation for data

32
00:01:13,720 --> 00:01:15,006 
engineering. And if there's

33
00:01:15,038 --> 00:01:15,938 
anything we've heard about in

34
00:01:15,944 --> 00:01:17,542 
the last, say, six to ten

35
00:01:17,596 --> 00:01:19,030 
months, it's all about

36
00:01:19,180 --> 00:01:20,562 
automation this, automation

37
00:01:20,626 --> 00:01:22,582 
that. Whether it's Chat, GPT or

38
00:01:22,636 --> 00:01:24,694 
any other kind of low code, no

39
00:01:24,732 --> 00:01:26,566 
code, automation is all the

40
00:01:26,588 --> 00:01:31,946 
rage. And he also very much like

41
00:01:31,968 --> 00:01:33,930 
a previous guest, has a cool

42
00:01:34,000 --> 00:01:35,946 
vendor tag from Gartner. So

43
00:01:35,968 --> 00:01:37,466 
we're going to talk to that.

44
00:01:37,568 --> 00:01:38,394 
Welcome to the show.

45
00:01:38,432 --> 00:01:41,146 
Soquette thank you Frank, and

46
00:01:41,168 --> 00:01:42,800 
thank you Andy. Good to be here.

47
00:01:44,130 --> 00:01:47,918 
So very fascinated about kind of

48
00:01:47,924 --> 00:01:49,294 
your story. In the virtual green

49
00:01:49,332 --> 00:01:51,086 
room we were talking about. You

50
00:01:51,108 --> 00:01:52,954 
used to write Linux drivers for

51
00:01:53,092 --> 00:01:56,674 
video card manufacturers and we

52
00:01:56,712 --> 00:01:58,846 
spent a few minutes on waxing

53
00:01:58,878 --> 00:02:01,554 
poetic about how easy Linux has

54
00:02:01,592 --> 00:02:06,354 
become. So what exactly does

55
00:02:06,392 --> 00:02:07,826 
automation for data Engineering

56
00:02:07,858 --> 00:02:09,126 
mean to you? I think let's start

57
00:02:09,148 --> 00:02:09,720 
there.

58
00:02:10,330 --> 00:02:12,920 
Yeah, I think when we look at

59
00:02:13,930 --> 00:02:15,526 
enterprises and companies out

60
00:02:15,548 --> 00:02:17,206 
there with a lot more data, with

61
00:02:17,228 --> 00:02:18,406 
a lot more people who need to

62
00:02:18,428 --> 00:02:20,026 
use data, there are two ways you

63
00:02:20,048 --> 00:02:22,442 
can achieve scale. One is

64
00:02:22,496 --> 00:02:23,786 
through automation and the other

65
00:02:23,808 --> 00:02:25,594 
is through collaboration and

66
00:02:25,632 --> 00:02:27,366 
automation. Or achieving scale

67
00:02:27,398 --> 00:02:28,474 
through automation means that

68
00:02:28,512 --> 00:02:30,794 
the tasks that we do today, can

69
00:02:30,832 --> 00:02:32,954 
they be automated, can they

70
00:02:32,992 --> 00:02:34,266 
become more intelligent? So for

71
00:02:34,288 --> 00:02:35,726 
example, if I had to create a

72
00:02:35,748 --> 00:02:37,502 
data pipeline and I have to

73
00:02:37,556 --> 00:02:38,846 
connect to a data system and

74
00:02:38,868 --> 00:02:40,398 
read that data, process it,

75
00:02:40,484 --> 00:02:41,838 
maybe transform that, push the

76
00:02:41,844 --> 00:02:43,486 
data somewhere, let's say it

77
00:02:43,508 --> 00:02:45,954 
takes me four weeks or six weeks

78
00:02:45,992 --> 00:02:47,918 
to write that code, test it QA,

79
00:02:48,014 --> 00:02:49,134 
take it to production.

80
00:02:49,262 --> 00:02:50,994 
Automation would basically mean

81
00:02:51,032 --> 00:02:53,602 
that can a lot of these things

82
00:02:53,656 --> 00:02:55,074 
be done automatically and

83
00:02:55,112 --> 00:02:56,594 
faster? So can I, for example,

84
00:02:56,712 --> 00:02:57,906 
not have to write a connector?

85
00:02:57,938 --> 00:02:59,590 
It can get auto generated there,

86
00:02:59,660 --> 00:03:01,606 
right? Can I not have to write,

87
00:03:01,788 --> 00:03:04,934 
test or error conditions and

88
00:03:04,972 --> 00:03:06,726 
check for them because the

89
00:03:06,748 --> 00:03:08,262 
system can look at the data,

90
00:03:08,316 --> 00:03:09,558 
understand its properties and

91
00:03:09,564 --> 00:03:10,534 
say, oh, this would be a good

92
00:03:10,572 --> 00:03:11,798 
validation for this type of

93
00:03:11,804 --> 00:03:13,674 
data. For example, or if I had

94
00:03:13,712 --> 00:03:16,266 
to process or run the same

95
00:03:16,288 --> 00:03:17,354 
pipeline. But now the data

96
00:03:17,392 --> 00:03:18,998 
volume has grown ten x, I don't

97
00:03:19,014 --> 00:03:20,278 
have to go and do a whole bunch

98
00:03:20,294 --> 00:03:22,106 
of engineering to manage that

99
00:03:22,128 --> 00:03:23,374 
scale. The system can

100
00:03:23,412 --> 00:03:24,606 
understand, oh, the scale is

101
00:03:24,628 --> 00:03:26,766 
increasing. My bottom neck is in

102
00:03:26,788 --> 00:03:28,126 
this part of the processing. Let

103
00:03:28,148 --> 00:03:29,566 
me allocate more containers to

104
00:03:29,588 --> 00:03:31,294 
that and just let it run

105
00:03:31,332 --> 00:03:33,634 
smoothly. So automation is a lot

106
00:03:33,672 --> 00:03:35,186 
about doing the same tasks that

107
00:03:35,208 --> 00:03:38,830 
we do, but doing that faster.

108
00:03:38,910 --> 00:03:40,674 
Why? Because something can

109
00:03:40,712 --> 00:03:42,562 
figure out certain tasks, do it

110
00:03:42,616 --> 00:03:45,038 
for us, create more reliability,

111
00:03:45,134 --> 00:03:46,382 
create more repeatability,

112
00:03:46,446 --> 00:03:47,422 
create better performance

113
00:03:47,486 --> 00:03:49,118 
without us having to do that

114
00:03:49,144 --> 00:03:50,534 
manual work. So when we go back

115
00:03:50,572 --> 00:03:51,974 
into automation for data

116
00:03:52,012 --> 00:03:53,510 
engineering and you understand

117
00:03:53,580 --> 00:03:54,902 
that there is so much data

118
00:03:54,956 --> 00:03:56,646 
engineering work to do, I think

119
00:03:56,668 --> 00:03:58,726 
it's almost impossible for the

120
00:03:58,748 --> 00:03:59,994 
data engineers out there to just

121
00:04:00,032 --> 00:04:01,466 
support all that demand that

122
00:04:01,488 --> 00:04:03,660 
they have. Automation for them

123
00:04:04,110 --> 00:04:05,366 
is like something that helps

124
00:04:05,398 --> 00:04:06,598 
them and supports them. And it's

125
00:04:06,614 --> 00:04:09,946 
like a lot of easy use cases can

126
00:04:09,968 --> 00:04:11,226 
be done automatically and

127
00:04:11,248 --> 00:04:12,378 
quickly. And a lot of difficult

128
00:04:12,464 --> 00:04:14,510 
use cases can have big chunks

129
00:04:15,010 --> 00:04:16,554 
taken care of in various

130
00:04:16,602 --> 00:04:18,094 
aspects. So that's kind of where

131
00:04:18,212 --> 00:04:19,706 
that direction is and automation

132
00:04:19,738 --> 00:04:21,006 
is one of the key parts to that

133
00:04:21,028 --> 00:04:21,790 
scale.

134
00:04:22,530 --> 00:04:23,280 
Interesting.

135
00:04:23,750 --> 00:04:26,594 
Yeah. I really like your

136
00:04:26,632 --> 00:04:28,078 
description of this. I'm

137
00:04:28,094 --> 00:04:29,186 
wondering if it's okay if you

138
00:04:29,208 --> 00:04:31,358 
share a little more detail. I've

139
00:04:31,454 --> 00:04:33,090 
worked some with automating data

140
00:04:33,160 --> 00:04:36,854 
engineering in the past and I

141
00:04:36,892 --> 00:04:40,802 
find that it's very applicable

142
00:04:40,866 --> 00:04:42,342 
when you're doing pretty much

143
00:04:42,396 --> 00:04:44,134 
straight one to one type stuff

144
00:04:44,252 --> 00:04:47,254 
and that's not throwing off on

145
00:04:47,292 --> 00:04:49,098 
your product by any stretch. I

146
00:04:49,104 --> 00:04:50,106 
don't know if you agree or

147
00:04:50,128 --> 00:04:51,562 
disagree, but I think about

148
00:04:51,616 --> 00:04:54,842 
staging data so I can pull data

149
00:04:54,896 --> 00:04:57,654 
from extracts, text files, flat

150
00:04:57,702 --> 00:05:00,218 
files and load that into some

151
00:05:00,304 --> 00:05:03,014 
data store, usually a database.

152
00:05:03,142 --> 00:05:05,754 
And once I get it there, I find

153
00:05:05,792 --> 00:05:06,878 
a couple of things are true. And

154
00:05:06,884 --> 00:05:07,806 
I may not pull it from an

155
00:05:07,828 --> 00:05:09,038 
extract, I may pull it from the

156
00:05:09,044 --> 00:05:10,686 
system of record. I want to get

157
00:05:10,708 --> 00:05:12,190 
in, get the data and get out

158
00:05:12,260 --> 00:05:13,794 
with doing as little harm as

159
00:05:13,832 --> 00:05:15,074 
possible, stealing as little

160
00:05:15,112 --> 00:05:17,246 
cycles. But once I get my copy

161
00:05:17,278 --> 00:05:19,534 
of it, then I can start applying

162
00:05:19,582 --> 00:05:23,470 
rules, looking for opportunities

163
00:05:23,550 --> 00:05:25,426 
to apply strong data types and

164
00:05:25,448 --> 00:05:28,322 
the like. And automation really

165
00:05:28,376 --> 00:05:30,930 
works well there. Your product,

166
00:05:31,080 --> 00:05:33,094 
I'm I'm assuming, does that and

167
00:05:33,132 --> 00:05:34,438 
does that part well?

168
00:05:34,604 --> 00:05:36,006 
Yeah, absolutely. But there are

169
00:05:36,028 --> 00:05:37,846 
also parts where you're getting

170
00:05:37,868 --> 00:05:39,398 
that extract and there is a

171
00:05:39,404 --> 00:05:41,206 
slight change in schema now and

172
00:05:41,308 --> 00:05:42,634 
it's more or less the same. But

173
00:05:42,672 --> 00:05:44,666 
can automation cache that for

174
00:05:44,688 --> 00:05:45,866 
you and take care of a few of

175
00:05:45,888 --> 00:05:47,146 
those things? Or do you have to

176
00:05:47,168 --> 00:05:49,386 
go back and write that piece of

177
00:05:49,408 --> 00:05:50,806 
code there? So there are many

178
00:05:50,848 --> 00:05:55,594 
places where you can benefit

179
00:05:55,642 --> 00:05:56,814 
from that. So there is what we

180
00:05:56,852 --> 00:05:58,206 
call when we talk about

181
00:05:58,228 --> 00:06:00,154 
automation, what is the driver

182
00:06:00,202 --> 00:06:01,486 
of automation, what is the

183
00:06:01,508 --> 00:06:03,214 
source of that? And we put that

184
00:06:03,252 --> 00:06:05,434 
on an aspect of applying

185
00:06:05,482 --> 00:06:07,454 
intelligence to the metadata. So

186
00:06:07,492 --> 00:06:08,946 
when we look at the data and we

187
00:06:08,968 --> 00:06:10,354 
understand that the metadata is

188
00:06:10,392 --> 00:06:11,966 
actually things like the Schema

189
00:06:11,998 --> 00:06:13,746 
or I have a price attribute in

190
00:06:13,768 --> 00:06:14,914 
my extract, but this is the

191
00:06:14,952 --> 00:06:16,258 
behavior of that attribute, this

192
00:06:16,264 --> 00:06:17,846 
is how it looks like, these are

193
00:06:17,868 --> 00:06:19,094 
the characteristics. And based

194
00:06:19,132 --> 00:06:21,126 
on that metadata, can I apply a

195
00:06:21,148 --> 00:06:22,966 
validation rule to it, for

196
00:06:22,988 --> 00:06:24,134 
example, automatically without

197
00:06:24,172 --> 00:06:25,734 
having to define that. And

198
00:06:25,772 --> 00:06:27,958 
something does it for me that is

199
00:06:28,044 --> 00:06:29,590 
bringing automation. So actually

200
00:06:29,660 --> 00:06:31,500 
the roots of that come from

201
00:06:32,030 --> 00:06:33,466 
there is so much information.

202
00:06:33,568 --> 00:06:35,638 
For example, your data extract

203
00:06:35,734 --> 00:06:37,914 
happens every day at 04:00 p.m.

204
00:06:37,952 --> 00:06:39,482 
And you expect the finished data

205
00:06:39,536 --> 00:06:41,354 
to be ready by five and on

206
00:06:41,392 --> 00:06:42,678 
someday. It doesn't happen. It's

207
00:06:42,694 --> 00:06:44,366 
06:00 and it's not there. So is

208
00:06:44,388 --> 00:06:45,566 
there automation going and

209
00:06:45,588 --> 00:06:48,334 
saying alert it didn't come

210
00:06:48,372 --> 00:06:49,566 
through. Oh, by the way, the

211
00:06:49,588 --> 00:06:50,574 
reason it didn't come through

212
00:06:50,612 --> 00:06:53,006 
was that your stuff was all

213
00:06:53,028 --> 00:06:55,506 
great except these 20 records in

214
00:06:55,528 --> 00:06:56,994 
between completely threw it off

215
00:06:57,032 --> 00:06:59,102 
because it was wrongly formatted

216
00:06:59,166 --> 00:07:02,674 
or whatever. So stuff like that

217
00:07:02,712 --> 00:07:04,366 
is where I think that automation

218
00:07:04,478 --> 00:07:07,186 
really becomes an assist in

219
00:07:07,208 --> 00:07:08,470 
this. So you know the business

220
00:07:08,540 --> 00:07:09,506 
problem, you know what you're

221
00:07:09,538 --> 00:07:11,286 
trying to do, this is how you

222
00:07:11,308 --> 00:07:12,002 
get there faster.

223
00:07:12,066 --> 00:07:14,386 
So I've heard that unexpected

224
00:07:14,418 --> 00:07:16,694 
changes in formats, which I see

225
00:07:16,732 --> 00:07:18,758 
it all the time because it's

226
00:07:18,774 --> 00:07:22,874 
what I do. But I hear that

227
00:07:22,912 --> 00:07:25,114 
addressed under the topic of

228
00:07:25,152 --> 00:07:27,562 
Schema drift. And it can happen

229
00:07:27,616 --> 00:07:28,986 
a couple of different ways. You

230
00:07:29,008 --> 00:07:31,822 
can miss a tab or comma whatever

231
00:07:31,876 --> 00:07:33,518 
delimiter you're using. You can

232
00:07:33,524 --> 00:07:34,846 
either miss one or an extra one

233
00:07:34,868 --> 00:07:37,066 
be inserted in the extract

234
00:07:37,098 --> 00:07:40,842 
process. You can have missing

235
00:07:40,906 --> 00:07:42,922 
files, completely missing files

236
00:07:42,986 --> 00:07:44,866 
for a variety of reasons, some

237
00:07:44,888 --> 00:07:46,846 
of which are legit, like you're

238
00:07:46,878 --> 00:07:49,154 
doing an incremental extract and

239
00:07:49,192 --> 00:07:51,906 
nothing changed and stuff like

240
00:07:51,928 --> 00:07:53,810 
that. And I guess your product

241
00:07:53,880 --> 00:07:55,554 
addresses that has rules for

242
00:07:55,592 --> 00:07:57,154 
saying if this is missing, just

243
00:07:57,192 --> 00:07:59,006 
keep going. It's a slowly

244
00:07:59,038 --> 00:08:00,238 
changing dimension and if we

245
00:08:00,264 --> 00:08:02,374 
miss ten in a row, it's no big

246
00:08:02,412 --> 00:08:03,094 
deal.

247
00:08:03,292 --> 00:08:04,886 
So what we do at a very high

248
00:08:04,908 --> 00:08:05,798 
level, right, when we think

249
00:08:05,804 --> 00:08:06,966 
about data engineering, one of

250
00:08:06,988 --> 00:08:08,354 
the key problems that it solves

251
00:08:08,402 --> 00:08:10,546 
is integrating data, getting

252
00:08:10,588 --> 00:08:12,426 
data from point A to point B and

253
00:08:12,528 --> 00:08:13,866 
making sure it's valid, it is

254
00:08:13,888 --> 00:08:15,306 
trusted, it can be used by the

255
00:08:15,328 --> 00:08:17,034 
downstream application. There is

256
00:08:17,072 --> 00:08:19,180 
often an implicit contract that

257
00:08:20,110 --> 00:08:21,386 
dashboard is relying on this

258
00:08:21,408 --> 00:08:23,194 
sort of information. So what we

259
00:08:23,232 --> 00:08:24,782 
do basically at a high level is

260
00:08:24,836 --> 00:08:27,534 
one we are like hey, we can

261
00:08:27,572 --> 00:08:29,374 
figure out how to connect to new

262
00:08:29,412 --> 00:08:30,926 
systems. This is a part where we

263
00:08:30,948 --> 00:08:31,886 
bring automation to the

264
00:08:31,908 --> 00:08:33,582 
connector creation. So instead

265
00:08:33,636 --> 00:08:34,954 
of writing code for connectors,

266
00:08:35,002 --> 00:08:36,318 
we are able to generate most of

267
00:08:36,324 --> 00:08:37,646 
the connectors out there. So

268
00:08:37,668 --> 00:08:39,010 
that's one part. But when we

269
00:08:39,080 --> 00:08:40,322 
scan so the data we understand,

270
00:08:40,376 --> 00:08:41,886 
we do understand what the Schema

271
00:08:41,918 --> 00:08:42,834 
is and all of that, and we

272
00:08:42,872 --> 00:08:44,654 
present that and automatically

273
00:08:44,702 --> 00:08:45,878 
sort of package that into what

274
00:08:45,884 --> 00:08:47,382 
we call as a logical data

275
00:08:47,436 --> 00:08:50,198 
product that becomes much more

276
00:08:50,364 --> 00:08:52,006 
easily understandable by an

277
00:08:52,028 --> 00:08:54,200 
average data user person who

278
00:08:54,890 --> 00:08:56,294 
understand it. So in that

279
00:08:56,332 --> 00:08:58,454 
process, in between that, yes,

280
00:08:58,492 --> 00:08:59,862 
the Schema Drift is an important

281
00:08:59,916 --> 00:09:01,274 
part, but it's not as

282
00:09:01,312 --> 00:09:02,474 
straightforward because what

283
00:09:02,512 --> 00:09:04,362 
happens is you're getting data

284
00:09:04,416 --> 00:09:06,026 
with first name and last name

285
00:09:06,048 --> 00:09:07,238 
and email address and suddenly

286
00:09:07,254 --> 00:09:09,174 
you get data maybe two records

287
00:09:09,222 --> 00:09:11,146 
which don't match that. Is that

288
00:09:11,168 --> 00:09:12,666 
an error? Is that a change? Is

289
00:09:12,688 --> 00:09:14,506 
that an evolution? You got first

290
00:09:14,528 --> 00:09:16,142 
name, last name, email address

291
00:09:16,196 --> 00:09:17,086 
and now you're also getting

292
00:09:17,108 --> 00:09:17,998 
phone number. Well, it's a

293
00:09:18,004 --> 00:09:19,246 
sparse schema potentially and

294
00:09:19,268 --> 00:09:20,574 
it's a drifting schema. Well

295
00:09:20,612 --> 00:09:21,674 
does it break the downstream

296
00:09:21,722 --> 00:09:24,066 
contract because something got

297
00:09:24,088 --> 00:09:25,406 
renamed or does it just simply

298
00:09:25,438 --> 00:09:26,786 
add to that? So there are a few

299
00:09:26,808 --> 00:09:28,242 
of those aspects. We do cover

300
00:09:28,296 --> 00:09:31,122 
all of those by sort of saying

301
00:09:31,176 --> 00:09:35,474 
that when I connect to a data

302
00:09:35,512 --> 00:09:36,482 
system, I'm going to present

303
00:09:36,536 --> 00:09:38,226 
that data in a certain sort of a

304
00:09:38,248 --> 00:09:39,926 
data product view is we call it

305
00:09:39,948 --> 00:09:41,718 
a logical data product. So

306
00:09:41,884 --> 00:09:43,222 
here's a logical data product,

307
00:09:43,276 --> 00:09:44,758 
this is all there is, these are

308
00:09:44,764 --> 00:09:45,958 
the characteristics and stuff

309
00:09:46,044 --> 00:09:47,446 
and you decide what you want to

310
00:09:47,468 --> 00:09:48,598 
do with it and how you want to

311
00:09:48,604 --> 00:09:49,938 
use it. But once you have a

312
00:09:49,964 --> 00:09:52,074 
consumer for a data product,

313
00:09:52,192 --> 00:09:53,526 
then it sort of implicitly

314
00:09:53,558 --> 00:09:55,114 
creates a contract and we keep

315
00:09:55,152 --> 00:09:58,106 
track of that. And there's some

316
00:09:58,128 --> 00:09:59,466 
interesting concepts that we do

317
00:09:59,488 --> 00:10:00,906 
there, which is you can take a

318
00:10:00,928 --> 00:10:02,106 
data product and create some

319
00:10:02,128 --> 00:10:03,626 
derivatives out of that. So you

320
00:10:03,648 --> 00:10:05,486 
can say I have an orders or

321
00:10:05,508 --> 00:10:07,134 
transactions and it has credit

322
00:10:07,172 --> 00:10:08,926 
card number, I'll mask it and I

323
00:10:08,948 --> 00:10:10,286 
have an order ID and I'll look

324
00:10:10,308 --> 00:10:11,982 
up the items from a different

325
00:10:12,116 --> 00:10:13,762 
entity and now I have a new data

326
00:10:13,816 --> 00:10:15,106 
product which is enriched and

327
00:10:15,128 --> 00:10:17,058 
which is maybe more Pisa. So

328
00:10:17,144 --> 00:10:18,946 
some interesting you know what

329
00:10:18,968 --> 00:10:19,058 
I.

330
00:10:19,064 --> 00:10:21,394 
Like about that is we think

331
00:10:21,432 --> 00:10:23,550 
about tools that visualize

332
00:10:23,630 --> 00:10:26,134 
lineage, Atlas and Purview and

333
00:10:26,172 --> 00:10:27,846 
tools like that, but those are

334
00:10:27,868 --> 00:10:30,294 
very reactive, even though both

335
00:10:30,332 --> 00:10:31,986 
Atlas and Microsoft's

336
00:10:32,018 --> 00:10:34,070 
implementation of that. Purview

337
00:10:35,130 --> 00:10:36,722 
introduce automatic scans,

338
00:10:36,786 --> 00:10:39,202 
automated scans and they manage

339
00:10:39,276 --> 00:10:40,522 
Schema Drift to a certain

340
00:10:40,576 --> 00:10:44,086 
extent. You know this if you've

341
00:10:44,118 --> 00:10:45,962 
ever tried to code a tool to

342
00:10:46,016 --> 00:10:47,498 
react to Schema drift, then you

343
00:10:47,504 --> 00:10:50,954 
know how complex that is. And I

344
00:10:50,992 --> 00:10:52,746 
can't wait to see large language

345
00:10:52,778 --> 00:10:54,174 
models integrated into that

346
00:10:54,212 --> 00:10:56,938 
process because I suspect it'll

347
00:10:56,954 --> 00:10:58,506 
do a much better job of managing

348
00:10:58,538 --> 00:11:01,006 
that than trying to guess what

349
00:11:01,028 --> 00:11:02,318 
this data type should be and

350
00:11:02,324 --> 00:11:03,586 
such. But what I hear you

351
00:11:03,608 --> 00:11:05,198 
describing in your contract

352
00:11:05,374 --> 00:11:08,146 
sounds like a proactive piece to

353
00:11:08,168 --> 00:11:10,674 
that where you meet with your

354
00:11:10,712 --> 00:11:13,566 
client and you say, yeah, here's

355
00:11:13,758 --> 00:11:15,042 
whatever you want to call it,

356
00:11:15,096 --> 00:11:16,566 
our data dictionary, what have

357
00:11:16,588 --> 00:11:18,806 
you of all of our source data

358
00:11:18,908 --> 00:11:20,514 
and our fields and columns,

359
00:11:20,642 --> 00:11:22,854 
fields, columns, metadata for

360
00:11:22,892 --> 00:11:24,626 
all of that. And then you're

361
00:11:24,658 --> 00:11:27,986 
saying these particular fields

362
00:11:28,018 --> 00:11:29,198 
are important because they're

363
00:11:29,234 --> 00:11:30,982 
used downstream in these dozen

364
00:11:31,046 --> 00:11:32,746 
reports at the very end of the

365
00:11:32,768 --> 00:11:34,842 
process. But what you're saying

366
00:11:34,896 --> 00:11:37,386 
is if another related field was

367
00:11:37,408 --> 00:11:39,898 
to show up in that list, then

368
00:11:39,984 --> 00:11:41,710 
you're going to be able to make

369
00:11:41,860 --> 00:11:44,302 
an educated guess at whether

370
00:11:44,356 --> 00:11:46,302 
that's schema drift or whether

371
00:11:46,356 --> 00:11:48,526 
it's additional attributes. Is

372
00:11:48,548 --> 00:11:49,886 
that what I'm understanding we

373
00:11:49,908 --> 00:11:50,094 
make.

374
00:11:50,132 --> 00:11:51,934 
An educated guess about? Is this

375
00:11:51,972 --> 00:11:53,086 
a one off thing and we should

376
00:11:53,108 --> 00:11:55,694 
treat it as an error? Hey, this

377
00:11:55,732 --> 00:11:56,746 
record is an error, it didn't

378
00:11:56,778 --> 00:11:57,878 
actually meet certain criteria

379
00:11:57,914 --> 00:11:59,794 
or this is a change that is

380
00:11:59,912 --> 00:12:01,314 
showing up. And the way you do

381
00:12:01,352 --> 00:12:02,754 
that is you observe that data

382
00:12:02,792 --> 00:12:04,210 
over a certain period of time to

383
00:12:04,280 --> 00:12:06,146 
make the determination and we do

384
00:12:06,168 --> 00:12:08,434 
a certain level of drift

385
00:12:08,482 --> 00:12:10,374 
analysis. And if the drift is

386
00:12:10,412 --> 00:12:12,006 
very significant, then what we

387
00:12:12,028 --> 00:12:14,774 
do is we today actually do not

388
00:12:14,812 --> 00:12:16,198 
go and make assumptions on

389
00:12:16,284 --> 00:12:18,614 
behalf of the user. We actually

390
00:12:18,652 --> 00:12:19,846 
create a notification and saying

391
00:12:19,868 --> 00:12:21,686 
we saw a significant change and

392
00:12:21,708 --> 00:12:23,114 
we think this might be a new

393
00:12:23,152 --> 00:12:24,746 
data product that we have

394
00:12:24,768 --> 00:12:26,346 
detected. So we are connected to

395
00:12:26,368 --> 00:12:27,338 
a store. So looking at the

396
00:12:27,344 --> 00:12:28,406 
schemas and stuff and we're

397
00:12:28,438 --> 00:12:29,562 
saying here's a data product,

398
00:12:29,616 --> 00:12:31,066 
that data product is

399
00:12:31,248 --> 00:12:32,746 
transactions and this is what it

400
00:12:32,768 --> 00:12:34,394 
looks like. Oh, some came here

401
00:12:34,432 --> 00:12:35,758 
and whatnot. But then at some

402
00:12:35,764 --> 00:12:36,814 
point we may say, hey, this

403
00:12:36,852 --> 00:12:37,822 
looks significantly different,

404
00:12:37,876 --> 00:12:39,886 
would you like to consider this

405
00:12:39,908 --> 00:12:41,406 
as a new entity? And then it

406
00:12:41,428 --> 00:12:42,686 
sort of notifies the user and

407
00:12:42,708 --> 00:12:44,554 
lets them do that. I think when

408
00:12:44,612 --> 00:12:46,274 
building automation, at least my

409
00:12:46,312 --> 00:12:47,314 
understanding is that's very

410
00:12:47,352 --> 00:12:50,626 
important to understand, when to

411
00:12:50,648 --> 00:12:52,194 
make a reasonable assumption and

412
00:12:52,232 --> 00:12:53,486 
when to actually let the user

413
00:12:53,518 --> 00:12:55,074 
decide. But even creating that

414
00:12:55,112 --> 00:12:56,562 
workflow is a big value add

415
00:12:56,616 --> 00:12:57,646 
because this is stuff that

416
00:12:57,688 --> 00:12:58,706 
actually would have gotten

417
00:12:58,738 --> 00:13:01,414 
missed maybe for a few days, but

418
00:13:01,452 --> 00:13:02,614 
you are getting notified about

419
00:13:02,652 --> 00:13:04,662 
that upfront. So we try to be

420
00:13:04,716 --> 00:13:05,494 
maybe a little bit more

421
00:13:05,532 --> 00:13:08,380 
conservative, if anything about

422
00:13:08,750 --> 00:13:10,218 
making assumptions on behalf of

423
00:13:10,224 --> 00:13:11,882 
the customer or the user because

424
00:13:12,016 --> 00:13:15,386 
you go wrong and it's not fun at

425
00:13:15,408 --> 00:13:15,594 
all.

426
00:13:15,632 --> 00:13:17,580 
Yeah, I really like it.

427
00:13:20,990 --> 00:13:22,218 
Now that's interesting. And I

428
00:13:22,224 --> 00:13:26,158 
see your company has been around

429
00:13:26,244 --> 00:13:27,502 
about seven years.

430
00:13:27,636 --> 00:13:28,320 
Yeah.

431
00:13:29,490 --> 00:13:31,054 
What has changed in those seven

432
00:13:31,092 --> 00:13:32,634 
years? Because I think seven

433
00:13:32,692 --> 00:13:36,562 
years ago AI was not on the top

434
00:13:36,616 --> 00:13:38,178 
of everyone's lips, obviously. I

435
00:13:38,184 --> 00:13:40,990 
think certainly since Chat GPT

436
00:13:41,070 --> 00:13:42,914 
came out. Right. It's a big part

437
00:13:42,952 --> 00:13:46,034 
of the conversation. Have you

438
00:13:46,072 --> 00:13:47,382 
also seen kind of the data

439
00:13:47,436 --> 00:13:48,806 
engineering world be kind of

440
00:13:48,828 --> 00:13:52,360 
touched by AI and if so, how?

441
00:13:53,290 --> 00:13:54,886 
Not as much, but I think it is

442
00:13:54,908 --> 00:13:55,846 
getting there. So when we

443
00:13:55,868 --> 00:13:58,246 
started very early, as I said,

444
00:13:58,348 --> 00:13:59,654 
seven years ago, we had actually

445
00:13:59,692 --> 00:14:03,254 
come in with the question when

446
00:14:03,292 --> 00:14:04,426 
starting the company was do we

447
00:14:04,448 --> 00:14:05,926 
want to build an AI or a machine

448
00:14:05,958 --> 00:14:06,762 
learning company? Because

449
00:14:06,816 --> 00:14:09,258 
actually in 2016 it was hot, it

450
00:14:09,264 --> 00:14:10,378 
was hype. There was a lot of

451
00:14:10,384 --> 00:14:11,226 
like, oh, this is going to

452
00:14:11,248 --> 00:14:12,906 
change the world. Always the

453
00:14:12,928 --> 00:14:14,394 
hype is ahead of the reality.

454
00:14:14,442 --> 00:14:16,094 
And it took a while before

455
00:14:16,292 --> 00:14:18,686 
things like LLM came around and

456
00:14:18,708 --> 00:14:19,758 
generative AI is starting to

457
00:14:19,764 --> 00:14:21,034 
succeed. But even otherwise

458
00:14:21,082 --> 00:14:22,058 
there are a lot of other AI

459
00:14:22,074 --> 00:14:22,926 
initiatives that are still

460
00:14:22,948 --> 00:14:24,538 
figuring their way out. But we

461
00:14:24,564 --> 00:14:25,858 
were very clear that we want to

462
00:14:25,864 --> 00:14:27,586 
build the technology for the

463
00:14:27,608 --> 00:14:29,202 
users of data. We want to focus

464
00:14:29,256 --> 00:14:34,306 
on users of data and allow the

465
00:14:34,328 --> 00:14:35,522 
user of data to get the data

466
00:14:35,576 --> 00:14:36,994 
wherever they need and do

467
00:14:37,032 --> 00:14:38,438 
whatever they want to do with it

468
00:14:38,444 --> 00:14:39,686 
in whatever tool they want to do

469
00:14:39,708 --> 00:14:41,254 
with. So we came with that

470
00:14:41,292 --> 00:14:43,926 
approach because we said that

471
00:14:43,948 --> 00:14:45,638 
the user of data is not going to

472
00:14:45,644 --> 00:14:49,766 
be necessarily very expert in

473
00:14:49,788 --> 00:14:51,266 
the data system. So there's

474
00:14:51,298 --> 00:14:52,726 
expertise in data, which is I

475
00:14:52,748 --> 00:14:53,818 
understand what the data is, and

476
00:14:53,824 --> 00:14:54,634 
there's expertise in data

477
00:14:54,672 --> 00:14:55,626 
systems, which is with the more

478
00:14:55,648 --> 00:14:57,114 
engineering side, we said the

479
00:14:57,152 --> 00:14:58,518 
data user will not be an expert

480
00:14:58,534 --> 00:14:59,970 
in data. So if you have a lot

481
00:14:59,980 --> 00:15:01,526 
more variety of data that's

482
00:15:01,558 --> 00:15:02,954 
coming, how do they use it?

483
00:15:02,992 --> 00:15:04,134 
We'll create an abstraction.

484
00:15:04,182 --> 00:15:05,386 
We'll create an abstraction that

485
00:15:05,408 --> 00:15:06,622 
will give them a clean,

486
00:15:06,676 --> 00:15:07,726 
consistent view of data no

487
00:15:07,748 --> 00:15:08,846 
matter where it comes from. So

488
00:15:08,868 --> 00:15:09,838 
now you're like, I don't care if

489
00:15:09,844 --> 00:15:11,486 
the data was a stream or API or

490
00:15:11,508 --> 00:15:13,486 
JSON or document or whatever, I

491
00:15:13,508 --> 00:15:14,606 
have something consistent to

492
00:15:14,628 --> 00:15:16,606 
work with. So we went in and

493
00:15:16,628 --> 00:15:18,014 
looked at metadata and started

494
00:15:18,052 --> 00:15:19,534 
to apply metadata intelligence

495
00:15:19,582 --> 00:15:21,346 
to create that. I would say that

496
00:15:21,368 --> 00:15:22,578 
when we were doing it, a lot of

497
00:15:22,584 --> 00:15:23,826 
people early on were like, why

498
00:15:23,848 --> 00:15:25,186 
are you doing that? Why not just

499
00:15:25,208 --> 00:15:26,494 
create like this straightforward

500
00:15:26,542 --> 00:15:28,066 
thing that everybody does? What

501
00:15:28,088 --> 00:15:29,398 
has changed for us is in the

502
00:15:29,404 --> 00:15:31,254 
last two years or so, our

503
00:15:31,292 --> 00:15:32,694 
approach of creating this

504
00:15:32,732 --> 00:15:34,534 
logical entity around data and

505
00:15:34,572 --> 00:15:36,726 
using that started to catch on

506
00:15:36,748 --> 00:15:37,782 
with the concept of data

507
00:15:37,836 --> 00:15:39,222 
products. So where we were

508
00:15:39,276 --> 00:15:40,854 
initially struggling to say what

509
00:15:40,892 --> 00:15:42,154 
does this mean? And why is it

510
00:15:42,192 --> 00:15:44,282 
valuable? Has suddenly become

511
00:15:44,336 --> 00:15:45,594 
like, oh, it makes so much sense

512
00:15:45,632 --> 00:15:46,746 
that you guys have done it this

513
00:15:46,768 --> 00:15:48,650 
way. So that has certainly

514
00:15:48,720 --> 00:15:50,118 
changed. I think that

515
00:15:50,224 --> 00:15:52,606 
application of intelligence to

516
00:15:52,628 --> 00:15:55,150 
the metadata itself to make data

517
00:15:55,220 --> 00:15:57,546 
tasks themselves more automated

518
00:15:57,738 --> 00:16:01,166 
is a very valid use case. The

519
00:16:01,188 --> 00:16:02,782 
generative models are doing

520
00:16:02,836 --> 00:16:05,026 
quite well in things like can

521
00:16:05,048 --> 00:16:06,434 
you generate a description for

522
00:16:06,472 --> 00:16:07,746 
this data if this data looks

523
00:16:07,768 --> 00:16:11,570 
like this? There's also been

524
00:16:11,640 --> 00:16:12,978 
some really interesting stuff in

525
00:16:12,984 --> 00:16:14,626 
terms of generating code as far

526
00:16:14,648 --> 00:16:15,586 
as data engineering is

527
00:16:15,608 --> 00:16:16,654 
concerned. Hey, can you generate

528
00:16:16,702 --> 00:16:19,126 
code that reads it up from here

529
00:16:19,148 --> 00:16:20,646 
and pushes it out there? So I

530
00:16:20,668 --> 00:16:21,846 
think there is that happening as

531
00:16:21,868 --> 00:16:23,366 
well. So there's a lot of, I

532
00:16:23,388 --> 00:16:26,454 
think, places transforming data.

533
00:16:26,492 --> 00:16:28,134 
For example, if the data looks

534
00:16:28,172 --> 00:16:29,222 
like this and it has to become

535
00:16:29,276 --> 00:16:30,826 
like that, can we figure out

536
00:16:30,848 --> 00:16:32,266 
what is the logic in between? So

537
00:16:32,288 --> 00:16:35,286 
I think I would say that it's

538
00:16:35,318 --> 00:16:36,474 
good that we're moving in that

539
00:16:36,512 --> 00:16:39,226 
direction because I believe that

540
00:16:39,248 --> 00:16:40,554 
the number of things to do is so

541
00:16:40,592 --> 00:16:42,318 
massive that some degree of

542
00:16:42,324 --> 00:16:43,710 
automation is essential.

543
00:16:44,050 --> 00:16:45,806 
I totally agree. And I was just

544
00:16:45,828 --> 00:16:47,550 
scrolling, so apologies, Frank,

545
00:16:48,130 --> 00:16:50,478 
scrolling around on Nexla.com

546
00:16:50,564 --> 00:16:51,710 
and looking at your data

547
00:16:51,780 --> 00:16:54,786 
operations. And first I want to

548
00:16:54,808 --> 00:16:56,770 
commend you for next sets.

549
00:16:57,110 --> 00:17:00,386 
That's a cool play. And some of

550
00:17:00,408 --> 00:17:02,580 
the fields that you cover here

551
00:17:05,050 --> 00:17:06,886 
in my career, I've been doing

552
00:17:06,908 --> 00:17:08,600 
this for a long time. I'm old

553
00:17:09,130 --> 00:17:12,354 
socket but Continuous Metadata

554
00:17:12,402 --> 00:17:15,766 
intelligence caught my eye. That

555
00:17:15,788 --> 00:17:17,798 
is a very cool concept. And I've

556
00:17:17,814 --> 00:17:19,354 
done what you described here

557
00:17:19,392 --> 00:17:20,646 
sounds like some stuff that I've

558
00:17:20,678 --> 00:17:23,734 
done but that's a much cooler

559
00:17:23,782 --> 00:17:28,502 
name. Continuous Metadata CMI

560
00:17:28,566 --> 00:17:30,410 
that just rolls off the tongue.

561
00:17:31,410 --> 00:17:34,154 
The idea, the automated Error

562
00:17:34,202 --> 00:17:35,498 
management and quarantine. I'm

563
00:17:35,514 --> 00:17:36,286 
just kind of going from the

564
00:17:36,308 --> 00:17:38,014 
bottom up here on the page for

565
00:17:38,052 --> 00:17:41,022 
data operations. Those are just

566
00:17:41,076 --> 00:17:44,434 
key pieces of functionality that

567
00:17:44,632 --> 00:17:47,042 
time and time again data

568
00:17:47,096 --> 00:17:50,162 
Warehouse ETL includes something

569
00:17:50,216 --> 00:17:51,646 
like that. But everybody's

570
00:17:51,678 --> 00:17:56,786 
rolling it from scratch. This is

571
00:17:56,808 --> 00:17:58,454 
just very cool. I love this idea

572
00:17:58,492 --> 00:18:02,338 
of abstracting that out and I'm

573
00:18:02,354 --> 00:18:03,286 
just going to throw this out

574
00:18:03,308 --> 00:18:04,226 
there. I'm going to be signing

575
00:18:04,258 --> 00:18:05,814 
up for a demo. I want to see

576
00:18:05,852 --> 00:18:06,486 
more.

577
00:18:06,668 --> 00:18:08,214 
Oh absolutely. You're very

578
00:18:08,252 --> 00:18:09,766 
welcome for that. And I think

579
00:18:09,788 --> 00:18:11,858 
the other problem that it helped

580
00:18:11,874 --> 00:18:13,258 
us solve, as I said, there are

581
00:18:13,264 --> 00:18:14,426 
two ways to get scale in

582
00:18:14,448 --> 00:18:15,386 
enterprise. One is through

583
00:18:15,408 --> 00:18:16,426 
automation and the other is

584
00:18:16,448 --> 00:18:18,074 
through collaboration. By

585
00:18:18,112 --> 00:18:19,462 
creating this abstracted entity

586
00:18:19,526 --> 00:18:21,210 
we were able to say hey, this is

587
00:18:21,280 --> 00:18:23,054 
a lot more easy to understand as

588
00:18:23,092 --> 00:18:25,854 
access control. I can be really

589
00:18:25,892 --> 00:18:27,230 
good at connecting to the data

590
00:18:27,300 --> 00:18:29,694 
from a transaction system and

591
00:18:29,732 --> 00:18:31,520 
cleaning it up and doing some

592
00:18:32,130 --> 00:18:33,566 
applying some compliance to it.

593
00:18:33,588 --> 00:18:35,310 
But then the output of my work,

594
00:18:35,380 --> 00:18:36,538 
which is the cool thing about

595
00:18:36,564 --> 00:18:37,634 
these next sets of the data

596
00:18:37,672 --> 00:18:39,746 
products, is you take one and

597
00:18:39,768 --> 00:18:41,106 
you apply some operations to the

598
00:18:41,128 --> 00:18:42,386 
output is another next set,

599
00:18:42,488 --> 00:18:44,270 
which is identical in behavior

600
00:18:44,350 --> 00:18:46,818 
and consistency, but it is a

601
00:18:46,824 --> 00:18:47,906 
slightly different view of the

602
00:18:47,928 --> 00:18:48,882 
data and you can give somebody

603
00:18:48,936 --> 00:18:50,166 
else access to that and you can

604
00:18:50,188 --> 00:18:51,558 
keep repeating that process. You

605
00:18:51,564 --> 00:18:53,398 
can imagine in a large company

606
00:18:53,484 --> 00:18:54,966 
people are finding these,

607
00:18:54,988 --> 00:18:55,766 
they're creating their own

608
00:18:55,788 --> 00:18:57,014 
variants and they're using it.

609
00:18:57,052 --> 00:18:58,146 
But what they are doing becomes

610
00:18:58,178 --> 00:18:59,526 
an input to somebody else and

611
00:18:59,548 --> 00:19:00,678 
they can go take it out of

612
00:19:00,684 --> 00:19:02,282 
there. And what we did a month

613
00:19:02,336 --> 00:19:04,006 
back was introduced the concept

614
00:19:04,038 --> 00:19:05,690 
of you can take all of these

615
00:19:05,760 --> 00:19:07,574 
logical data products and Nexus

616
00:19:07,622 --> 00:19:08,906 
that we are creating and make

617
00:19:08,928 --> 00:19:10,186 
them into, put them into a

618
00:19:10,208 --> 00:19:11,578 
marketplace that is internal to

619
00:19:11,584 --> 00:19:12,762 
the company. You can allow

620
00:19:12,816 --> 00:19:15,006 
people to go find it request to

621
00:19:15,028 --> 00:19:16,014 
get that you're not really

622
00:19:16,052 --> 00:19:17,198 
buying. You're saying hey, can I

623
00:19:17,204 --> 00:19:18,800 
get access to that? And there is

624
00:19:19,330 --> 00:19:21,614 
a mechanism to approve and give

625
00:19:21,652 --> 00:19:23,374 
people access to that. Now the

626
00:19:23,412 --> 00:19:24,466 
interesting thing is that these

627
00:19:24,488 --> 00:19:27,154 
are all very importantly, these

628
00:19:27,192 --> 00:19:28,482 
Nexus or Data products are

629
00:19:28,536 --> 00:19:30,114 
logical entities. They're not

630
00:19:30,152 --> 00:19:31,506 
making copies of data. So

631
00:19:31,528 --> 00:19:33,074 
they're bringing the same sort

632
00:19:33,112 --> 00:19:34,226 
of benefits that

633
00:19:34,328 --> 00:19:35,682 
containerization for example,

634
00:19:35,736 --> 00:19:37,106 
has done on the compute side is

635
00:19:37,128 --> 00:19:38,046 
like you have an abstracted

636
00:19:38,078 --> 00:19:39,666 
entity, it's consistent, you

637
00:19:39,688 --> 00:19:40,694 
don't have to worry about what

638
00:19:40,732 --> 00:19:42,086 
was under the hood. Where did it

639
00:19:42,108 --> 00:19:43,926 
come from, was it XML data, was

640
00:19:43,948 --> 00:19:45,942 
it CSV? Now you have something

641
00:19:45,996 --> 00:19:47,526 
consistent to work with and it

642
00:19:47,548 --> 00:19:48,566 
opens up a whole bunch of

643
00:19:48,588 --> 00:19:50,022 
interfaces. I can take that data

644
00:19:50,076 --> 00:19:51,418 
product, that next set and say,

645
00:19:51,504 --> 00:19:52,938 
I would like this data in a

646
00:19:52,944 --> 00:19:55,066 
warehouse or I would like this

647
00:19:55,088 --> 00:19:57,030 
data as an API. And you realize

648
00:19:57,110 --> 00:19:59,114 
that the same entity can have

649
00:19:59,152 --> 00:20:00,906 
benefits for different users and

650
00:20:00,928 --> 00:20:01,946 
they can approach it in

651
00:20:01,968 --> 00:20:03,418 
different ways. So we think that

652
00:20:03,424 --> 00:20:04,234 
is what is bringing

653
00:20:04,282 --> 00:20:05,486 
collaboration. So when you bring

654
00:20:05,508 --> 00:20:07,246 
together automation on one hand,

655
00:20:07,348 --> 00:20:08,446 
collaboration on other, and then

656
00:20:08,468 --> 00:20:10,126 
you really get the benefits of

657
00:20:10,148 --> 00:20:12,302 
scale from both technology and

658
00:20:12,356 --> 00:20:12,960 
process.

659
00:20:13,350 --> 00:20:16,402 
I absolutely love you. Now I get

660
00:20:16,456 --> 00:20:18,420 
why you keep saying data product

661
00:20:18,790 --> 00:20:21,090 
and it makes perfect sense now.

662
00:20:21,240 --> 00:20:23,650 
You're creating a very

663
00:20:23,720 --> 00:20:26,006 
interesting, almost an

664
00:20:26,028 --> 00:20:29,414 
integration layer in between the

665
00:20:29,452 --> 00:20:33,286 
idea of Containerized for code

666
00:20:33,468 --> 00:20:35,718 
and you're containerizing data.

667
00:20:35,804 --> 00:20:37,046 
And that's what I believe your

668
00:20:37,068 --> 00:20:39,558 
data product represents. And now

669
00:20:39,644 --> 00:20:40,934 
I'm really interested in that

670
00:20:40,972 --> 00:20:41,654 
demo.

671
00:20:41,852 --> 00:20:44,646 
Well, especially if my recent

672
00:20:44,678 --> 00:20:46,746 
forays into OpenShift and kind

673
00:20:46,768 --> 00:20:48,746 
of what Containerization has

674
00:20:48,768 --> 00:20:51,166 
done for developers, I think

675
00:20:51,188 --> 00:20:52,382 
it's only a matter of time

676
00:20:52,436 --> 00:20:53,838 
before containerization kind of

677
00:20:53,844 --> 00:20:56,906 
hits the data world. And there's

678
00:20:56,938 --> 00:21:02,702 
something I want to point out is

679
00:21:02,756 --> 00:21:05,566 
that it was very smart, I think,

680
00:21:05,588 --> 00:21:09,614 
of you to focus on the data

681
00:21:09,652 --> 00:21:10,786 
engineering side, right. Because

682
00:21:10,808 --> 00:21:12,866 
AI is the hype machine, right. I

683
00:21:12,888 --> 00:21:14,098 
fully admit that I say this as a

684
00:21:14,104 --> 00:21:16,326 
data scientist, right? But one

685
00:21:16,348 --> 00:21:17,814 
of the things that we've kind of

686
00:21:17,852 --> 00:21:20,678 
discovered, both Andy and I, and

687
00:21:20,764 --> 00:21:22,418 
in both our professional careers

688
00:21:22,514 --> 00:21:25,366 
is that data scientists will

689
00:21:25,388 --> 00:21:27,254 
tend to brush aside the simple

690
00:21:27,372 --> 00:21:30,470 
basics of it's. Five words,

691
00:21:30,540 --> 00:21:31,802 
right? First you get the data,

692
00:21:31,856 --> 00:21:33,820 
or first we got the data. Right.

693
00:21:34,190 --> 00:21:36,794 
Behind that is months of work.

694
00:21:36,832 --> 00:21:38,218 
It's orders of magnitude of

695
00:21:38,224 --> 00:21:39,558 
work. In fact, one of the spiels

696
00:21:39,574 --> 00:21:42,958 
I have now is kind of like the

697
00:21:42,964 --> 00:21:44,602 
idea of rock stars and roadies,

698
00:21:44,746 --> 00:21:46,430 
right? And for every rock star

699
00:21:46,500 --> 00:21:48,286 
there's, there's an army of

700
00:21:48,308 --> 00:21:50,714 
Rhodeies that set up the lights

701
00:21:50,762 --> 00:21:54,158 
to move the chairs around and

702
00:21:54,244 --> 00:21:55,742 
take set up the equipment and

703
00:21:55,796 --> 00:21:56,946 
manage the sound. So I think

704
00:21:56,968 --> 00:21:58,738 
that the data engineering, I

705
00:21:58,744 --> 00:22:00,034 
think, is one of those things

706
00:22:00,072 --> 00:22:02,018 
that I think has not it's like

707
00:22:02,024 --> 00:22:04,066 
the Rodney Dangerfield of kind

708
00:22:04,088 --> 00:22:06,406 
of the data world where it

709
00:22:06,428 --> 00:22:07,782 
didn't get any respect up until

710
00:22:07,836 --> 00:22:11,366 
lately because Chat GPT will get

711
00:22:11,388 --> 00:22:12,870 
all the headlines, right? But

712
00:22:12,940 --> 00:22:14,166 
think about the data that went

713
00:22:14,188 --> 00:22:16,982 
into it. I've heard numbers,

714
00:22:17,116 --> 00:22:18,486 
billions and trillions of

715
00:22:18,508 --> 00:22:21,786 
parameters for four. I think

716
00:22:21,808 --> 00:22:23,226 
it's smart that you picked that

717
00:22:23,248 --> 00:22:24,154 
and I think that actually worked

718
00:22:24,192 --> 00:22:25,418 
out pretty well for you.

719
00:22:25,584 --> 00:22:27,340 
Actually what worked,

720
00:22:27,790 --> 00:22:30,826 
fortunately for us, was that I

721
00:22:30,848 --> 00:22:32,170 
actually really looked at

722
00:22:32,240 --> 00:22:35,078 
machine learning as a way to I'd

723
00:22:35,094 --> 00:22:36,210 
been an entrepreneur before. I'd

724
00:22:36,230 --> 00:22:37,866 
built a company I really enjoyed

725
00:22:37,978 --> 00:22:39,386 
sort of building the data aspect

726
00:22:39,418 --> 00:22:40,526 
of it. I had built a company in

727
00:22:40,548 --> 00:22:41,790 
the advertising at Tech Space,

728
00:22:41,860 --> 00:22:42,986 
built one of the earliest mobile

729
00:22:43,018 --> 00:22:45,342 
ad servers. We became part of

730
00:22:45,396 --> 00:22:46,786 
the largest ad exchange at the

731
00:22:46,808 --> 00:22:48,098 
time outside of Google. So we

732
00:22:48,104 --> 00:22:49,742 
were processing over 300 billion

733
00:22:49,806 --> 00:22:52,206 
records a day, and my co founder

734
00:22:52,238 --> 00:22:52,994 
here was running that

735
00:22:53,032 --> 00:22:54,402 
infrastructure. And we're like,

736
00:22:54,536 --> 00:22:55,906 
at some level, be candid, I

737
00:22:55,928 --> 00:22:56,946 
didn't really enjoy being in

738
00:22:56,968 --> 00:22:58,546 
advertising, but I did like the

739
00:22:58,568 --> 00:22:59,698 
sort of data challenges that

740
00:22:59,704 --> 00:23:01,062 
were there. So we were looking

741
00:23:01,116 --> 00:23:02,486 
to figure out what is the next

742
00:23:02,508 --> 00:23:03,286 
because we had taken that

743
00:23:03,308 --> 00:23:04,326 
company public and all that

744
00:23:04,348 --> 00:23:05,846 
stuff, and we moved on. I was

745
00:23:05,868 --> 00:23:07,046 
like, okay, what is the next

746
00:23:07,068 --> 00:23:08,614 
thing you want to work on? And I

747
00:23:08,652 --> 00:23:09,862 
really seriously looked at

748
00:23:09,916 --> 00:23:10,982 
building something because

749
00:23:11,036 --> 00:23:12,522 
Machine Learning and AI was hot

750
00:23:12,576 --> 00:23:14,906 
topic even in 20, 15, 16. But

751
00:23:14,928 --> 00:23:15,866 
when I looked at it, I

752
00:23:15,888 --> 00:23:17,594 
understood that at some level,

753
00:23:17,712 --> 00:23:19,354 
it is so specific to that

754
00:23:19,392 --> 00:23:21,018 
business and that company and

755
00:23:21,024 --> 00:23:22,726 
that problem. It almost becomes

756
00:23:22,758 --> 00:23:24,014 
consultative. And I was like,

757
00:23:24,052 --> 00:23:25,418 
how do you platform it? It's

758
00:23:25,434 --> 00:23:27,422 
hard to platform because even

759
00:23:27,556 --> 00:23:28,682 
you can take two retail

760
00:23:28,746 --> 00:23:30,074 
companies and they're solving

761
00:23:30,122 --> 00:23:31,166 
maybe the same problem of

762
00:23:31,188 --> 00:23:32,638 
recommending products to people

763
00:23:32,724 --> 00:23:34,174 
and their models will be very

764
00:23:34,212 --> 00:23:35,546 
different. And what you do there

765
00:23:35,588 --> 00:23:38,386 
is not easily translatable. So I

766
00:23:38,408 --> 00:23:39,954 
hesitated for that reason. And

767
00:23:39,992 --> 00:23:41,374 
when I look back, in hindsight,

768
00:23:41,422 --> 00:23:43,042 
almost some of the major

769
00:23:43,096 --> 00:23:44,034 
companies that came out on

770
00:23:44,072 --> 00:23:44,894 
machine Learning, ultimately,

771
00:23:44,942 --> 00:23:46,146 
when you look deeper into it,

772
00:23:46,168 --> 00:23:47,534 
there are large professional

773
00:23:47,582 --> 00:23:49,766 
services organizations under the

774
00:23:49,788 --> 00:23:52,598 
hood. And that's what gave us a

775
00:23:52,604 --> 00:23:53,526 
hesitation. I'm more of an

776
00:23:53,548 --> 00:23:54,726 
engineer. I like to build a

777
00:23:54,748 --> 00:23:56,406 
platform and make it once and

778
00:23:56,428 --> 00:23:58,054 
let people use it. And the data

779
00:23:58,092 --> 00:23:59,494 
engineering part of it is what

780
00:23:59,532 --> 00:24:01,878 
looked like that thing. But I

781
00:24:01,884 --> 00:24:02,714 
don't know if you have read this

782
00:24:02,752 --> 00:24:04,794 
paper called The Hidden Debt in

783
00:24:04,832 --> 00:24:06,378 
Machine Learning Systems. It's a

784
00:24:06,384 --> 00:24:07,930 
Google paper and it actually

785
00:24:08,000 --> 00:24:09,722 
talks about there's a very cool

786
00:24:09,776 --> 00:24:10,938 
diagram in there. I don't know

787
00:24:10,944 --> 00:24:12,460 
if you can screen share here,

788
00:24:13,390 --> 00:24:15,386 
but actually it's a diagram

789
00:24:15,418 --> 00:24:16,414 
which shows that in all these

790
00:24:16,452 --> 00:24:18,106 
different boxes, there's a tiny

791
00:24:18,138 --> 00:24:19,278 
box called Machine Learning, and

792
00:24:19,284 --> 00:24:20,606 
there's huge boxes around it

793
00:24:20,628 --> 00:24:22,560 
which are all essentially data.

794
00:24:22,930 --> 00:24:24,590 
I think it's a 2016 paper.

795
00:24:24,660 --> 00:24:27,618 
If I'm not, I'm looking it up.

796
00:24:27,784 --> 00:24:29,474 
Yeah. Hidden technical debt in

797
00:24:29,512 --> 00:24:31,554 
machine learning systems is the

798
00:24:31,592 --> 00:24:35,398 
paper. And there's a diagram on

799
00:24:35,404 --> 00:24:39,174 
the third page which shows that.

800
00:24:39,212 --> 00:24:42,230 
But it is interesting that even

801
00:24:42,300 --> 00:24:44,086 
at that time, people who were

802
00:24:44,108 --> 00:24:46,422 
working on these things were

803
00:24:46,476 --> 00:24:48,810 
seeing the same pattern.

804
00:24:50,110 --> 00:24:52,810 
Yes, I have seen this diagram.

805
00:24:57,230 --> 00:24:59,146 
This is something that comes up

806
00:24:59,168 --> 00:25:00,586 
in my day job quite a bit, where

807
00:25:00,608 --> 00:25:03,218 
we talk about how the machine

808
00:25:03,254 --> 00:25:04,670 
learning is only one part of it,

809
00:25:04,740 --> 00:25:07,886 
right? There's a whole lot that

810
00:25:07,908 --> 00:25:09,760 
has to go into that. So yeah,

811
00:25:10,370 --> 00:25:11,966 
and that's a good point. And I

812
00:25:11,988 --> 00:25:13,726 
think that everybody wants to be

813
00:25:13,748 --> 00:25:15,182 
the rock star, right? Everybody

814
00:25:15,236 --> 00:25:16,514 
wants to have their name up on

815
00:25:16,552 --> 00:25:19,202 
lights, but the amount of people

816
00:25:19,256 --> 00:25:20,594 
that goes to make that rock star

817
00:25:20,632 --> 00:25:22,818 
look good, there's a lot of

818
00:25:22,824 --> 00:25:25,966 
opportunity in there. And he's

819
00:25:25,998 --> 00:25:28,706 
been a guest on the show. He's

820
00:25:28,738 --> 00:25:30,066 
famous within a certain Internet

821
00:25:30,098 --> 00:25:32,040 
circle. John Lee Dumas. Right?

822
00:25:32,410 --> 00:25:34,934 
He has a phrase where I like

823
00:25:34,972 --> 00:25:37,046 
boring boring is good because no

824
00:25:37,068 --> 00:25:38,746 
one is competing to do the

825
00:25:38,768 --> 00:25:40,442 
boring stuff. Not that data

826
00:25:40,496 --> 00:25:42,778 
engineering is boring. I want to

827
00:25:42,784 --> 00:25:45,994 
head off the hate mail right

828
00:25:46,032 --> 00:25:48,298 
there. But no, I mean it's one

829
00:25:48,304 --> 00:25:51,782 
of those things where there's

830
00:25:51,846 --> 00:25:53,498 
enormous opportunity. If you

831
00:25:53,504 --> 00:25:57,326 
look that box, it's one part of

832
00:25:57,348 --> 00:25:59,358 
the the whole operation and.

833
00:25:59,444 --> 00:26:01,470 
There'S a tiny little ML code

834
00:26:01,540 --> 00:26:01,966 
box.

835
00:26:02,068 --> 00:26:03,978 
In the it takes out all the

836
00:26:04,004 --> 00:26:05,780 
oxygen in the room. But

837
00:26:06,150 --> 00:26:07,666 
realistically, in order to have

838
00:26:07,688 --> 00:26:08,866 
that little box, you need to

839
00:26:08,888 --> 00:26:11,602 
stand on a lot of other

840
00:26:11,656 --> 00:26:15,874 
operations. Right. Andy and I

841
00:26:15,912 --> 00:26:18,194 
kind of had this back and forth

842
00:26:18,242 --> 00:26:20,246 
and obviously data science is

843
00:26:20,268 --> 00:26:21,506 
important, blah, blah, blah. AI

844
00:26:21,538 --> 00:26:22,374 
is important, machine learning

845
00:26:22,412 --> 00:26:24,566 
is important, but it stands on

846
00:26:24,588 --> 00:26:28,150 
the shoulders of giants. Another

847
00:26:28,300 --> 00:26:30,520 
analogy I've seen where it shows

848
00:26:31,310 --> 00:26:32,666 
like a rocket, right, in a

849
00:26:32,688 --> 00:26:34,006 
little capsule that holds

850
00:26:34,038 --> 00:26:36,426 
people. But it's sitting on top

851
00:26:36,448 --> 00:26:38,186 
of a massive rocket, which of

852
00:26:38,208 --> 00:26:39,866 
course has the launch pad and

853
00:26:39,888 --> 00:26:41,820 
all the other accessories to it.

854
00:26:42,830 --> 00:26:44,046 
That's another way to look at

855
00:26:44,068 --> 00:26:44,158 
it.

856
00:26:44,164 --> 00:26:44,286 
Right.

857
00:26:44,308 --> 00:26:47,066 
It is crucial. And I think it's

858
00:26:47,098 --> 00:26:49,182 
a shame that we kind of not we,

859
00:26:49,236 --> 00:26:50,634 
but it doesn't get the attention

860
00:26:50,682 --> 00:26:51,754 
that it deserves.

861
00:26:51,882 --> 00:26:54,990 
Yeah, I think data engineering

862
00:26:55,330 --> 00:26:58,030 
is complex. It is also painful.

863
00:26:58,110 --> 00:26:59,458 
It is also something that has to

864
00:26:59,464 --> 00:27:00,814 
be done at a massive, massive

865
00:27:00,862 --> 00:27:02,690 
scale. And it's challenging. But

866
00:27:02,760 --> 00:27:03,986 
remember, it's a means to the

867
00:27:04,008 --> 00:27:05,806 
end and people get fascinated

868
00:27:05,838 --> 00:27:07,058 
about that end. That happens

869
00:27:07,144 --> 00:27:08,534 
eventually, but the means to the

870
00:27:08,572 --> 00:27:10,246 
end takes a lot of work. And

871
00:27:10,268 --> 00:27:11,574 
sometimes, to be candid, it

872
00:27:11,612 --> 00:27:13,960 
becomes tankless work. Because

873
00:27:15,210 --> 00:27:16,674 
why are we such a big proponent

874
00:27:16,722 --> 00:27:18,054 
of bringing automation into data

875
00:27:18,092 --> 00:27:19,026 
engineering work? You cannot

876
00:27:19,058 --> 00:27:19,814 
automate all of data

877
00:27:19,852 --> 00:27:21,014 
engineering, just to be clear.

878
00:27:21,132 --> 00:27:22,186 
But when you bring in

879
00:27:22,208 --> 00:27:23,306 
automation, you're saying that

880
00:27:23,328 --> 00:27:24,746 
if I'm running literally, I have

881
00:27:24,768 --> 00:27:25,866 
customers running thousands of

882
00:27:25,888 --> 00:27:27,306 
pipelines in our system, for

883
00:27:27,328 --> 00:27:29,818 
example. And you don't want to

884
00:27:29,824 --> 00:27:30,966 
be waken up on that Saturday

885
00:27:30,998 --> 00:27:32,154 
night because one of those is

886
00:27:32,192 --> 00:27:33,826 
not working. You want automation

887
00:27:33,878 --> 00:27:35,054 
in there. Right? You want that

888
00:27:35,092 --> 00:27:36,446 
system to work for you.

889
00:27:36,468 --> 00:27:38,702 
Otherwise all you get in data

890
00:27:38,756 --> 00:27:40,526 
engineering is a lot of lot of

891
00:27:40,548 --> 00:27:42,494 
work to do and complaints when

892
00:27:42,532 --> 00:27:44,606 
it doesn't work. But it is a

893
00:27:44,628 --> 00:27:46,734 
very key niece then. So I do

894
00:27:46,772 --> 00:27:48,146 
celebrate the work that they do.

895
00:27:48,168 --> 00:27:49,378 
And if you look at OpenAI, for

896
00:27:49,384 --> 00:27:50,434 
example, because it's such a hot

897
00:27:50,472 --> 00:27:51,966 
company right now, that billion

898
00:27:51,998 --> 00:27:54,002 
dollar plus in funding, very few

899
00:27:54,056 --> 00:27:55,746 
companies could have done what

900
00:27:55,768 --> 00:27:57,026 
they have done because they had

901
00:27:57,048 --> 00:27:58,418 
that kind of money okay.

902
00:27:58,504 --> 00:27:58,754 
Right.

903
00:27:58,792 --> 00:28:01,046 
To begin with. But I bet if you

904
00:28:01,068 --> 00:28:02,214 
look at how that money got

905
00:28:02,252 --> 00:28:04,166 
spent, I'm sure a big chunk is

906
00:28:04,188 --> 00:28:06,342 
in pipelines because of data and

907
00:28:06,396 --> 00:28:07,766 
processing that and moving that

908
00:28:07,788 --> 00:28:09,786 
around and we don't talk about

909
00:28:09,808 --> 00:28:12,374 
it. But the reason that Chat GPD

910
00:28:12,422 --> 00:28:14,170 
works so well is because it can

911
00:28:14,320 --> 00:28:17,866 
look into all of that data and

912
00:28:18,048 --> 00:28:19,466 
talk intelligently about it,

913
00:28:19,488 --> 00:28:20,060 
right?

914
00:28:20,610 --> 00:28:23,886 
No, absolutely. I don't know

915
00:28:23,908 --> 00:28:25,438 
when this switch happened, but

916
00:28:25,524 --> 00:28:28,174 
in terms of staffing, a.

917
00:28:28,212 --> 00:28:29,646 
Previous job, at the end of

918
00:28:29,668 --> 00:28:33,774 
2021, they wanted to do

919
00:28:33,812 --> 00:28:34,878 
something. Can't say what it

920
00:28:34,884 --> 00:28:36,146 
was, but they wanted to do

921
00:28:36,168 --> 00:28:37,460 
something. And they said, oh,

922
00:28:38,870 --> 00:28:40,546 
it'll be challenging to find the

923
00:28:40,568 --> 00:28:42,530 
data scientist for this job. And

924
00:28:42,680 --> 00:28:43,938 
my manager and I can look each

925
00:28:43,944 --> 00:28:45,346 
other. Actually, at this point

926
00:28:45,368 --> 00:28:46,514 
in time, it's going to be a lot

927
00:28:46,552 --> 00:28:47,774 
harder to find a data engineers

928
00:28:47,822 --> 00:28:48,918 
that you're going to need for

929
00:28:48,924 --> 00:28:49,046 
that.

930
00:28:49,068 --> 00:28:49,206 
Right.

931
00:28:49,228 --> 00:28:50,774 
Because they only really needed

932
00:28:50,812 --> 00:28:52,534 
two data scientists. But just

933
00:28:52,572 --> 00:28:54,034 
based on what the aggressive

934
00:28:54,082 --> 00:28:54,966 
thing that they were trying to

935
00:28:54,988 --> 00:28:59,014 
build, they would need, I would

936
00:28:59,212 --> 00:29:00,598 
just spitball and I would say a

937
00:29:00,604 --> 00:29:01,990 
dozen data engineers.

938
00:29:02,350 --> 00:29:04,038 
But from a technology provider

939
00:29:04,054 --> 00:29:05,238 
and a tool provider perspective,

940
00:29:05,254 --> 00:29:06,170 
I would say the interesting

941
00:29:06,240 --> 00:29:07,386 
thing about data engineering is

942
00:29:07,408 --> 00:29:09,674 
it is very complex, but the

943
00:29:09,712 --> 00:29:11,094 
challenges are very consistent.

944
00:29:11,142 --> 00:29:12,894 
I can look at our customers in

945
00:29:12,932 --> 00:29:14,686 
retail, like a BedBath, or

946
00:29:14,708 --> 00:29:16,494 
Forever 21, or in delivery like

947
00:29:16,532 --> 00:29:19,198 
DoorDash, or in Pharma like

948
00:29:19,284 --> 00:29:21,594 
Yansen JNJ, or in financial

949
00:29:21,642 --> 00:29:23,982 
services, or in cybersecurity,

950
00:29:24,126 --> 00:29:26,114 
all of them. The challenges at

951
00:29:26,152 --> 00:29:27,666 
fundamental level are very

952
00:29:27,768 --> 00:29:29,838 
similar, which is large amounts

953
00:29:29,854 --> 00:29:31,940 
of diverse heterogeneous data.

954
00:29:32,310 --> 00:29:33,938 
Being able to take that process

955
00:29:34,024 --> 00:29:35,534 
that do that, reliably detect

956
00:29:35,582 --> 00:29:37,378 
the issues, all of that stuff.

957
00:29:37,464 --> 00:29:40,158 
Have data quality monitoring,

958
00:29:40,334 --> 00:29:42,280 
make that data usable by people

959
00:29:42,890 --> 00:29:44,454 
scaling. All of those are very

960
00:29:44,492 --> 00:29:46,614 
similar. Which means that it

961
00:29:46,652 --> 00:29:47,926 
fits very nicely into the

962
00:29:47,948 --> 00:29:49,530 
traditional sort of problems

963
00:29:49,600 --> 00:29:51,142 
that can be solved by software,

964
00:29:51,206 --> 00:29:52,474 
problems that can be solved by

965
00:29:52,512 --> 00:29:55,420 
automation sort of model. Right.

966
00:29:56,350 --> 00:29:58,586 
So I think that is definitely a

967
00:29:58,608 --> 00:29:59,946 
part where the challenges are

968
00:29:59,968 --> 00:30:02,122 
not very unique to a certain

969
00:30:02,176 --> 00:30:03,118 
problem. And of course, there

970
00:30:03,124 --> 00:30:04,686 
are unique flavors to it. Some

971
00:30:04,708 --> 00:30:05,806 
have real time data, some have

972
00:30:05,828 --> 00:30:07,406 
data from devices, some have

973
00:30:07,428 --> 00:30:09,726 
data from legacy systems and so

974
00:30:09,748 --> 00:30:11,280 
on. But yeah, there is

975
00:30:12,130 --> 00:30:13,238 
structure.

976
00:30:13,434 --> 00:30:14,014 
Excellent.

977
00:30:14,062 --> 00:30:14,660 
Cool.

978
00:30:15,510 --> 00:30:17,966 
Very cool. All right, I'm

979
00:30:17,998 --> 00:30:19,678 
tweeting about Nexla.

980
00:30:19,854 --> 00:30:20,242 
Cool.

981
00:30:20,296 --> 00:30:20,962 
Thank you.

982
00:30:21,096 --> 00:30:22,020 
Right now.

983
00:30:24,150 --> 00:30:25,846 
And for all the stalkers, we are

984
00:30:25,868 --> 00:30:31,574 
recording this on April 12, just

985
00:30:31,612 --> 00:30:33,058 
FYI. So if you look at Andy's

986
00:30:33,074 --> 00:30:33,878 
feed and you're like, where is

987
00:30:33,884 --> 00:30:37,000 
the tweet? You have to go back.

988
00:30:38,250 --> 00:30:40,246 
So at this point in the show, we

989
00:30:40,268 --> 00:30:41,370 
want to switch to kind of the

990
00:30:41,440 --> 00:30:43,642 
pre found questions. And given

991
00:30:43,696 --> 00:30:46,282 
your background, the first one,

992
00:30:46,336 --> 00:30:47,178 
I really have to know the

993
00:30:47,184 --> 00:30:48,522 
answer, right? We always ask,

994
00:30:48,656 --> 00:30:50,042 
how did you find your way into

995
00:30:50,096 --> 00:30:52,410 
data? Did the data life find you

996
00:30:52,560 --> 00:30:54,160 
or did you find data?

997
00:30:54,530 --> 00:30:55,582 
I think it was happening

998
00:30:55,636 --> 00:30:57,934 
together, right? I guess so. My

999
00:30:57,972 --> 00:30:59,886 
decision to start a company, and

1000
00:30:59,988 --> 00:31:01,066 
I mentioned to you guys, I'd

1001
00:31:01,098 --> 00:31:03,386 
been at Nvidia, on the compute

1002
00:31:03,418 --> 00:31:05,614 
side of the world, really, and

1003
00:31:05,732 --> 00:31:07,178 
at some point I decided to start

1004
00:31:07,204 --> 00:31:08,866 
my own company. And when I was

1005
00:31:08,888 --> 00:31:10,818 
looking to do that in 2009, I

1006
00:31:10,824 --> 00:31:13,794 
was like, where do I go? Build a

1007
00:31:13,832 --> 00:31:15,186 
platform, if you will. And I

1008
00:31:15,208 --> 00:31:16,846 
felt that at the time, in 2009,

1009
00:31:16,888 --> 00:31:17,958 
apps were new and I'm going to

1010
00:31:17,964 --> 00:31:19,238 
build a monetization platform

1011
00:31:19,324 --> 00:31:21,106 
for app developers in the mobile

1012
00:31:21,138 --> 00:31:24,838 
space. That whole approach about

1013
00:31:25,004 --> 00:31:26,614 
building something around ad

1014
00:31:26,652 --> 00:31:28,006 
servers and one of the early ad

1015
00:31:28,028 --> 00:31:30,326 
servers in the mobile space, you

1016
00:31:30,348 --> 00:31:31,942 
realize that it's a very data

1017
00:31:31,996 --> 00:31:33,642 
driven world in advertising. And

1018
00:31:33,696 --> 00:31:35,306 
there is a reason why a lot of

1019
00:31:35,328 --> 00:31:36,538 
the data innovation, I would

1020
00:31:36,544 --> 00:31:37,926 
say, if you trace its roots,

1021
00:31:38,038 --> 00:31:39,354 
come from advertising. Whether

1022
00:31:39,392 --> 00:31:40,658 
it was Yahoo, whether it was

1023
00:31:40,684 --> 00:31:42,330 
Google, whether it was Facebook,

1024
00:31:43,310 --> 00:31:44,666 
what were these guys doing with

1025
00:31:44,688 --> 00:31:45,726 
data in the first place? They

1026
00:31:45,748 --> 00:31:47,374 
were dealing with a huge number

1027
00:31:47,412 --> 00:31:48,826 
of people visiting those pages

1028
00:31:48,858 --> 00:31:50,238 
and clicking on those ads, and

1029
00:31:50,244 --> 00:31:51,518 
they had to really figure out

1030
00:31:51,604 --> 00:31:53,006 
how to show the performance and

1031
00:31:53,028 --> 00:31:54,094 
say, which ad should you spend

1032
00:31:54,132 --> 00:31:55,406 
more money in, where should you

1033
00:31:55,428 --> 00:31:57,166 
not? And a lot of machine

1034
00:31:57,198 --> 00:31:58,546 
learning systems that we built

1035
00:31:58,648 --> 00:32:00,610 
early on back in 2011, twelve

1036
00:32:00,680 --> 00:32:02,510 
actually were for that purpose.

1037
00:32:02,670 --> 00:32:04,226 
We ran one of the largest ad

1038
00:32:04,248 --> 00:32:06,354 
auction systems at the time, and

1039
00:32:06,472 --> 00:32:07,774 
if you're running an automated

1040
00:32:07,822 --> 00:32:09,442 
ad auction with 15 billion

1041
00:32:09,506 --> 00:32:11,734 
auctions happening and 300

1042
00:32:11,772 --> 00:32:13,878 
billion bids on that, you have a

1043
00:32:13,884 --> 00:32:15,430 
lot of data. But you can also

1044
00:32:15,580 --> 00:32:17,366 
figure out that, hey, based on

1045
00:32:17,388 --> 00:32:20,594 
certain patterns, I can decide

1046
00:32:20,642 --> 00:32:21,978 
who to invite for an auction, I

1047
00:32:21,984 --> 00:32:23,274 
can decide what the floor price

1048
00:32:23,312 --> 00:32:24,634 
should be. And those were all

1049
00:32:24,672 --> 00:32:25,866 
machine learning systems. So I

1050
00:32:25,888 --> 00:32:28,314 
ended up actually building this

1051
00:32:28,352 --> 00:32:30,586 
advertising technology and

1052
00:32:30,608 --> 00:32:32,906 
system in those days to solve

1053
00:32:33,018 --> 00:32:34,606 
that developer problem of like,

1054
00:32:34,628 --> 00:32:36,654 
I'm building apps, how do I

1055
00:32:36,692 --> 00:32:38,606 
monetize it? But realizing that

1056
00:32:38,628 --> 00:32:40,400 
a whole chunk of it was data,

1057
00:32:41,330 --> 00:32:42,814 
and this was a lot of the data

1058
00:32:42,852 --> 00:32:45,110 
stuff that we did was pre kafka,

1059
00:32:45,290 --> 00:32:47,186 
even when big data hadoop was

1060
00:32:47,208 --> 00:32:48,706 
relatively early at the time, so

1061
00:32:48,728 --> 00:32:50,606 
the technologies were limited.

1062
00:32:50,798 --> 00:32:52,498 
Did a lot of homegrown stuff at

1063
00:32:52,504 --> 00:32:54,786 
the time, but realized that this

1064
00:32:54,808 --> 00:32:56,994 
is a massive problem. And I

1065
00:32:57,032 --> 00:32:59,638 
think data and I sort of met in

1066
00:32:59,644 --> 00:33:00,966 
that time, but I'm coming from

1067
00:33:00,988 --> 00:33:02,680 
this compute land. I'm building

1068
00:33:03,050 --> 00:33:04,626 
software for embedded systems

1069
00:33:04,658 --> 00:33:06,354 
where you are trained to squeeze

1070
00:33:06,402 --> 00:33:09,014 
every single kilobyte and every

1071
00:33:09,052 --> 00:33:10,406 
single ounce of performance in

1072
00:33:10,428 --> 00:33:11,418 
some of these systems. Like I

1073
00:33:11,424 --> 00:33:12,362 
mentioned, I was building

1074
00:33:12,416 --> 00:33:13,494 
software for the PlayStation

1075
00:33:13,542 --> 00:33:16,058 
Three when I was at Nvidia. And

1076
00:33:16,064 --> 00:33:17,386 
you come from that mindset of

1077
00:33:17,408 --> 00:33:19,386 
high performance squeezing the

1078
00:33:19,408 --> 00:33:20,906 
most of the system, and you see

1079
00:33:20,928 --> 00:33:22,506 
the data challenges. And I

1080
00:33:22,528 --> 00:33:24,226 
thought it sort of intersected

1081
00:33:24,278 --> 00:33:25,886 
nicely for me in sort of a

1082
00:33:25,908 --> 00:33:28,138 
developer or a product approach.

1083
00:33:28,314 --> 00:33:29,934 
And then we said, well, more

1084
00:33:29,972 --> 00:33:31,310 
people need to be using data.

1085
00:33:31,380 --> 00:33:32,286 
That's where the world is

1086
00:33:32,308 --> 00:33:33,518 
headed. Everybody is going to

1087
00:33:33,524 --> 00:33:35,022 
become a data user. I could see

1088
00:33:35,076 --> 00:33:37,442 
my second grade kid at the time,

1089
00:33:37,496 --> 00:33:39,266 
and they do simple survey in the

1090
00:33:39,288 --> 00:33:40,946 
class, and they created a

1091
00:33:40,968 --> 00:33:42,034 
histogram like, okay,

1092
00:33:42,072 --> 00:33:43,682 
everybody's going to be data

1093
00:33:43,736 --> 00:33:45,474 
user. How do you really get

1094
00:33:45,512 --> 00:33:47,138 
there? Not everybody's going to

1095
00:33:47,144 --> 00:33:48,706 
be technical and engineer. So

1096
00:33:48,728 --> 00:33:52,134 
that is kind of where my sort of

1097
00:33:52,172 --> 00:33:54,054 
direct experience in the data

1098
00:33:54,092 --> 00:33:55,126 
world started to happen, is

1099
00:33:55,148 --> 00:33:56,134 
like, we want to solve that

1100
00:33:56,172 --> 00:33:57,254 
problem. We want to make it

1101
00:33:57,292 --> 00:33:58,822 
possible for anybody to use

1102
00:33:58,876 --> 00:34:00,486 
data. And what is standing in

1103
00:34:00,508 --> 00:34:01,526 
their way is that data is

1104
00:34:01,548 --> 00:34:03,514 
complex. It's everywhere. It is

1105
00:34:03,552 --> 00:34:05,162 
hard to work with. Only

1106
00:34:05,216 --> 00:34:06,682 
developers are able to do that.

1107
00:34:06,736 --> 00:34:08,154 
So we're going to automate that.

1108
00:34:08,192 --> 00:34:09,306 
We're going to bring that and

1109
00:34:09,328 --> 00:34:10,986 
present that data to this user

1110
00:34:11,098 --> 00:34:12,286 
and they'll be able to use it

1111
00:34:12,308 --> 00:34:14,206 
wherever they want. And that was

1112
00:34:14,228 --> 00:34:15,582 
the driver for me.

1113
00:34:15,716 --> 00:34:20,606 
Wow, that's fascinating. Our

1114
00:34:20,628 --> 00:34:22,334 
next question is what's your

1115
00:34:22,372 --> 00:34:24,162 
favorite part of your current

1116
00:34:24,216 --> 00:34:24,820 
job?

1117
00:34:26,070 --> 00:34:30,194 
I think the CEO job and the co

1118
00:34:30,232 --> 00:34:33,326 
founder job is a new challenge,

1119
00:34:33,358 --> 00:34:36,514 
I would say every day and all

1120
00:34:36,552 --> 00:34:38,046 
sorts of unexpected things. I'm

1121
00:34:38,078 --> 00:34:39,286 
still a very product person at

1122
00:34:39,308 --> 00:34:41,174 
heart, so it's like those things

1123
00:34:41,212 --> 00:34:44,166 
are always fun to look at. But I

1124
00:34:44,188 --> 00:34:46,246 
would say no day is similar to

1125
00:34:46,268 --> 00:34:47,606 
the last one is the best part of

1126
00:34:47,628 --> 00:34:49,974 
it. I think about a month back,

1127
00:34:50,092 --> 00:34:51,226 
all of a sudden we were like,

1128
00:34:51,248 --> 00:34:52,518 
oh, the bank that we are banking

1129
00:34:52,534 --> 00:34:53,978 
with is going under and what do

1130
00:34:53,984 --> 00:34:59,674 
you do? Did I go in or did any

1131
00:34:59,712 --> 00:35:01,050 
of those CEOs go in on that

1132
00:35:01,120 --> 00:35:02,506 
Thursday or Friday morning to

1133
00:35:02,528 --> 00:35:03,578 
say, oh, this is what I'm going

1134
00:35:03,584 --> 00:35:04,718 
to do. I'm going to spend the

1135
00:35:04,724 --> 00:35:06,286 
whole weekend figuring out if we

1136
00:35:06,308 --> 00:35:07,518 
have any money as a company or

1137
00:35:07,524 --> 00:35:10,094 
we're out. All of a sudden the

1138
00:35:10,132 --> 00:35:11,518 
rug pulled underneath us. So I

1139
00:35:11,524 --> 00:35:13,610 
think that's the challenge. But

1140
00:35:13,620 --> 00:35:17,394 
that's also the fun of this

1141
00:35:17,432 --> 00:35:19,506 
particular role. I do enjoy the

1142
00:35:19,528 --> 00:35:20,594 
thought that we are doing some

1143
00:35:20,632 --> 00:35:23,506 
very cool stuff in data. And the

1144
00:35:23,528 --> 00:35:24,386 
number one source of

1145
00:35:24,408 --> 00:35:25,746 
satisfaction for me is when our

1146
00:35:25,768 --> 00:35:27,480 
customers come to us and say,

1147
00:35:29,050 --> 00:35:31,298 
this user of ours in this pharma

1148
00:35:31,314 --> 00:35:32,486 
industry, they're like, you know

1149
00:35:32,508 --> 00:35:34,150 
what, this data that we're using

1150
00:35:34,220 --> 00:35:36,482 
with you guys, it was processing

1151
00:35:36,546 --> 00:35:38,486 
in multiple hours and now it

1152
00:35:38,508 --> 00:35:39,718 
happens in nine minutes. I was

1153
00:35:39,724 --> 00:35:42,474 
like, wow, my goodness. And when

1154
00:35:42,512 --> 00:35:43,526 
I hear those kind of stories,

1155
00:35:43,558 --> 00:35:44,682 
I'm like, okay, we are actually

1156
00:35:44,736 --> 00:35:46,266 
delivering value. I don't know

1157
00:35:46,288 --> 00:35:48,586 
how to make a medicine or a

1158
00:35:48,608 --> 00:35:50,170 
medical device, but these guys

1159
00:35:50,240 --> 00:35:51,934 
who know how to do that, we are

1160
00:35:51,972 --> 00:35:53,486 
somehow enabling them to do

1161
00:35:53,508 --> 00:35:55,566 
their job better. And that is, I

1162
00:35:55,588 --> 00:35:56,574 
think, ultimately the

1163
00:35:56,612 --> 00:35:57,822 
satisfaction of the work.

1164
00:35:57,876 --> 00:35:58,574 
Right?

1165
00:35:58,772 --> 00:35:59,886 
Very cool.

1166
00:36:00,068 --> 00:36:03,120 
Interesting. So we have three

1167
00:36:04,370 --> 00:36:06,226 
complete the sentences. And the

1168
00:36:06,248 --> 00:36:07,554 
first one is when I'm not

1169
00:36:07,592 --> 00:36:09,730 
working, I enjoy blank.

1170
00:36:10,950 --> 00:36:13,074 
So many things. I love road

1171
00:36:13,112 --> 00:36:14,386 
biking. I do that a lot right

1172
00:36:14,408 --> 00:36:17,940 
now, but I love snowboarding and

1173
00:36:19,210 --> 00:36:21,218 
a bunch of activities. I don't

1174
00:36:21,234 --> 00:36:22,726 
do flying anymore actively, but

1175
00:36:22,748 --> 00:36:24,102 
that's another one I would do.

1176
00:36:24,236 --> 00:36:27,110 
Nice. Very cool. Our next

1177
00:36:27,180 --> 00:36:29,034 
complete the sentence. I think

1178
00:36:29,072 --> 00:36:30,554 
the coolest thing in technology

1179
00:36:30,672 --> 00:36:32,570 
today is blank.

1180
00:36:36,750 --> 00:36:38,314 
I would have to say that

1181
00:36:38,432 --> 00:36:40,026 
Generative AI is certainly one

1182
00:36:40,048 --> 00:36:41,322 
of the coolest things out there.

1183
00:36:41,376 --> 00:36:42,686 
I think I'm still trying to

1184
00:36:42,708 --> 00:36:44,042 
understand from a technical

1185
00:36:44,106 --> 00:36:45,902 
engineering perspective, like

1186
00:36:46,036 --> 00:36:47,966 
the ins and outs of it, but it

1187
00:36:47,988 --> 00:36:49,214 
is fascinating. It is also

1188
00:36:49,252 --> 00:36:52,154 
scary, to be honest. I wouldn't

1189
00:36:52,202 --> 00:36:53,300 
deny that either.

1190
00:36:53,670 --> 00:36:55,634 
Well, it's funny you mentioned

1191
00:36:55,672 --> 00:36:57,586 
that because I was thinking on

1192
00:36:57,608 --> 00:37:00,100 
that earlier today. Even. And

1193
00:37:01,110 --> 00:37:03,282 
the whole idea, the moving

1194
00:37:03,336 --> 00:37:05,654 
parts, when you start thinking

1195
00:37:05,692 --> 00:37:07,686 
about the image generation, just

1196
00:37:07,708 --> 00:37:10,646 
take a subset and you think

1197
00:37:10,668 --> 00:37:12,182 
about what goes into that. You

1198
00:37:12,236 --> 00:37:14,534 
describe something, so it has to

1199
00:37:14,572 --> 00:37:16,150 
understand what you describe,

1200
00:37:16,810 --> 00:37:19,366 
and that has that LLM component

1201
00:37:19,398 --> 00:37:22,874 
to it. Right. And then it

1202
00:37:22,912 --> 00:37:24,394 
interprets that in such a way

1203
00:37:24,432 --> 00:37:27,866 
and then probably tokenizes it,

1204
00:37:27,888 --> 00:37:29,194 
and then it generates this

1205
00:37:29,232 --> 00:37:32,238 
image. And I was reading a

1206
00:37:32,244 --> 00:37:35,726 
blurb, a quote from someone at

1207
00:37:35,748 --> 00:37:38,014 
Nvidia today, and that's what

1208
00:37:38,052 --> 00:37:40,814 
kind of got me off doing

1209
00:37:40,852 --> 00:37:42,926 
Billable work, mind you, and

1210
00:37:43,108 --> 00:37:44,890 
running down the rabbit hole.

1211
00:37:44,970 --> 00:37:47,074 
And I think it was a guy from

1212
00:37:47,112 --> 00:37:48,626 
Nvidia. And if it wasn't, then I

1213
00:37:48,648 --> 00:37:50,354 
apologize. But it was a person

1214
00:37:50,392 --> 00:37:52,414 
at Nvidia who made the statement

1215
00:37:52,542 --> 00:37:56,614 
that we're approaching that

1216
00:37:56,652 --> 00:37:57,750 
point where we're no longer

1217
00:37:57,820 --> 00:38:01,074 
rendering the pixels we're

1218
00:38:01,122 --> 00:38:03,862 
generating at the pixel level.

1219
00:38:03,916 --> 00:38:05,762 
So now they're rendering

1220
00:38:05,906 --> 00:38:08,982 
splotches of it. Yes, generated,

1221
00:38:09,046 --> 00:38:11,290 
but probably pulled in from

1222
00:38:11,360 --> 00:38:12,634 
someplace based on the

1223
00:38:12,672 --> 00:38:15,286 
description. A tokenized image

1224
00:38:15,318 --> 00:38:17,110 
from a tokenized description.

1225
00:38:17,270 --> 00:38:18,442 
But they're talking about

1226
00:38:18,496 --> 00:38:20,618 
generating the pixels suck in.

1227
00:38:20,784 --> 00:38:21,258 
Wow.

1228
00:38:21,344 --> 00:38:22,474 
I would like to understand how

1229
00:38:22,512 --> 00:38:23,686 
lighting because lighting has

1230
00:38:23,728 --> 00:38:25,006 
always been the biggest part in

1231
00:38:25,028 --> 00:38:27,534 
doing this, right? And how that

1232
00:38:27,572 --> 00:38:28,606 
applies to it. But I would say

1233
00:38:28,628 --> 00:38:30,174 
the reason I used to have

1234
00:38:30,212 --> 00:38:31,326 
unconditional love for

1235
00:38:31,348 --> 00:38:33,054 
technology innovation was ten

1236
00:38:33,092 --> 00:38:34,414 
years ago. Everything that is

1237
00:38:34,452 --> 00:38:36,114 
better is always or faster is

1238
00:38:36,152 --> 00:38:37,940 
better. But I would say that

1239
00:38:38,470 --> 00:38:41,182 
post social media and YouTube

1240
00:38:41,246 --> 00:38:42,898 
and all of the stuff have become

1241
00:38:42,984 --> 00:38:44,226 
a little bit concerned that we

1242
00:38:44,248 --> 00:38:45,634 
really have to understand what

1243
00:38:45,672 --> 00:38:47,378 
is this technology going to do?

1244
00:38:47,464 --> 00:38:48,854 
And nothing scares me more than

1245
00:38:48,892 --> 00:38:50,902 
that about genetic AI. Is that

1246
00:38:51,036 --> 00:38:52,566 
okay? It is a cool piece of

1247
00:38:52,588 --> 00:38:55,186 
innovation, but unlike a faster

1248
00:38:55,218 --> 00:38:56,326 
chip, which was almost a no

1249
00:38:56,348 --> 00:38:57,478 
brainer, I think now the

1250
00:38:57,484 --> 00:38:59,466 
question is like, oh, okay, what

1251
00:38:59,488 --> 00:39:00,758 
is it going to do that we can't

1252
00:39:00,774 --> 00:39:02,058 
even think about today?

1253
00:39:02,224 --> 00:39:04,362 
Yeah, I'm with you. And I get

1254
00:39:04,416 --> 00:39:08,218 
the hesitancy and the thinking

1255
00:39:08,384 --> 00:39:10,330 
part of our population calling

1256
00:39:10,400 --> 00:39:12,494 
for a moratorium, kind of a six

1257
00:39:12,532 --> 00:39:15,374 
month pause. Knowing what I know

1258
00:39:15,412 --> 00:39:17,326 
about geeks and engineers, even

1259
00:39:17,428 --> 00:39:18,574 
knowing what I know about me,

1260
00:39:18,612 --> 00:39:20,446 
that's not going to happen. I

1261
00:39:20,468 --> 00:39:21,920 
found a quote. It's from

1262
00:39:22,370 --> 00:39:25,760 
digitalnative Substac.com, and

1263
00:39:26,930 --> 00:39:28,386 
here's a quote from it. It's in

1264
00:39:28,408 --> 00:39:31,010 
a talk with Sequoia last week,

1265
00:39:31,160 --> 00:39:34,500 
nvidia CEO Jensen Huang said,

1266
00:39:34,870 --> 00:39:36,534 
every single pixel will be

1267
00:39:36,572 --> 00:39:38,514 
generated soon. Not rendered,

1268
00:39:38,642 --> 00:39:41,318 
generated. And that was from, I

1269
00:39:41,324 --> 00:39:43,766 
think, at the time of this

1270
00:39:43,788 --> 00:39:44,806 
recording. It's either the

1271
00:39:44,828 --> 00:39:46,614 
latest from Digital native or

1272
00:39:46,732 --> 00:39:48,550 
next to the latest.

1273
00:39:50,250 --> 00:39:51,706 
And he's a very smart guy, and

1274
00:39:51,728 --> 00:39:53,514 
he knows the stuff better than

1275
00:39:53,552 --> 00:39:55,126 
anybody out there. So if he's

1276
00:39:55,158 --> 00:39:56,762 
saying it, I believe it. Okay?

1277
00:39:56,816 --> 00:39:56,986 
Yeah.

1278
00:39:57,008 --> 00:39:59,738 
But I don't know if everybody

1279
00:39:59,824 --> 00:40:02,400 
gets how big of a leap that is

1280
00:40:02,850 --> 00:40:05,182 
to go from what we were doing

1281
00:40:05,236 --> 00:40:10,478 
before to generating pixels. I

1282
00:40:10,484 --> 00:40:12,478 
don't know. Maybe I'm making too

1283
00:40:12,484 --> 00:40:13,754 
much out of it, but it boggles

1284
00:40:13,802 --> 00:40:14,510 
my mind.

1285
00:40:14,660 --> 00:40:16,046 
At least the game logic and

1286
00:40:16,068 --> 00:40:17,266 
stuff is going there. I was

1287
00:40:17,288 --> 00:40:18,546 
reading something about how they

1288
00:40:18,568 --> 00:40:20,946 
put these AI agents in the game.

1289
00:40:20,968 --> 00:40:22,146 
In this paper that came out, I

1290
00:40:22,168 --> 00:40:23,202 
think, two or three days back,

1291
00:40:23,256 --> 00:40:25,346 
and they figured out to do a

1292
00:40:25,368 --> 00:40:26,706 
Valentine's Day party in there

1293
00:40:26,728 --> 00:40:27,954 
and invite people and all that

1294
00:40:27,992 --> 00:40:29,734 
sort of social mechanics were

1295
00:40:29,772 --> 00:40:30,598 
happening in that sort of

1296
00:40:30,604 --> 00:40:31,526 
generative way. So they're not

1297
00:40:31,548 --> 00:40:34,150 
pre programming these goodness.

1298
00:40:34,810 --> 00:40:36,182 
That was also sort of crazy

1299
00:40:36,236 --> 00:40:37,126 
fascinating, because when it

1300
00:40:37,148 --> 00:40:38,486 
means to game development and

1301
00:40:38,508 --> 00:40:39,766 
the gaming experience, you can

1302
00:40:39,788 --> 00:40:42,718 
have a truly sort of a multiplat

1303
00:40:42,754 --> 00:40:44,218 
storyline of any sort and you

1304
00:40:44,224 --> 00:40:45,366 
don't know what everybody's

1305
00:40:45,398 --> 00:40:46,266 
gaming experience is going to do

1306
00:40:46,288 --> 00:40:47,386 
different. I mean, there was a

1307
00:40:47,408 --> 00:40:49,290 
thing at a time when you would

1308
00:40:49,360 --> 00:40:50,506 
code those things and make

1309
00:40:50,528 --> 00:40:51,242 
people have a different

1310
00:40:51,296 --> 00:40:51,862 
experience.

1311
00:40:52,016 --> 00:40:54,206 
Right. One more thing and then

1312
00:40:54,228 --> 00:40:55,822 
we'll shut up. Do you remember

1313
00:40:55,876 --> 00:40:58,254 
that Black Mirror episode where

1314
00:40:58,372 --> 00:41:00,894 
the lady's social score was just

1315
00:41:00,932 --> 00:41:02,494 
crashing as she went to some

1316
00:41:02,532 --> 00:41:04,674 
gathering and by the time she

1317
00:41:04,712 --> 00:41:06,338 
got there, she couldn't get in

1318
00:41:06,504 --> 00:41:07,666 
because she didn't have a high

1319
00:41:07,688 --> 00:41:11,426 
enough social score? It's like

1320
00:41:11,528 --> 00:41:13,314 
we're getting to more and more

1321
00:41:13,352 --> 00:41:16,174 
to that point in some place.

1322
00:41:16,232 --> 00:41:16,982 
Of the world that already

1323
00:41:17,036 --> 00:41:17,750 
exists.

1324
00:41:22,330 --> 00:41:23,778 
Your next fill in the blank. I'm

1325
00:41:23,794 --> 00:41:24,214 
sorry.

1326
00:41:24,332 --> 00:41:26,710 
Sure. No, it's all good stuff.

1327
00:41:26,860 --> 00:41:28,278 
That's what makes this podcast I

1328
00:41:28,284 --> 00:41:29,298 
mean, dare I say, it makes this

1329
00:41:29,324 --> 00:41:30,230 
podcast look cool. But that's

1330
00:41:30,230 --> 00:41:30,794 
what makes this field

1331
00:41:30,832 --> 00:41:32,234 
interesting. Right. It's not

1332
00:41:32,272 --> 00:41:33,574 
just about the bits anymore.

1333
00:41:33,622 --> 00:41:36,378 
Right. There's social

1334
00:41:36,464 --> 00:41:38,586 
connotations. Now, I used to

1335
00:41:38,608 --> 00:41:40,954 
also be unquestioned fan of

1336
00:41:40,992 --> 00:41:42,294 
technology. It makes everything

1337
00:41:42,352 --> 00:41:43,406 
better, it's going to solve the

1338
00:41:43,428 --> 00:41:45,374 
problems. And here we are some

1339
00:41:45,412 --> 00:41:47,166 
ten years later. It's like we

1340
00:41:47,188 --> 00:41:48,158 
actually created a whole bunch

1341
00:41:48,164 --> 00:41:53,218 
of new problems. And I don't

1342
00:41:53,224 --> 00:41:55,186 
know, is that maturity? I'm ten

1343
00:41:55,208 --> 00:41:57,620 
years older, or is that kind of

1344
00:41:58,550 --> 00:42:01,554 
the state of the technology that

1345
00:42:01,592 --> 00:42:05,234 
we have created? No, I'll let

1346
00:42:05,272 --> 00:42:06,546 
the philosophers debate that

1347
00:42:06,568 --> 00:42:07,140 
one.

1348
00:42:08,070 --> 00:42:09,494 
I think it's evolution because

1349
00:42:09,532 --> 00:42:10,994 
we went from building the basic

1350
00:42:11,042 --> 00:42:12,006 
infrastructure, which is like

1351
00:42:12,028 --> 00:42:13,526 
chips and compute and stuff, and

1352
00:42:13,548 --> 00:42:15,586 
those things are but we didn't

1353
00:42:15,618 --> 00:42:16,998 
see that end application to the

1354
00:42:17,004 --> 00:42:18,166 
level that we are seeing now. So

1355
00:42:18,188 --> 00:42:20,454 
it's better building technology.

1356
00:42:20,572 --> 00:42:21,866 
Bricks, cement, all that is

1357
00:42:21,888 --> 00:42:22,986 
good, but suddenly cities are

1358
00:42:23,008 --> 00:42:24,794 
built that look crazy and

1359
00:42:24,832 --> 00:42:25,914 
whatever, and you're like, oh,

1360
00:42:25,952 --> 00:42:27,114 
is that what we're trying to do

1361
00:42:27,152 --> 00:42:30,666 
here? So I think it's just

1362
00:42:30,688 --> 00:42:31,854 
applying technology to every

1363
00:42:31,892 --> 00:42:32,800 
kind of problem.

1364
00:42:33,730 --> 00:42:35,886 
Yeah, no, absolutely. All right.

1365
00:42:35,908 --> 00:42:37,454 
And our third and final complete

1366
00:42:37,492 --> 00:42:39,326 
the sentence I look forward to

1367
00:42:39,348 --> 00:42:40,766 
today when I can use technology

1368
00:42:40,868 --> 00:42:42,030 
to blank.

1369
00:42:46,050 --> 00:42:49,266 
I think that I would say, like,

1370
00:42:49,288 --> 00:42:50,498 
the self driving car. I think it

1371
00:42:50,504 --> 00:42:51,826 
would save me a good bunch of

1372
00:42:51,848 --> 00:42:53,054 
time if it really becomes

1373
00:42:53,102 --> 00:42:54,286 
reliable. I don't know if I'll

1374
00:42:54,318 --> 00:42:55,380 
trust it, but.

1375
00:42:57,370 --> 00:42:59,010 
Why is it that all the engineers

1376
00:42:59,090 --> 00:43:00,722 
are suspicious of self driving

1377
00:43:00,786 --> 00:43:01,510 
cars?

1378
00:43:04,170 --> 00:43:05,686 
I'll be candid with you. At some

1379
00:43:05,708 --> 00:43:06,578 
level I feel like when I'm

1380
00:43:06,594 --> 00:43:08,134 
driving the car, it's not that

1381
00:43:08,172 --> 00:43:09,298 
much work. And I'm sitting

1382
00:43:09,314 --> 00:43:10,746 
there, I might as well just have

1383
00:43:10,768 --> 00:43:11,786 
my hand on, because most of the

1384
00:43:11,808 --> 00:43:13,786 
stuff is got an automated like

1385
00:43:13,968 --> 00:43:16,022 
those adaptive cruises and lane

1386
00:43:16,086 --> 00:43:17,706 
management and stuff. So, yeah,

1387
00:43:17,808 --> 00:43:18,602 
that much work.

1388
00:43:18,656 --> 00:43:22,302 
But yeah, adaptive cruise

1389
00:43:22,356 --> 00:43:25,886 
control is not self driving in

1390
00:43:25,908 --> 00:43:27,358 
the purest sense, but I will

1391
00:43:27,364 --> 00:43:29,902 
tell you, I can't live without

1392
00:43:29,956 --> 00:43:32,894 
it. Now, my car, this is first

1393
00:43:32,932 --> 00:43:34,350 
world problems, right? My car

1394
00:43:34,420 --> 00:43:38,226 
will break all the way to 0.

1395
00:43:38,248 --> 00:43:39,602 
Wife's car will stop at around

1396
00:43:39,656 --> 00:43:42,766 
22 when I'm stuck in a traffic

1397
00:43:42,798 --> 00:43:44,338 
jam in my car. I wouldn't say

1398
00:43:44,344 --> 00:43:46,226 
it's no big deal, but I can just

1399
00:43:46,248 --> 00:43:48,294 
kind of sit back and let the car

1400
00:43:48,332 --> 00:43:50,546 
handle the braking in my wife's

1401
00:43:50,578 --> 00:43:52,294 
car when it goes below a certain

1402
00:43:52,332 --> 00:43:53,846 
speed. Now I'm on the hook. It

1403
00:43:53,868 --> 00:43:55,366 
actually is annoying, which is

1404
00:43:55,388 --> 00:44:01,850 
pretty funny. But next question.

1405
00:44:01,920 --> 00:44:02,582 
Andy?

1406
00:44:02,726 --> 00:44:04,586 
I'm in. Yeah, I was going to

1407
00:44:04,608 --> 00:44:05,654 
interrupt you there, Frank.

1408
00:44:05,702 --> 00:44:07,386 
Before you yell at you to put

1409
00:44:07,408 --> 00:44:09,180 
down the shovel and climb out,

1410
00:44:12,770 --> 00:44:14,142 
share something different about

1411
00:44:14,196 --> 00:44:16,298 
yourself. Socket. But we remind

1412
00:44:16,394 --> 00:44:19,806 
our interviewees that it's a

1413
00:44:19,828 --> 00:44:21,758 
family podcast, so we want to

1414
00:44:21,764 --> 00:44:24,158 
keep our clean ratings. So

1415
00:44:24,324 --> 00:44:25,710 
something different about you.

1416
00:44:25,780 --> 00:44:27,380 
And you already mentioned a few.

1417
00:44:28,790 --> 00:44:32,526 
Yeah, from a family perspective,

1418
00:44:32,638 --> 00:44:34,594 
I'm a dad of three kids, two

1419
00:44:34,632 --> 00:44:37,766 
boys who are 13 and ten and a

1420
00:44:37,788 --> 00:44:41,046 
daughter who is six. And there

1421
00:44:41,068 --> 00:44:43,622 
is definitely a joy to seeing

1422
00:44:43,676 --> 00:44:45,526 
all of that happen. So, yeah,

1423
00:44:45,708 --> 00:44:47,126 
I'm a pretty regular individual

1424
00:44:47,228 --> 00:44:49,900 
in that way, but definitely

1425
00:44:51,150 --> 00:44:54,762 
bitten by the desire to build.

1426
00:44:54,816 --> 00:44:56,266 
And I'm like, I see a problem, I

1427
00:44:56,288 --> 00:44:57,526 
have to solve it. And that's

1428
00:44:57,558 --> 00:44:59,162 
kind of what landed me in this

1429
00:44:59,296 --> 00:45:00,958 
entrepreneur boat. Okay.

1430
00:45:01,044 --> 00:45:02,720 
Yeah, that's awesome.

1431
00:45:04,930 --> 00:45:09,470 
Awesome. And the next question,

1432
00:45:09,540 --> 00:45:10,622 
although technically speaking,

1433
00:45:10,676 --> 00:45:11,966 
it's out of order, so I need to

1434
00:45:11,988 --> 00:45:14,802 
fix that. Audible sponsors data

1435
00:45:14,856 --> 00:45:17,342 
driven. Do you do audiobooks?

1436
00:45:17,486 --> 00:45:19,586 
And if so, can you recommend a

1437
00:45:19,608 --> 00:45:21,266 
good one? And if you don't do

1438
00:45:21,288 --> 00:45:23,394 
audiobooks, any book

1439
00:45:23,432 --> 00:45:24,740 
recommendation will do.

1440
00:45:25,670 --> 00:45:27,458 
I do actually love them. That's

1441
00:45:27,474 --> 00:45:28,886 
where the driving part comes in,

1442
00:45:28,908 --> 00:45:30,486 
right? I mean, that's the way to

1443
00:45:30,668 --> 00:45:32,226 
to make the most of your driving

1444
00:45:32,258 --> 00:45:35,366 
time is to be listening into a

1445
00:45:35,388 --> 00:45:38,742 
book. I do a lot of technical

1446
00:45:38,806 --> 00:45:40,666 
stuff, actually, I think the

1447
00:45:40,688 --> 00:45:45,180 
most recent book that I was

1448
00:45:45,630 --> 00:45:48,986 
listening into was I'm just

1449
00:45:49,008 --> 00:45:50,082 
looking back at my bookshelf

1450
00:45:50,086 --> 00:45:51,114 
because always get a physical

1451
00:45:51,162 --> 00:45:52,240 
copy as well.

1452
00:45:54,290 --> 00:45:56,000 
I do the same thing.

1453
00:45:56,610 --> 00:46:01,754 
Yeah, I think my most recent

1454
00:46:01,802 --> 00:46:05,278 
book that I really enjoyed was I

1455
00:46:05,284 --> 00:46:06,978 
think the book called Sapiens. I

1456
00:46:06,984 --> 00:46:08,114 
don't know if you have read yes.

1457
00:46:08,152 --> 00:46:09,250 
Oh, I've heard of that.

1458
00:46:09,320 --> 00:46:11,650 
Yeah, it is an older book, but

1459
00:46:11,800 --> 00:46:14,210 
yeah, I got to it more recently.

1460
00:46:14,710 --> 00:46:15,526 
I've got it.

1461
00:46:15,548 --> 00:46:18,514 
Haven't read it yet. That author

1462
00:46:18,562 --> 00:46:20,246 
wrote something else that I

1463
00:46:20,268 --> 00:46:24,054 
read. Was it the guns? Germs and

1464
00:46:24,092 --> 00:46:25,720 
steel or something like that?

1465
00:46:28,570 --> 00:46:30,038 
That's an amazing book. It's a

1466
00:46:30,044 --> 00:46:31,354 
difficult read, I would say.

1467
00:46:31,392 --> 00:46:32,362 
That'S why it is.

1468
00:46:32,496 --> 00:46:34,474 
We can sit back and listen to

1469
00:46:34,512 --> 00:46:34,714 
it.

1470
00:46:34,752 --> 00:46:37,210 
Yeah, crank that dude up to 1.25

1471
00:46:37,280 --> 00:46:38,330 
and let it go.

1472
00:46:38,480 --> 00:46:41,342 
Yeah, exactly. I actually just

1473
00:46:41,396 --> 00:46:43,934 
finished The Wolf of Wall

1474
00:46:43,972 --> 00:46:46,906 
Street, the Abridged version,

1475
00:46:47,018 --> 00:46:49,566 
and I'd seen the movie, and I

1476
00:46:49,588 --> 00:46:50,734 
follow a lot of the other things

1477
00:46:50,772 --> 00:46:52,766 
that Jordan Belford had kind of

1478
00:46:52,788 --> 00:46:55,026 
done. But after listening to the

1479
00:46:55,048 --> 00:46:56,366 
story, there's a lot that didn't

1480
00:46:56,398 --> 00:46:58,322 
make it into the movie. And the

1481
00:46:58,376 --> 00:47:00,354 
impression I'm left with was

1482
00:47:00,552 --> 00:47:02,274 
truth really is stranger than

1483
00:47:02,312 --> 00:47:06,082 
fiction. I think they didn't put

1484
00:47:06,136 --> 00:47:07,686 
some of that stuff in because no

1485
00:47:07,708 --> 00:47:11,526 
one would have believed it. And

1486
00:47:11,548 --> 00:47:12,726 
if you've seen the movie, it's a

1487
00:47:12,748 --> 00:47:14,214 
pretty wild story. Anyway. So

1488
00:47:14,252 --> 00:47:16,182 
there's like even crazier stuff

1489
00:47:16,236 --> 00:47:16,758 
in there.

1490
00:47:16,844 --> 00:47:17,762 
Oh, my goodness.

1491
00:47:17,906 --> 00:47:19,770 
Yeah. So it's hard to imagine

1492
00:47:21,150 --> 00:47:21,658 
there's a.

1493
00:47:21,664 --> 00:47:23,334 
Second edition of The Black Swan

1494
00:47:23,382 --> 00:47:23,978 
out.

1495
00:47:24,144 --> 00:47:24,906 
Oh, really?

1496
00:47:25,008 --> 00:47:26,474 
Yeah, it's got a few extra stuff

1497
00:47:26,512 --> 00:47:29,066 
in it. There's definitely an

1498
00:47:29,088 --> 00:47:30,366 
appendix, I think a brand new

1499
00:47:30,388 --> 00:47:31,918 
appendix at the end where he

1500
00:47:31,924 --> 00:47:33,214 
talks about implications and

1501
00:47:33,252 --> 00:47:35,694 
applications. But there's new

1502
00:47:35,732 --> 00:47:36,846 
stuff all the way. Even in the

1503
00:47:36,868 --> 00:47:39,054 
introduction. He added some

1504
00:47:39,092 --> 00:47:43,086 
nassim. Nicholas Tilleb and I

1505
00:47:43,108 --> 00:47:44,334 
believe the first book of his

1506
00:47:44,372 --> 00:47:47,780 
encerta. So very interesting

1507
00:47:48,230 --> 00:47:49,540 
listening to that.

1508
00:47:50,870 --> 00:47:52,798 
I have to pick up the New Blacks

1509
00:47:52,814 --> 00:47:53,746 
one. That'll be interesting.

1510
00:47:53,848 --> 00:47:55,846 
Yeah. As a product person, I

1511
00:47:55,868 --> 00:47:58,006 
actually love also listening to

1512
00:47:58,028 --> 00:48:00,278 
a lot of these books on the

1513
00:48:00,284 --> 00:48:01,846 
business side of things. So I

1514
00:48:01,868 --> 00:48:02,902 
was listening to this book

1515
00:48:02,956 --> 00:48:04,662 
called The Ultimate Sales

1516
00:48:04,716 --> 00:48:08,822 
Machine maybe a couple of months

1517
00:48:08,876 --> 00:48:10,354 
back. And what I really enjoyed

1518
00:48:10,402 --> 00:48:12,066 
was that while you're listening

1519
00:48:12,098 --> 00:48:13,466 
to these books in the car and

1520
00:48:13,488 --> 00:48:14,906 
you have this free time, which

1521
00:48:14,928 --> 00:48:16,186 
is so hard to get, by the way,

1522
00:48:16,208 --> 00:48:19,034 
these days, is that gives me the

1523
00:48:19,072 --> 00:48:21,994 
space to think about things and

1524
00:48:22,032 --> 00:48:23,306 
reconstruct my ideas and

1525
00:48:23,328 --> 00:48:24,606 
thoughts. Most of the ideas that

1526
00:48:24,628 --> 00:48:26,506 
I get about maybe writing a blog

1527
00:48:26,538 --> 00:48:28,062 
post and stuff like that happen

1528
00:48:28,196 --> 00:48:29,786 
typically in the car, listening

1529
00:48:29,818 --> 00:48:31,006 
to something that triggers some

1530
00:48:31,028 --> 00:48:32,994 
idea in my head and do that

1531
00:48:33,032 --> 00:48:34,146 
because otherwise the whole day

1532
00:48:34,168 --> 00:48:35,774 
is like meetings and chasing

1533
00:48:35,822 --> 00:48:36,866 
stuff and whatever.

1534
00:48:37,048 --> 00:48:37,780 
Sure.

1535
00:48:39,030 --> 00:48:43,586 
Awesome. That's a good point.

1536
00:48:43,688 --> 00:48:46,482 
It's interesting how it always

1537
00:48:46,536 --> 00:48:47,626 
good to stretch your brain,

1538
00:48:47,678 --> 00:48:49,960 
because our brain is pretty much

1539
00:48:50,970 --> 00:48:53,558 
I'm not physically fit, but I

1540
00:48:53,564 --> 00:48:54,678 
like to think my brain is, at

1541
00:48:54,684 --> 00:48:57,320 
least. It's just good to kind of

1542
00:48:57,690 --> 00:48:59,266 
I get a lot out of just walking

1543
00:48:59,298 --> 00:49:00,774 
around or driving and kind of

1544
00:49:00,812 --> 00:49:02,006 
thinking about things and

1545
00:49:02,028 --> 00:49:03,226 
listening to books that are kind

1546
00:49:03,248 --> 00:49:05,146 
of outside my norm. Right. Which

1547
00:49:05,168 --> 00:49:06,266 
is probably why I picked up The

1548
00:49:06,288 --> 00:49:08,086 
Wolf of Wall Street because I'd

1549
00:49:08,118 --> 00:49:09,206 
been doing so many technical

1550
00:49:09,238 --> 00:49:10,358 
books and kind of sales books,

1551
00:49:10,374 --> 00:49:12,394 
and I was like, Let me check

1552
00:49:12,432 --> 00:49:13,200 
that out.

1553
00:49:14,610 --> 00:49:16,414 
Sometimes reading something or

1554
00:49:16,532 --> 00:49:17,694 
listening to something in one

1555
00:49:17,732 --> 00:49:18,718 
context, which is completely

1556
00:49:18,804 --> 00:49:20,026 
different, suddenly starts

1557
00:49:20,058 --> 00:49:21,086 
connecting to all the stuff that

1558
00:49:21,108 --> 00:49:22,880 
we are doing, like day to day.

1559
00:49:24,150 --> 00:49:25,666 
Absolutely. Yeah.

1560
00:49:25,688 --> 00:49:26,980 
I totally get that.

1561
00:49:28,630 --> 00:49:33,362 
Well, cool. So with that, is

1562
00:49:33,416 --> 00:49:36,146 
there anything where people can

1563
00:49:36,168 --> 00:49:38,580 
find you? I know@nextla.com.

1564
00:49:39,270 --> 00:49:41,682 
Yeah. next.com. And on LinkedIn,

1565
00:49:41,746 --> 00:49:43,846 
actually, I do engage with a lot

1566
00:49:43,868 --> 00:49:45,174 
of folks in conversations over

1567
00:49:45,212 --> 00:49:46,134 
there as well.

1568
00:49:46,252 --> 00:49:47,170 
Excellent.

1569
00:49:47,330 --> 00:49:48,806 
Awesome. Well, thank you for

1570
00:49:48,828 --> 00:49:50,006 
being on the show. And we'll let

1571
00:49:50,028 --> 00:49:52,454 
Bailey wrap us up now.

1572
00:49:52,492 --> 00:49:54,566 
That was some show. Is it me, or

1573
00:49:54,588 --> 00:49:56,406 
are the shows getting better? It

1574
00:49:56,428 --> 00:49:58,006 
could be my bias that leads me

1575
00:49:58,028 --> 00:49:59,686 
to say that. But I figured I

1576
00:49:59,708 --> 00:50:01,546 
would ask to get more input.

1577
00:50:01,738 --> 00:50:03,982 
After all, what's an AI without

1578
00:50:04,036 --> 00:50:06,042 
good input and a feedback loop?

1579
00:50:06,186 --> 00:50:07,966 
Speaking of feedback, have you

1580
00:50:07,988 --> 00:50:09,594 
checked out Data Driven Magazine

1581
00:50:09,642 --> 00:50:09,740 
yet?