1
00:00:00,000 --> 00:00:04,337
Audrow Nash: Hi everyone, Audrow
here. Isn't it amazing how far

2
00:00:04,337 --> 00:00:08,743
Tesla gets in their self driving
by leveraging vision. Vision is

3
00:00:08,743 --> 00:00:13,149
a powerful sense. There's a lot
information there. But its a lot

4
00:00:13,149 --> 00:00:17,141
of work. From designing the
hardware, to routing the data,

5
00:00:17,141 --> 00:00:20,928
to running AI and algorithms,
and to syncing with other

6
00:00:20,928 --> 00:00:25,127
sensors. That's where today's
guest Luxonis comes in. Luxonis

7
00:00:25,127 --> 00:00:29,257
makes vision platforms that
solve a lot of the hard problems

8
00:00:29,257 --> 00:00:33,319
I mentioned. And increasingly,
they're handling more of the

9
00:00:33,319 --> 00:00:37,449
on-device compute, with vision
sensors and compute being put

10
00:00:37,449 --> 00:00:41,236
together in one device. This is
super exciting for edge

11
00:00:41,236 --> 00:00:45,710
applications like most robots. I
talk with Luxonis’s CEO, Bradley

12
00:00:45,710 --> 00:00:49,909
Dillon. You'll like this episode
if you’re - someone building

13
00:00:49,909 --> 00:00:54,040
robots on your own - or part of
a company that has robots in

14
00:00:54,040 --> 00:00:58,101
places where seeing would be
valuable. Or if you're curious

15
00:00:58,101 --> 00:01:02,163
about the state of hardware in
robotics. This is the second

16
00:01:02,163 --> 00:01:06,362
full interview that I've done
with Bradley. It's very cool to

17
00:01:06,362 --> 00:01:10,630
see how much progress they've
made in the last couple years. I

18
00:01:10,630 --> 00:01:14,554
think Luxonis is making a great
platform, and that having

19
00:01:14,554 --> 00:01:18,891
companies like Luxonis makes it
easier to just get something of

20
00:01:18,891 --> 00:01:23,159
the shelf that works to build
your application. Without having

21
00:01:23,159 --> 00:01:27,152
to be an expert in cameras and
all the related engineering

22
00:01:27,152 --> 00:01:31,214
work. This isn't dissimilar in
my mind to using a great and

23
00:01:31,214 --> 00:01:35,620
developed software library like
Eigen for linear algebra, tensor

24
00:01:35,620 --> 00:01:39,750
flow for machine learning, or
Three.js for 3d visualization.

25
00:01:39,750 --> 00:01:43,812
You can just use them without
solving all of the underlying

26
00:01:43,812 --> 00:01:47,047
hard problems they solve.
Anyway, make yourself

27
00:01:47,047 --> 00:01:50,627
comfortable. And here's my
conversation with Bradley

28
00:01:50,627 --> 00:01:54,000
Dillon. Hi Bradley, would you
introduce yourself?

29
00:01:54,000 --> 00:01:57,536
Bradley Dillon: Hi, my name is
Bradley Dillon. I'm the CEO at

30
00:01:57,536 --> 00:01:58,000
Luxonis.

31
00:01:58,000 --> 00:02:01,754
Audrow Nash: Awesome. Tell me a
bit about Luxonis. What's the

32
00:02:01,754 --> 00:02:05,631
mission? How large are you guys?
How, long. Is the company been

33
00:02:05,631 --> 00:02:06,000
going?

34
00:02:06,000 --> 00:02:09,536
Bradley Dillon: Yeah, great.
Great question. So we were

35
00:02:09,536 --> 00:02:13,587
founded back in 2018. So we're
coming up on our sixth birthday,

36
00:02:13,587 --> 00:02:17,574
and what we do is we provide,
artificial intelligence enabled,

37
00:02:17,574 --> 00:02:21,431
cameras. And what these cameras
do is they help, things like

38
00:02:21,431 --> 00:02:24,968
robots, be able to see and
navigate the world. So we're

39
00:02:24,968 --> 00:02:28,890
trying to be the basically the
eyes and the ears in the brain

40
00:02:28,890 --> 00:02:32,876
behind these robots. So you can
basically think of us as being

41
00:02:32,876 --> 00:02:36,670
kind of a visual cortex. And,
you know, for us and our, you

42
00:02:36,670 --> 00:02:40,592
know, startup journey, we have
scaled up a lot. So we've sold

43
00:02:40,592 --> 00:02:44,000
more than 150,000 of these,
these AI enabled cameras.

44
00:02:44,000 --> 00:02:45,000
Audrow Nash: Amazing.

45
00:02:45,000 --> 00:02:48,675
Bradley Dillon: Yeah, it's been
quite, quite, quite the journey

46
00:02:48,675 --> 00:02:52,175
for sure. And we're up to a team
of about 80 people. And or,

47
00:02:52,175 --> 00:02:55,675
sorry, 100 people. And 80 of
those people are our engineers.

48
00:02:55,675 --> 00:02:59,000
And yeah, that's kind of a brief
overview of the company.

49
00:02:59,000 --> 00:03:02,810
Audrow Nash: Now, you said six
years. Tell me about kind of the

50
00:03:02,810 --> 00:03:06,560
trajectory of the company and
kind of how things are changing.

51
00:03:06,560 --> 00:03:10,069
Like your focus, I suppose, how
it's changed over the last

52
00:03:10,069 --> 00:03:13,395
years, because I know you guys
are going in an exciting

53
00:03:13,395 --> 00:03:14,000
direction.

54
00:03:14,000 --> 00:03:17,512
Bradley Dillon: So things have
definitely changed a lot in the

55
00:03:17,512 --> 00:03:21,195
first few years as a company and
not really that very like 0 to 1

56
00:03:21,195 --> 00:03:24,254
time frame. It was all about
being able to to nail the

57
00:03:24,254 --> 00:03:27,767
hardware and firmware. And so we
got off the ground by doing a

58
00:03:27,767 --> 00:03:31,336
couple of Kickstarter campaigns.
And so we were able to kind of

59
00:03:31,336 --> 00:03:34,735
crowdfund, our idea and the idea
was, is that you could have

60
00:03:34,735 --> 00:03:37,851
this, this low cost and
performant, you know, 3D stereo

61
00:03:37,851 --> 00:03:41,250
camera, you know, that could
help, you know, robots navigate

62
00:03:41,250 --> 00:03:44,593
and see in the world and that
you could do it in a way that

63
00:03:44,593 --> 00:03:48,048
could actually have, you know,
artificial intelligence on it.

64
00:03:48,048 --> 00:03:51,391
So when we were, in that first
few year period with a small

65
00:03:51,391 --> 00:03:54,847
team, you know, 5 to 10 people
and the biggest obstacles were

66
00:03:54,847 --> 00:03:58,189
really on the firmware side of
things. So first off, it was

67
00:03:58,189 --> 00:04:01,645
getting to, you know, being able
to abstract the hardware, to

68
00:04:01,645 --> 00:04:04,988
make it so that you could have
meaningful software that our

69
00:04:04,988 --> 00:04:08,274
customers could use. And then
the next part of our journey

70
00:04:08,274 --> 00:04:11,559
was, how the heck do you do
manufacturing? And we drew up,

71
00:04:11,559 --> 00:04:15,072
for the first time, you know,
procuring and building products,

72
00:04:15,072 --> 00:04:17,791
in the midst of probably the
craziest, you know,

73
00:04:17,791 --> 00:04:21,360
semiconductor electronic supply
shortage and, and Sandy B that,

74
00:04:21,360 --> 00:04:24,816
you know, the world is seem a
century. And so that was during

75
00:04:24,816 --> 00:04:28,442
kind of in the post Covid times.
And so we had to figure out how

76
00:04:28,442 --> 00:04:31,785
do you actually build, you know,
tens of thousands of these

77
00:04:31,785 --> 00:04:35,467
devices in a, in a reliable way.
So that was really the first few

78
00:04:35,467 --> 00:04:38,753
years of the journey. And in the
the last few years of the

79
00:04:38,753 --> 00:04:42,152
journey, what we've done is
we've started to build on top of

80
00:04:42,152 --> 00:04:45,778
that backbone, the hardware and
firmware being our backbone. And

81
00:04:45,778 --> 00:04:49,120
so that means as we started to,
really work with businesses

82
00:04:49,120 --> 00:04:52,576
tackling, you know, specific,
industry specific use cases and

83
00:04:52,576 --> 00:04:56,202
get it to where these, these our
type of equipment is able to be

84
00:04:56,202 --> 00:04:59,375
deployed at scale. And so
initially we were more focused

85
00:04:59,375 --> 00:05:02,830
on kind of a hobbyist, consumer
hobbyist use case. And that's

86
00:05:02,830 --> 00:05:05,776
how we kind of initially
started. But now our entire

87
00:05:05,776 --> 00:05:09,232
focus is on, you know, the big,
business to business, type of

88
00:05:09,232 --> 00:05:12,461
applications that are out there.
So the companies you can

89
00:05:12,461 --> 00:05:16,031
imagine, we've had a transition
to where, you know, things like

90
00:05:16,031 --> 00:05:19,090
reliability and stability,
become really important. So

91
00:05:19,090 --> 00:05:22,716
we've had to start to, you know,
care and invest more in quality

92
00:05:22,716 --> 00:05:26,285
manufacturing operations, things
like that. And then we've also

93
00:05:26,285 --> 00:05:29,571
sort of the vertical as our
product, meaning that there's,

94
00:05:29,571 --> 00:05:32,913
you know, software that can sit
on top of that back and saw

95
00:05:32,913 --> 00:05:36,086
those, you know, industry use
cases like I mentioned. So

96
00:05:36,086 --> 00:05:39,598
that's a bit of the bit of the
journey. It's you know, startup

97
00:05:39,598 --> 00:05:42,941
life is definitely a roller
coaster. But, yeah, that's kind

98
00:05:42,941 --> 00:05:46,000
of the main transition, kind of
in the last six years.

99
00:05:46,000 --> 00:05:49,888
Audrow Nash: And where are you
guys at in terms of are you

100
00:05:49,888 --> 00:05:54,246
taking investment. Are you, cash
flow like earning good cash flow

101
00:05:54,246 --> 00:05:58,000
from all the sales or how are
you guys funded I suppose.

102
00:05:58,000 --> 00:06:01,508
Bradley Dillon: So we have been
funded by kind of bootstrapped

103
00:06:01,508 --> 00:06:05,185
plus friends and family. So, you
know, personal. Yeah. It's crazy

104
00:06:05,185 --> 00:06:08,636
crazy six years deep doing that.
So personally, I've invested

105
00:06:08,636 --> 00:06:11,918
twice. I've had eight. My
friends and family, who invested

106
00:06:11,918 --> 00:06:15,482
as well and, you know, mix and
mix things like, say, going to a

107
00:06:15,482 --> 00:06:18,763
wedding a little bit awkward
because they want to know how

108
00:06:18,763 --> 00:06:22,158
other how their money's doing.
So we have luckily this great

109
00:06:22,158 --> 00:06:25,553
network of, of angels and some
of these angels have cut, you

110
00:06:25,553 --> 00:06:29,004
know, bigger tickets to invest
in us. And that's been able to

111
00:06:29,004 --> 00:06:32,568
get us to where we're at today.
So we've had to be very scrappy

112
00:06:32,568 --> 00:06:36,132
to be able to do that. And, you
know, actually, you know, build

113
00:06:36,132 --> 00:06:39,527
and have, hardware inventory.
And then as we look ahead, you

114
00:06:39,527 --> 00:06:42,469
know, we yeah, we are
considering all options, where

115
00:06:42,469 --> 00:06:45,976
we potentially would be looking
to do a fundraise, in the, you

116
00:06:45,976 --> 00:06:49,032
know, not so distant future,
but, you know, for is it,

117
00:06:49,032 --> 00:06:52,370
financially we're, we're, we're
doing well, we're not quite

118
00:06:52,370 --> 00:06:55,990
currently, you know, profitable
or cash flow positive. But we're

119
00:06:55,990 --> 00:06:59,498
relatively close. It's really
just that we have so much things

120
00:06:59,498 --> 00:07:02,836
that we're so excited to develop
from a product development

121
00:07:02,836 --> 00:07:06,344
perspective that, you know, we
we keep just plowing, you know,

122
00:07:06,344 --> 00:07:09,795
every available dollar we can
justify into a building for the

123
00:07:09,795 --> 00:07:13,359
future. But yeah, I think, you
know, the company is very stable

124
00:07:13,359 --> 00:07:15,000
from a financial perspective.

125
00:07:15,000 --> 00:07:17,481
Audrow Nash: Yeah. That's
awesome to hear. That's

126
00:07:17,481 --> 00:07:20,418
absolutely wild that you've made
it this far on family and

127
00:07:20,418 --> 00:07:23,456
friends and some individual
angel investors. Do you think if

128
00:07:23,456 --> 00:07:26,646
you were to go for investment,
would it be like an A round or a

129
00:07:26,646 --> 00:07:29,886
B round or something? Because it
sounds like you have. I mean, I

130
00:07:29,886 --> 00:07:32,873
think you have pretty good
traction. You've kind of skipped

131
00:07:32,873 --> 00:07:35,000
a little bit. But what do you
think of it.

132
00:07:35,000 --> 00:07:38,668
Bradley Dillon: Yeah, it's funny
the amount raised and you know,

133
00:07:38,668 --> 00:07:42,107
series A, B rounds has changed
so much over the last 5 to 10

134
00:07:42,107 --> 00:07:45,604
years. So, you know, a series A
that used to maybe have been,

135
00:07:45,604 --> 00:07:49,157
you know, $5 million that maybe
is now kind of $25 million. So

136
00:07:49,157 --> 00:07:52,768
yeah, it's funny how that funny
that that's happened, but yeah,

137
00:07:52,768 --> 00:07:56,093
I mean, in the, probably in the
venture capital world, you

138
00:07:56,093 --> 00:07:59,474
would, you would call it, you
know, probably a series A and

139
00:07:59,474 --> 00:08:02,799
you would think the terms are
kind of funny, but you'd say

140
00:08:02,799 --> 00:08:06,352
maybe we previously done like,
you know, a pre-seed a seed and

141
00:08:06,352 --> 00:08:09,791
a pre-series A, so something
along those lines. But yeah, it

142
00:08:09,791 --> 00:08:13,345
would be, it would be a larger
institutional raise. That would

143
00:08:13,345 --> 00:08:16,784
be, you know, raising a, you
know, more significant chunk of

144
00:08:16,784 --> 00:08:20,452
capital and likely with likely,
the kind of venture capital side

145
00:08:20,452 --> 00:08:22,000
of institutional investors.

146
00:08:22,000 --> 00:08:25,385
Audrow Nash: Okay. Very
interesting, super exciting.

147
00:08:25,385 --> 00:08:29,225
That you guys are there, like
funding by that way. And have

148
00:08:29,225 --> 00:08:33,195
been scrappy to get there. I
really like that. I want to talk

149
00:08:33,195 --> 00:08:36,580
more about that a bit later.
But. Tell me about your

150
00:08:36,580 --> 00:08:40,550
equipment. Tell me about your
cameras. Tell me about how they

151
00:08:40,550 --> 00:08:44,000
generally work and what kinds of
sensors you support.

152
00:08:44,000 --> 00:08:46,830
Bradley Dillon: Yeah,
absolutely. So I'll hold out for

153
00:08:46,830 --> 00:08:50,132
any of your audience ideas. You
know, watching the video stream

154
00:08:50,132 --> 00:08:53,433
here. This is an example of one
of our, our cameras right here.

155
00:08:53,433 --> 00:08:56,578
And I'll hold it up to my eyes
and start with to say, if you

156
00:08:56,578 --> 00:08:59,722
can kind of see there's a left
camera and a right camera and

157
00:08:59,722 --> 00:09:02,971
it's actually we designed it is
exactly the width of our human

158
00:09:02,971 --> 00:09:06,273
eyes. And so our thinking was,
is, you know, for some reason we

159
00:09:06,273 --> 00:09:09,575
evolved to have human eyes that
were, you know, this far apart.

160
00:09:09,575 --> 00:09:12,929
So might as well just pick that
for these pretty stereo cameras.

161
00:09:12,929 --> 00:09:16,073
So from a sensor perspective,
the first big thing is, is the

162
00:09:16,073 --> 00:09:18,956
left camera on the right camera
day. That data is fuzed

163
00:09:18,956 --> 00:09:22,100
together, and it makes it so
that you can create a 3D map of

164
00:09:22,100 --> 00:09:25,087
the world in front of you. So it
makes it so that you can

165
00:09:25,087 --> 00:09:28,232
understand, okay, this laptop
screen is, you know, 15in away

166
00:09:28,232 --> 00:09:31,219
for me. And then from a sensor
perspective, we also have,

167
00:09:31,219 --> 00:09:34,520
there's another camera here at
the center. I made a little hard

168
00:09:34,520 --> 00:09:37,927
to see. And that's a really high
resolution camera. So this would

169
00:09:37,927 --> 00:09:41,229
be something that would be, you
know, is high resolution is the

170
00:09:41,229 --> 00:09:44,478
latest and greatest iPhone. So
be, you know, our latest models

171
00:09:44,478 --> 00:09:47,570
in fact 48 megapixel. So it's a
really high resolution. And

172
00:09:47,570 --> 00:09:50,714
we're pushing the boundaries
that are to our newest hardware

173
00:09:50,714 --> 00:09:53,544
can support up to 108
megapixels. And so what's pretty

174
00:09:53,544 --> 00:09:56,584
incredible about the journey
they're on pixels is they say

175
00:09:56,584 --> 00:09:59,728
that the human eye has about,
you know, five 680 pixels. And

176
00:09:59,728 --> 00:10:02,820
so it won't be too long before
it will basically be able to

177
00:10:02,820 --> 00:10:05,912
have cameras that, you know,
have higher resolution and are

178
00:10:05,912 --> 00:10:08,952
human eyes, which is pretty
incredible to think about. And

179
00:10:08,952 --> 00:10:10,000
then to continue on.

180
00:10:10,000 --> 00:10:13,529
Audrow Nash: We we probably do
have. Cameras that are higher

181
00:10:13,529 --> 00:10:16,941
resolution than a human, I
would. Think, but it's not, Not

182
00:10:16,941 --> 00:10:19,000
a small form factor. Is it true
or.

183
00:10:19,000 --> 00:10:22,378
Bradley Dillon: That's right.
Yeah. That's not so we support

184
00:10:22,378 --> 00:10:26,038
kind of sensors that we can, you
know, manufacture at scale that,

185
00:10:26,038 --> 00:10:29,360
you know, at an economic price.
And so there are, you know,

186
00:10:29,360 --> 00:10:32,514
there are cameras like you
mentioned at do go that high.

187
00:10:32,514 --> 00:10:35,948
And it's really for, you know,
probably another four years in

188
00:10:35,948 --> 00:10:39,383
our journey, we'd be able to
offer that, I would say with our

189
00:10:39,383 --> 00:10:42,705
platform based on the, you know,
the cost curve and the way

190
00:10:42,705 --> 00:10:44,000
things are progressing.

191
00:10:44,000 --> 00:10:45,000
Audrow Nash: So exciting.

192
00:10:45,000 --> 00:10:47,911
Bradley Dillon: Yeah, it's
really incredibly exciting, for

193
00:10:47,911 --> 00:10:50,721
sure. And then some other
sensors on these products. And

194
00:10:50,721 --> 00:10:53,581
so there's, there's, some
infrared here and the infrared,

195
00:10:53,581 --> 00:10:56,492
what it does is, is that, two
things. One is it's infrared

196
00:10:56,492 --> 00:10:59,653
illumination and it's basically
night vision. So it makes it so

197
00:10:59,653 --> 00:11:02,765
that you're if you're at night,
you can still operate with it.

198
00:11:02,765 --> 00:11:05,575
And then there's this thing
called an infrared laser dot

199
00:11:05,575 --> 00:11:08,435
projector. And that's for,
something that's called active

200
00:11:08,435 --> 00:11:11,547
stereo. And what that does is it
basically, you can imagine it

201
00:11:11,547 --> 00:11:14,407
spray paints a bunch of dots on
a surface. And, you know,

202
00:11:14,407 --> 00:11:17,468
sometimes if you're like looking
at a like, say, like a white

203
00:11:17,468 --> 00:11:20,530
wall, you get kind of you get
kind of woozy because you can't

204
00:11:20,530 --> 00:11:23,691
tell how far it is. You kind of
get your your mind gets kind of

205
00:11:23,691 --> 00:11:26,752
warped out so that this does the
infrared laser dot projector

206
00:11:26,752 --> 00:11:29,813
makes it so that you get texture
on that surface, and you can

207
00:11:29,813 --> 00:11:33,025
actually see how far it is. And
then on the inside of the device

208
00:11:33,025 --> 00:11:36,036
we have additional sensors. So
we have like EMUs. So it's an

209
00:11:36,036 --> 00:11:38,997
inertial armament unit. And what
that does is it, you know,

210
00:11:38,997 --> 00:11:41,907
measures basically can measure
like the pitch and roll and

211
00:11:41,907 --> 00:11:44,868
measure how things are moving.
So that's the thing, like on

212
00:11:44,868 --> 00:11:47,829
your iPhone where, you know, it
knows that when you turn it

213
00:11:47,829 --> 00:11:51,041
sideways, it knows to like, you
know, turn your turn your screen

214
00:11:51,041 --> 00:11:53,751
sideways. And then we some
other, other new sensors as

215
00:11:53,751 --> 00:11:56,611
well, they're excited about. So
we have our first thermal

216
00:11:56,611 --> 00:11:59,622
sensor. And then we have also
our first, lidar based sensor.

217
00:11:59,622 --> 00:12:02,633
And that's, basically where a
bunch of light is shot off and

218
00:12:02,633 --> 00:12:05,744
then it bounces back. And then
based on how much time it takes

219
00:12:05,744 --> 00:12:08,655
for that to happen, I can create
a nice 3D, you know, high

220
00:12:08,655 --> 00:12:11,616
precision model of things. So,
yeah, for, for a full sensor

221
00:12:11,616 --> 00:12:14,476
stack, that's what I mentioned.
I forgot to mention this.

222
00:12:14,476 --> 00:12:17,236
Microphones. So these are,
incredible amount of sensors

223
00:12:17,236 --> 00:12:20,197
that are all fuzing the data
together on the edge, which is

224
00:12:20,197 --> 00:12:21,000
really exciting.

225
00:12:21,000 --> 00:12:25,487
Audrow Nash: Do you have the is
the thermal camera, so my video

226
00:12:25,487 --> 00:12:29,546
is a bit, jagged, so I can't
always see everything you're

227
00:12:29,546 --> 00:12:33,748
holding up, but the, like, it's
like, stop and go. But the,

228
00:12:33,748 --> 00:12:38,163
thermal camera, is that on one
of the like, do you have all of

229
00:12:38,163 --> 00:12:42,151
the sensors you mentioned on
one, device, or is it, that

230
00:12:42,151 --> 00:12:45,000
there's like a dedicated thermal
camera?

231
00:12:45,000 --> 00:12:47,842
Bradley Dillon: Oh, we have a
dedicated thermal camera, so I'm

232
00:12:47,842 --> 00:12:50,639
holding it up here. Oh, I see it
now. Okay, so this one, that

233
00:12:50,639 --> 00:12:53,206
kind of looks like it's
reflecting like a mirror. That's

234
00:12:53,206 --> 00:12:56,003
the that's the actual thermal
sensor. And then we have a high

235
00:12:56,003 --> 00:12:58,754
resolution color camera. And so
then that data is able to be

236
00:12:58,754 --> 00:13:01,000
fuzed together. So yeah it's a
dedicated product.

237
00:13:01,000 --> 00:13:05,135
Audrow Nash: That's so cool
okay. And so with all of these

238
00:13:05,135 --> 00:13:09,484
sensors that you have, I guess
it's very interesting that you

239
00:13:09,484 --> 00:13:13,976
have so many different types of
sensors, the the microphone the

240
00:13:13,976 --> 00:13:18,039
IMU for figuring out how it's
tilted. Tell me a bit about

241
00:13:18,039 --> 00:13:22,103
having all of those sensors
because it's, it's going more

242
00:13:22,103 --> 00:13:26,238
from. Like being a vision sensor
to being a sensor node or

243
00:13:26,238 --> 00:13:30,231
something like that, like a
little platform you strap on

244
00:13:30,231 --> 00:13:34,009
that has everything. Tell me
about that, because it's

245
00:13:34,009 --> 00:13:37,788
interesting that you're
integrating so many different

246
00:13:37,788 --> 00:13:39,000
types of sensors.

247
00:13:39,000 --> 00:13:42,330
Bradley Dillon: Yeah. So it's a
great question. And it kind of

248
00:13:42,330 --> 00:13:45,606
is at the heart of one of our
durable product strategies. And

249
00:13:45,606 --> 00:13:48,399
so one of our long term
strategies that we're really

250
00:13:48,399 --> 00:13:51,622
bought into is, is that we want
to be supporting new sensors

251
00:13:51,622 --> 00:13:54,952
every year. And more performance
answers for the existing ones

252
00:13:54,952 --> 00:13:58,175
that we have in the lineup.
Well, we hear from our customers

253
00:13:58,175 --> 00:14:01,129
is, is that when you have to
have an array of different

254
00:14:01,129 --> 00:14:04,566
sensors that come from different
providers and different systems

255
00:14:04,566 --> 00:14:07,413
and different setups, it's
really difficult to manage

256
00:14:07,413 --> 00:14:10,475
those. And so by integrating
them all into one device, it

257
00:14:10,475 --> 00:14:13,912
really makes things a lot easier
to be able to operate and makes

258
00:14:13,912 --> 00:14:17,188
it so that you can really fuze
that data together and base it

259
00:14:17,188 --> 00:14:20,411
so that you know exactly from a
like a timing perspective on

260
00:14:20,411 --> 00:14:23,795
where you're getting each piece
of data, where if you're trying

261
00:14:23,795 --> 00:14:26,856
to synchronize that across
different devices, it's really

262
00:14:26,856 --> 00:14:30,294
hard to get that to be done in a
really low latency, really, you

263
00:14:30,294 --> 00:14:33,624
know, really high performance
type, the type of case. And, you

264
00:14:33,624 --> 00:14:37,062
know, we believe that by having
the, you know, doing the painful

265
00:14:37,062 --> 00:14:40,553
thing of constantly adding these
new drivers to support all these

266
00:14:40,553 --> 00:14:43,561
sensors that we're making it
easier for, for our clients

267
00:14:43,561 --> 00:14:46,729
because they can just buy a
device that's you're all in one

268
00:14:46,729 --> 00:14:49,737
sensor, sensor device. And then
from a, you know, a more

269
00:14:49,737 --> 00:14:52,853
abstract perspective, we're
ultimately, in a lot of cases,

270
00:14:52,853 --> 00:14:56,129
we're competing with humans,
humans maybe would have been the

271
00:14:56,129 --> 00:14:59,566
ones that were, you know, doing
a type of task before. And so we

272
00:14:59,566 --> 00:15:02,896
want these systems to be able to
compete with humans. And that

273
00:15:02,896 --> 00:15:06,012
means that you have sensors
that, you know, can do some of

274
00:15:06,012 --> 00:15:09,181
what humans can, but then you
can add additional ones where

275
00:15:09,181 --> 00:15:12,081
you basically can imagine a
human, but they have super

276
00:15:12,081 --> 00:15:15,250
powers, right? So, you know,
humans, you know, don't have a

277
00:15:15,250 --> 00:15:18,580
thermal sensor. We can't, you
know, look at something and know

278
00:15:18,580 --> 00:15:21,964
how hot or cold it is. Whereas
but you add it to our device and

279
00:15:21,964 --> 00:15:24,918
you can and so that that's
pretty exciting for us to be

280
00:15:24,918 --> 00:15:28,194
able to unlock these use cases
that these sensors can provide

281
00:15:28,194 --> 00:15:29,000
to our clients.

282
00:15:29,000 --> 00:15:33,203
Audrow Nash: Yeah. I like that
perspective a lot. And with the

283
00:15:33,203 --> 00:15:37,068
thing brought in. Which is just
how crazy large the labor

284
00:15:37,068 --> 00:15:41,203
shortages are across lots of
industries, and that makes it so

285
00:15:41,203 --> 00:15:45,339
that there's a very good, reason
to try to get robots or some

286
00:15:45,339 --> 00:15:48,661
sort of automation to be
replacing humans or just

287
00:15:48,661 --> 00:15:52,729
providing work where there are
no humans for all this. Like,

288
00:15:52,729 --> 00:15:57,000
that's the thing that I think is
often, I don't know, should be

289
00:15:57,000 --> 00:16:01,339
thrown in there because there is
some fear around taking all the

290
00:16:01,339 --> 00:16:05,000
jobs, I think by robots taking
all the jobs of people.

291
00:16:05,000 --> 00:16:07,955
Bradley Dillon: It's a great
topic. A great point is

292
00:16:07,955 --> 00:16:11,307
something that we talk a lot
about and something that we're

293
00:16:11,307 --> 00:16:14,659
pretty passionate about. Our
perspective is, is that these,

294
00:16:14,659 --> 00:16:18,068
you know, robotics and, and
automation systems are going to,

295
00:16:18,068 --> 00:16:21,705
you know, make the world a much
better place. And the reason for

296
00:16:21,705 --> 00:16:25,170
it is, is that there, as you
mentioned, in a lot of the a lot

297
00:16:25,170 --> 00:16:28,466
of the rich world, you know,
labor has gotten more scarce.

298
00:16:28,466 --> 00:16:31,818
And the people that are doing
physical labor and what these

299
00:16:31,818 --> 00:16:35,057
types of robot robotics and
automation systems do is they

300
00:16:35,057 --> 00:16:38,239
they're able to take the most
kind of like mind numbing,

301
00:16:38,239 --> 00:16:41,591
repetitive, dangerous, tasks.
They're able to take that off

302
00:16:41,591 --> 00:16:45,227
from the human. And what they do
then is they're able to make it

303
00:16:45,227 --> 00:16:48,864
so that collectively, the world
can become a more productive and

304
00:16:48,864 --> 00:16:52,159
a richer place. And so you seen
this throughout, you know,

305
00:16:52,159 --> 00:16:55,739
technology history over the last
couple hundred years that each

306
00:16:55,739 --> 00:16:59,261
new wave of technology, what it
is, is it's very deflation. So

307
00:16:59,261 --> 00:17:02,273
it makes it so that we can be
more efficient and more

308
00:17:02,273 --> 00:17:05,795
productive. And then what that
does is then it, you know, that

309
00:17:05,795 --> 00:17:09,261
that makes it so that there's
excess capital, it gets formed,

310
00:17:09,261 --> 00:17:12,500
that excess capital gets
invested into new things, to new

311
00:17:12,500 --> 00:17:16,080
investments and new innovation.
And the world, you know, keeps,

312
00:17:16,080 --> 00:17:19,773
keeps having more and more jobs.
And the example I like to use is

313
00:17:19,773 --> 00:17:23,295
I got to recently visit South
Korea and, and South Korea. They

314
00:17:23,295 --> 00:17:26,818
actually have, five times more
robots per human than any other

315
00:17:26,818 --> 00:17:30,000
country in the world. It's like
walking into the future.

316
00:17:30,000 --> 00:17:31,000
Audrow Nash: Amazing.

317
00:17:30,000 --> 00:17:33,143
Bradley Dillon: And so I, you
know, at the airport, they have

318
00:17:33,143 --> 00:17:36,286
robots helping with things. I
went to a restaurant where they

319
00:17:36,286 --> 00:17:39,275
had, the, the food brought to
our table and taken from our

320
00:17:39,275 --> 00:17:42,521
table by a robot. And the dishes
were cleaned by a robot. And I

321
00:17:42,521 --> 00:17:45,561
don't think anybody in that
restaurant is saying I'm really

322
00:17:45,561 --> 00:17:48,447
upset. I really wanted to wash
the dishes right now, and

323
00:17:48,447 --> 00:17:51,538
instead that employee can say,
oh, I can try to maybe go and

324
00:17:51,538 --> 00:17:54,681
improve the client experience. I
can go try to upsell bam and

325
00:17:54,681 --> 00:17:57,618
have a better, you know,
personal touch. And the owner of

326
00:17:57,618 --> 00:18:00,813
that restaurant, they can pay
their workers more and they can,

327
00:18:00,813 --> 00:18:04,162
you know, use that extra savings
to maybe expand, you know, etc.,

328
00:18:04,162 --> 00:18:07,048
etc.. So we think it's an
incredible positive thing. And

329
00:18:07,048 --> 00:18:10,294
we know it's I think it's easy
sometimes to think, oh yeah, we,

330
00:18:10,294 --> 00:18:13,437
we believe in tech history. We
know what happened before, but

331
00:18:13,437 --> 00:18:16,632
this time is different. And I
don't I don't think this time is

332
00:18:16,632 --> 00:18:19,826
different. I think I think it's
another way that's going to be

333
00:18:19,826 --> 00:18:22,918
very similar. And I think it's
going to make humanity in the

334
00:18:22,918 --> 00:18:24,000
world a better place.

335
00:18:24,000 --> 00:18:27,803
Audrow Nash: Totally. I think
this time is a little different

336
00:18:27,803 --> 00:18:31,855
in that demographically we're in
a way that the world hasn't been

337
00:18:31,855 --> 00:18:35,345
in the past. When we have
increasing technologies, which

338
00:18:35,345 --> 00:18:39,086
is we have a large in most of
the, rich world say. There's a

339
00:18:39,086 --> 00:18:42,764
lot of older people and, that's
kind of a new thing. And so

340
00:18:42,764 --> 00:18:46,379
there's a lot of people leaving
the workforce and a lot of

341
00:18:46,379 --> 00:18:50,244
people to be taken care of, and
so, like, if you look at South

342
00:18:50,244 --> 00:18:53,922
Korea. As you were saying,
they're. One that has aged very,

343
00:18:53,922 --> 00:18:57,974
very fast. I think their average
age is far, far older than, say,

344
00:18:57,974 --> 00:19:01,153
the US. And so it's an
interesting thing there that

345
00:19:01,153 --> 00:19:04,519
they have compensated so
significantly with automation

346
00:19:04,519 --> 00:19:08,322
and are still very competitive,
like they're doing great work

347
00:19:08,322 --> 00:19:12,000
over there in general. And my
opinion are from what I know.

348
00:19:12,000 --> 00:19:14,422
Bradley Dillon: Yeah, I
completely agree. And

349
00:19:14,422 --> 00:19:17,543
considering on the South Korea
use case, it's interesting.

350
00:19:17,543 --> 00:19:20,503
Their their birth rates are
below like one. So it means

351
00:19:20,503 --> 00:19:23,517
they're like well below the
replacement rate. And what's

352
00:19:23,517 --> 00:19:26,639
interesting is again, as the
population is aged there, the

353
00:19:26,639 --> 00:19:30,029
unemployment rate there is very
low. It's been consistent makes

354
00:19:30,029 --> 00:19:32,774
sense below 5%. And you're
totally right. So as the

355
00:19:32,774 --> 00:19:35,895
population gets older we're
saying like who or who's going

356
00:19:35,895 --> 00:19:39,178
to who's going to care for this
aging population. And I think

357
00:19:39,178 --> 00:19:42,623
that's where, you know, robotics
and automation can come in. You

358
00:19:42,623 --> 00:19:45,744
can have a humanoid robot that,
you know, if I'm imagining

359
00:19:45,744 --> 00:19:48,865
myself, you know, kicking it in,
you know, an elderly care

360
00:19:48,865 --> 00:19:52,148
facility one day, like, I'd have
my humanoid robot buddy that

361
00:19:52,148 --> 00:19:55,162
would help me with, you know,
various tasks you can have

362
00:19:55,162 --> 00:19:58,122
things like, you know, automated
systems that dispense,

363
00:19:58,122 --> 00:20:01,567
prescriptions, for example. And,
that can improve the safety of,

364
00:20:01,567 --> 00:20:04,473
oh, maybe you had somebody
elderly that's accidentally

365
00:20:04,473 --> 00:20:07,648
taking too many or not taking
them. So there's all kinds of

366
00:20:07,648 --> 00:20:10,608
things where you can both
improve the, the, the elderly

367
00:20:10,608 --> 00:20:13,729
care experience. And as you
mentioned, as a population, as

368
00:20:13,729 --> 00:20:16,905
more and more people are not
working, and it means that the

369
00:20:16,905 --> 00:20:19,972
the existing workforce is not
enough to sustain the whole

370
00:20:19,972 --> 00:20:23,094
economy. And that's, you know,
that that's where robotics,

371
00:20:23,094 --> 00:20:26,000
honestly, is going to be able to
come in in a big way.

372
00:20:26,000 --> 00:20:29,200
Audrow Nash: Yeah. That's why
it's. This is a big reason I'm

373
00:20:29,200 --> 00:20:32,453
so bullish on robotics for the
near future. I think necessity

374
00:20:32,453 --> 00:20:35,653
is going to pull robotics to be
bigger and bigger because of

375
00:20:35,653 --> 00:20:38,853
this kind of thing, this large
demographic trend. But so you

376
00:20:38,853 --> 00:20:42,000
mentioned pill counting and
things like that. Tell me about

377
00:20:42,000 --> 00:20:45,307
some of the things your cameras
are being used. For, like cool

378
00:20:45,307 --> 00:20:48,400
automation tasks. Obviously
they're used in quite a bit of

379
00:20:48,400 --> 00:20:51,653
quite a number of robots. But
tell me about some of the tasks

380
00:20:51,653 --> 00:20:54,000
that your sensor that your
platforms enable.

381
00:20:54,000 --> 00:20:57,320
Bradley Dillon: Yeah. So the
I'll start by sharing an example

382
00:20:57,320 --> 00:21:00,204
which is in the industrial
equipment and agricultural

383
00:21:00,204 --> 00:21:03,251
space. And so we're helping a
large industrial equipment

384
00:21:03,251 --> 00:21:06,571
manufacturer where the machines
are used for, for things like

385
00:21:06,571 --> 00:21:10,108
agriculture and mining and stuff
like that. We're helping them on

386
00:21:10,108 --> 00:21:12,992
a, on a multiyear. It'll
probably end up being like a

387
00:21:12,992 --> 00:21:16,257
decade long journey of doing
three key steps. The first step

388
00:21:16,257 --> 00:21:19,685
is we're making it so that the
machines stop harming the human.

389
00:21:19,685 --> 00:21:22,787
And it's pretty shocking. Around
the world, machines kill

390
00:21:22,787 --> 00:21:25,780
millions of humans per year
today, and we unfortunately

391
00:21:25,780 --> 00:21:29,100
tolerate it now. A lot of those
obviously are like automobile

392
00:21:29,100 --> 00:21:32,093
accidents, for example. But
there's a lot of industrial

393
00:21:32,093 --> 00:21:35,358
equipment as well. So step one
is to make it so that there's

394
00:21:35,358 --> 00:21:38,623
human machine safety. And so
that the machine's top priority

395
00:21:38,623 --> 00:21:41,888
at all times is to make sure
that with that, there's a human

396
00:21:41,888 --> 00:21:45,153
around that they're not going to
harm it. And so so what our

397
00:21:45,153 --> 00:21:48,473
cameras do is they make it so
that the device, the industrial

398
00:21:48,473 --> 00:21:51,683
equipment, it can it can see
what's around them, and it can

399
00:21:51,683 --> 00:21:54,894
make sure that, you know, arms,
limbs, you know, humans are

400
00:21:54,894 --> 00:21:58,268
avoiding the equipment, doesn't
harm them. And then the second

401
00:21:58,268 --> 00:22:01,642
step with, that company is that
and the industry in general is

402
00:22:01,642 --> 00:22:05,070
to make it so that the equipment
can operate remotely. And so a

403
00:22:05,070 --> 00:22:08,499
lot of these equipments, pieces
of equipment are they're pretty

404
00:22:08,499 --> 00:22:11,927
difficult to operate over a long
time. They they vibrate a lot.

405
00:22:11,927 --> 00:22:15,301
They're, they're pretty hard on
the operators. And so by being

406
00:22:15,301 --> 00:22:18,784
able to operate it remotely, it
makes it so you can operate them

407
00:22:18,784 --> 00:22:21,614
more safely and more
efficiently. And then the final

408
00:22:21,614 --> 00:22:24,988
frontier is to make it so the
equipment can operate completely

409
00:22:24,988 --> 00:22:28,089
autonomously. And so that's
where you can set it off on a

410
00:22:28,089 --> 00:22:31,463
task, and you maybe have some
remote folks that can, you know,

411
00:22:31,463 --> 00:22:34,892
take over if needed. But they're
mostly operating autonomously.

412
00:22:34,892 --> 00:22:38,157
And so I think you see that
across different industries. But

413
00:22:38,157 --> 00:22:41,422
that's, you know, as an example,
in the industrial equipment

414
00:22:41,422 --> 00:22:43,000
space that we're helping one.

415
00:22:43,000 --> 00:22:47,472
Audrow Nash: I really like that.
Yeah. I, I'm just thinking back

416
00:22:47,472 --> 00:22:51,803
to an interview with Polymath
Robotics where they were talking

417
00:22:51,803 --> 00:22:55,786
about just driving massive
amounts of say toxic chemicals

418
00:22:55,786 --> 00:23:00,118
or like molten metal or crazy
things that you don't want to be

419
00:23:00,118 --> 00:23:04,450
exposed to. And the pay is high,
but the cancer rates are high

420
00:23:04,450 --> 00:23:08,921
and the workers in this kind of
thing. And so if you can make it

421
00:23:08,921 --> 00:23:12,974
so that we can automate or,
remotely operate some of these

422
00:23:12,974 --> 00:23:15,000
tasks, like that's huge wins.

423
00:23:15,000 --> 00:23:17,586
Bradley Dillon: Yeah,
absolutely. And what's also

424
00:23:17,586 --> 00:23:20,859
interesting is the equipment in
a lot of cases can be designed

425
00:23:20,859 --> 00:23:23,973
better and more performant. And
an example I'll use is with

426
00:23:23,973 --> 00:23:27,193
fighter pilots. We've you know,
we've we're used to kind of a

427
00:23:27,193 --> 00:23:30,465
pseudo top gun fighter pilot.
And what's interesting today is,

428
00:23:30,465 --> 00:23:33,580
is that, you know, basically,
you know, I operated, fighter

429
00:23:33,580 --> 00:23:36,905
jets were able to outperform a
human. And what's so interesting

430
00:23:36,905 --> 00:23:40,178
is you can design them so that
they could say, be moving at a,

431
00:23:40,178 --> 00:23:43,450
like, you know, more than ten
G's. It's something that a human

432
00:23:43,450 --> 00:23:46,670
couldn't handle, but the machine
doesn't care. If the machine

433
00:23:46,670 --> 00:23:49,890
doesn't have a human, it can
operate more performatively. And

434
00:23:49,890 --> 00:23:53,057
so I think in a lot of cases,
this industrial equipment, you

435
00:23:53,057 --> 00:23:56,171
can reimagine the way it's
operated, because you don't need

436
00:23:56,171 --> 00:23:59,443
to have to have a space where
the operator and you can make it

437
00:23:59,443 --> 00:24:02,874
operate in a way that, you know,
it's you'd say, oh, maybe that's

438
00:24:02,874 --> 00:24:06,041
too high vibration, or it's
going somewhere that wouldn't be

439
00:24:06,041 --> 00:24:08,997
safe. And so it unlocks, you
know, new opportunities for

440
00:24:08,997 --> 00:24:10,000
efficiency as well.

441
00:24:10,000 --> 00:24:13,430
Audrow Nash: Yeah. Very true.
And especially if we can

442
00:24:13,430 --> 00:24:17,177
automate stuff I'm thinking of
like electric sheep having a

443
00:24:17,177 --> 00:24:21,115
bunch of little lawnmowers that
go and mow. Because it's safer

444
00:24:21,115 --> 00:24:24,925
and it's easier to manage and
everything else. And then they

445
00:24:24,925 --> 00:24:28,990
can't cause as much damage. And
so if you do that kind of thing,

446
00:24:28,990 --> 00:24:33,055
then you don't have to have the
big rider that requires a person

447
00:24:33,055 --> 00:24:37,056
to operate it. And they're big
because there aren't many people

448
00:24:37,056 --> 00:24:41,184
for these roles for this kind of
thing. So yeah, imagine the form

449
00:24:41,184 --> 00:24:45,122
factor is changing quite a bit.
And those jets that they'll go

450
00:24:45,122 --> 00:24:49,123
and turn incredibly quickly and
it would like turn a human into

451
00:24:49,123 --> 00:24:52,870
mush or something like that at
those speeds. Those are just

452
00:24:52,870 --> 00:24:56,999
bonkers. What you can do with AI
or with, with a computer driving

453
00:24:56,999 --> 00:25:00,428
everything or even being
remotely operated. I suppose.

454
00:25:00,428 --> 00:25:01,000
How wild.

455
00:25:01,000 --> 00:25:03,000
Bradley Dillon: What a time of
your life. It's incredible.

456
00:25:03,000 --> 00:25:06,353
Audrow Nash: Now, another thing
that's interesting about what I

457
00:25:06,353 --> 00:25:09,333
think you guys are doing is
you're also putting a lot of

458
00:25:09,333 --> 00:25:12,420
compute. So we've talked about
all the sensors that you're

459
00:25:12,420 --> 00:25:15,613
enabling, but. You've been also
putting the compute with the

460
00:25:15,613 --> 00:25:18,541
sensors. So some of the
platforms, have like a powerful

461
00:25:18,541 --> 00:25:21,681
computer attached to them.
Correct? Would you tell me about

462
00:25:21,681 --> 00:25:22,000
those.

463
00:25:22,000 --> 00:25:25,199
Bradley Dillon: Yeah, that's
that's right. And so that's

464
00:25:25,199 --> 00:25:28,627
really the other kind of second
key durable product strategy

465
00:25:28,627 --> 00:25:31,826
that we have, which is, is that
we enable, essentially a

466
00:25:31,826 --> 00:25:35,254
computer that's, that's right
there on the edge, right, with

467
00:25:35,254 --> 00:25:38,738
out there with all the sensors.
And so what that means is, is

468
00:25:38,738 --> 00:25:41,880
that you're able to run
applications and business logic

469
00:25:41,880 --> 00:25:45,079
right on the device, and it
means that you're able to do

470
00:25:45,079 --> 00:25:48,678
artificial intelligence models,
you're able actually to run the

471
00:25:48,678 --> 00:25:52,163
inference where it's, you know,
actually, you know, operating

472
00:25:52,163 --> 00:25:55,705
you're able to also run that on
the edge on the device. And so

473
00:25:55,705 --> 00:25:58,733
our thesis this is that
customers are always going to

474
00:25:58,733 --> 00:26:02,046
want to have more of that.
They're always going to want to

475
00:26:02,046 --> 00:26:05,474
have more on the edge because
they are able to solve one use

476
00:26:05,474 --> 00:26:09,073
case and they're like, this is
exciting. We want to solve more.

477
00:26:09,073 --> 00:26:12,101
And so what Luxonis equipment
does is it provides the

478
00:26:12,101 --> 00:26:15,128
opportunity to eliminate an
external dependency, like

479
00:26:15,128 --> 00:26:18,213
needing to have cloud. And so
customers that are maybe

480
00:26:18,213 --> 00:26:21,298
spending a lot of money today,
on cloud compute, cloud

481
00:26:21,298 --> 00:26:24,840
inference, they're able to take
that workload and they're able

482
00:26:24,840 --> 00:26:28,211
to take it to the edge. And
they're able to, you know, they

483
00:26:28,211 --> 00:26:31,410
own the hardware and they're
able to have a massive, you

484
00:26:31,410 --> 00:26:34,666
know, savings over time when
they're going to operate, in

485
00:26:34,666 --> 00:26:37,979
their production environments.
And so for our products, we

486
00:26:37,979 --> 00:26:41,578
actually had a new generation of
hardware. It's our series four

487
00:26:41,578 --> 00:26:45,006
of hardware. It's, this this is
an example of one right here

488
00:26:45,006 --> 00:26:48,148
that I'm holding up. And so this
little device has more

489
00:26:48,148 --> 00:26:51,690
processing power than the laptop
that I'm on here. And it has,

490
00:26:51,690 --> 00:26:54,603
and has a 48 top for AI
inference. And so what that

491
00:26:54,603 --> 00:26:57,974
means is that they're able to
run, several vision models in

492
00:26:57,974 --> 00:27:01,402
parallel. So you could have a
model that could do one thing,

493
00:27:01,402 --> 00:27:04,943
another thing, another thing,
you're able to run, run all that

494
00:27:04,943 --> 00:27:07,000
simultaneously, which is
incredible.

495
00:27:07,000 --> 00:27:09,762
Audrow Nash: What you said,
TOPS? is that like terabyte of

496
00:27:09,762 --> 00:27:11,000
floating point operations?

497
00:27:11,000 --> 00:27:14,563
Bradley Dillon: Yeah. So it's
trillions of operations per

498
00:27:14,563 --> 00:27:15,000
second.

499
00:27:15,000 --> 00:27:18,370
Audrow Nash: Wow. Yeah. That's
wild. Yeah. So you can run lots

500
00:27:18,370 --> 00:27:20,000
of models all at once on this.

501
00:27:20,000 --> 00:27:21,000
Bradley Dillon: Yeah.

502
00:27:20,000 --> 00:27:24,019
Audrow Nash: Exactly. Amazing.
And so so then with this you're

503
00:27:24,019 --> 00:27:28,038
you're empowering on the edge.
So you're making it so that you

504
00:27:28,038 --> 00:27:31,798
can have. Really good compute
demands. Or, really, really.

505
00:27:31,798 --> 00:27:35,882
Sophisticated compute occurring
at the edge. And this helps you

506
00:27:35,882 --> 00:27:39,318
get away from the cloud. The
thing the thing that I'm

507
00:27:39,318 --> 00:27:43,531
imagining with this is it's like
You have your, like sensor stack

508
00:27:43,531 --> 00:27:47,745
that is one of your devices, and
then you just have it connecting

509
00:27:47,745 --> 00:27:51,634
to a lot of things. And by doing
that you get it’s like this

510
00:27:51,634 --> 00:27:55,718
little. Almost parasite that can
be put on to any other kind of

511
00:27:55,718 --> 00:27:59,673
equipment. And then it can do a
lot of the automation really,

512
00:27:59,673 --> 00:28:03,692
really well. And it's all just
like you basically have the one

513
00:28:03,692 --> 00:28:07,581
node and then you have little
wires running everywhere else,

514
00:28:07,581 --> 00:28:11,341
like this kind of thing as a
potential workflow or, way of

515
00:28:11,341 --> 00:28:15,360
having these devices that kind
of the, the vision, having this

516
00:28:15,360 --> 00:28:19,379
little thing sit somewhere and
then automate something bigger.

517
00:28:19,379 --> 00:28:21,000
Or how do you imagine it.

518
00:28:21,000 --> 00:28:24,100
Bradley Dillon: Where you
described is is definitely one

519
00:28:24,100 --> 00:28:27,145
of the main to use cases. And so
the use case is really

520
00:28:27,145 --> 00:28:30,521
bifurcate. Between one is where
you have kind of a standalone

521
00:28:30,521 --> 00:28:33,677
automation device. And that
would be where you maybe have

522
00:28:33,677 --> 00:28:37,164
this camera that would be up and
it'd be connected to the maybe

523
00:28:37,164 --> 00:28:40,319
the infrastructure that you have
locally, and it would be

524
00:28:40,319 --> 00:28:43,530
operating everything. And it
could be, say, like observing

525
00:28:43,530 --> 00:28:47,017
something in the world. So this
could be somewhere and it could

526
00:28:47,017 --> 00:28:50,173
be doing, say, license plate
recognition, and it could be

527
00:28:50,173 --> 00:28:53,549
running an application that's,
you know, detecting a vehicle,

528
00:28:53,549 --> 00:28:56,871
detecting a license plate. And
then it's sending the license

529
00:28:56,871 --> 00:29:00,192
plate data somewhere, what you
described as the second, main

530
00:29:00,192 --> 00:29:03,403
use case, which is where our
devices are integrated into a

531
00:29:03,403 --> 00:29:06,503
larger, a larger assembly. And
that's really the kind of

532
00:29:06,503 --> 00:29:09,880
canonical example is, you know,
things like autonomous mobile

533
00:29:09,880 --> 00:29:13,090
robots. And so that's where you
have, you know, one of our

534
00:29:13,090 --> 00:29:16,467
cameras is there and it's like
you said, it's kind of a node,

535
00:29:16,467 --> 00:29:19,899
that's, you know, a peripheral
device that's capturing all the

536
00:29:19,899 --> 00:29:22,833
sensor data. It's helping
offload a little bit of the

537
00:29:22,833 --> 00:29:25,988
compute and some of the AI
inference. And, and then it's,

538
00:29:25,988 --> 00:29:29,420
you know, combined to a larger,
larger assembly, the area that

539
00:29:29,420 --> 00:29:32,410
is one area of kind of robotics
that's pretty exciting

540
00:29:32,410 --> 00:29:35,731
humanoids, for example. And so
we work with several humanoid

541
00:29:35,731 --> 00:29:39,108
customers. And, you know, these
humanoids are they need to be

542
00:29:39,108 --> 00:29:42,318
able to kind of navigate the
world. So to frequently have,

543
00:29:42,318 --> 00:29:45,806
you know, one of our cameras in
the front and then one of their

544
00:29:45,806 --> 00:29:49,238
camera, our cameras in the back,
and, and that's then combined

545
00:29:49,238 --> 00:29:52,559
to, of course, you have a lot of
other sensors that's on the

546
00:29:52,559 --> 00:29:55,715
humanoid, and you have a lot
more compute that's on it as

547
00:29:55,715 --> 00:29:58,925
well. And so it's kind of one
sensor in a, in a broad, you

548
00:29:58,925 --> 00:30:02,302
know, kind of a broad ecosystem
to do something which is very

549
00:30:02,302 --> 00:30:05,845
complicated, which is, you know,
have you tried to try to make a

550
00:30:05,845 --> 00:30:09,000
robot that can navigate the
world in the way a human can?

551
00:30:09,000 --> 00:30:12,411
Audrow Nash: Yeah. That's very
exciting. How do you so when you

552
00:30:12,411 --> 00:30:15,660
have. Multiple ones, the one in
the front and the one in the

553
00:30:15,660 --> 00:30:19,179
back, or. Say you're. Automating
a tractor or something like this

554
00:30:19,179 --> 00:30:22,428
that requires several around. So
you can see everywhere it's

555
00:30:22,428 --> 00:30:25,893
going. There's, see all sides of
it while it's operating. How do

556
00:30:25,893 --> 00:30:29,413
you, how do you connect multiple
devices and does this change if.

557
00:30:29,413 --> 00:30:32,662
Those devices have compute? Or
would you imagine having like

558
00:30:32,662 --> 00:30:35,964
one with compute or a bigger
computer somewhere else and then

559
00:30:35,964 --> 00:30:38,942
all these other ones, as.
Peripherals do it or how does

560
00:30:38,942 --> 00:30:41,000
how does connecting lots of them
work?

561
00:30:41,000 --> 00:30:44,243
Bradley Dillon: That's a great
question. So our devices have

562
00:30:44,243 --> 00:30:47,378
which I'll hold up here. They,
they have in this case with

563
00:30:47,378 --> 00:30:50,622
series four they have usb-C,
which is like what you'd use to

564
00:30:50,622 --> 00:30:53,541
power your iPhone for
connectivity. And then they also

565
00:30:53,541 --> 00:30:56,892
have these ports here. And this
part here is an Ethernet port.

566
00:30:56,892 --> 00:31:00,135
So you can use Ethernet for
power and connectivity. And then

567
00:31:00,135 --> 00:31:03,108
there's another port here which
is called like an image

568
00:31:03,108 --> 00:31:06,459
connector. And what that is, is
it's a connector that makes it

569
00:31:06,459 --> 00:31:09,865
so you can actually daisy chain
the devices together. And so by

570
00:31:09,865 --> 00:31:13,054
daisy chaining them together,
you can have an array of them

571
00:31:13,054 --> 00:31:16,297
that are pointed at different
things. And then that data can

572
00:31:16,297 --> 00:31:19,324
all be captured and kind of
fuzed together into one, you

573
00:31:19,324 --> 00:31:22,568
know, one broad picture, one
broad scene, to be able to, you

574
00:31:22,568 --> 00:31:25,919
know, navigate whatever you're
working on. So that is one, one

575
00:31:25,919 --> 00:31:29,162
example. There are also cases
where our devices are and it's

576
00:31:29,162 --> 00:31:32,243
more like kind of this device
here, it's really used as a

577
00:31:32,243 --> 00:31:35,703
peripheral device. And multiple
of them would be connected in to

578
00:31:35,703 --> 00:31:39,108
some type of a central compute
box that would be processing all

579
00:31:39,108 --> 00:31:41,000
of the data and making sense of
it.

580
00:31:41,000 --> 00:31:45,094
Audrow Nash: So tell me about
the AI models that. You have and

581
00:31:45,094 --> 00:31:49,255
how those play in so. Especially
for the ones with compute. But

582
00:31:49,255 --> 00:31:53,415
it sounds like you do you have
the option to do a lot of fairly

583
00:31:53,415 --> 00:31:57,377
sophisticated, Observation of
whatever the sensor is seeing.

584
00:31:57,377 --> 00:32:01,538
And I'd love to hear how you do
that and how you expose that to

585
00:32:01,538 --> 00:32:02,000
a user.

586
00:32:02,000 --> 00:32:05,055
Bradley Dillon: Yeah,
absolutely. So we have a team of

587
00:32:05,055 --> 00:32:08,507
about 20 artificial intelligence
engineers. We also have some

588
00:32:08,507 --> 00:32:11,958
folks that work on, creating
synthetic data. And so I'll kind

589
00:32:11,958 --> 00:32:15,070
of start with, you know, if
you're if you're a business

590
00:32:15,070 --> 00:32:18,578
customer and you have no data
and you're like, how do how do I

591
00:32:18,578 --> 00:32:21,973
go about creating an AI model?
So what you can do is you can

592
00:32:21,973 --> 00:32:25,198
grab some maybe publicly
available data that's out there.

593
00:32:25,198 --> 00:32:28,423
And so for example, at use like
license plate recognition

594
00:32:28,423 --> 00:32:31,592
technology. So you've maybe you
can maybe buy a publicly

595
00:32:31,592 --> 00:32:34,930
available data set, that has
some, you know, license plates

596
00:32:34,930 --> 00:32:37,929
for example. And then what
you're wanting to do to to

597
00:32:37,929 --> 00:32:41,211
develop a model is you maybe
start to want to add your own

598
00:32:41,211 --> 00:32:44,606
proprietary data. And this is
data that's unique to your own

599
00:32:44,606 --> 00:32:48,283
operation that you're wanting to
basically take existing data out

600
00:32:48,283 --> 00:32:51,508
there and you're wanting to
customize it for exactly your

601
00:32:51,508 --> 00:32:55,016
purpose. And so there's two ways
that you go about doing that.

602
00:32:55,016 --> 00:32:58,072
The first is you create
synthetic data, which we help,

603
00:32:58,072 --> 00:33:01,636
do for our customers. And that's
where you create these scenes.

604
00:33:01,636 --> 00:33:05,031
We actually use a tool called
unity. And it's the same thing

605
00:33:05,031 --> 00:33:08,313
that's used for creating video
games. So you basically you

606
00:33:08,313 --> 00:33:11,878
imagine is you basically create
a bunch of scenes that you that

607
00:33:11,878 --> 00:33:15,555
would that would be as good as a
video game. And those scenes are

608
00:33:15,555 --> 00:33:18,950
good enough quality that they
can be used to help train the,

609
00:33:18,950 --> 00:33:22,402
the AI model, along with the
publicly available data that you

610
00:33:22,402 --> 00:33:25,910
have, that gets you a proof of
concept that gets you something

611
00:33:25,910 --> 00:33:29,078
that says, okay, we have a
model, we can go out and test

612
00:33:29,078 --> 00:33:32,473
it. Maybe it works 80 or 90% and
8,090% is nowhere near good

613
00:33:32,473 --> 00:33:35,528
enough. If you want to have
something that's out there

614
00:33:35,528 --> 00:33:38,867
that's, you know, adapting to
the real world, handling edge

615
00:33:38,867 --> 00:33:42,148
cases, you know, the world is
always changing. You need to

616
00:33:42,148 --> 00:33:45,657
capture actual real world data.
That's, you know, from the the

617
00:33:45,657 --> 00:33:48,938
actual situation that you're
trying to navigate. So that's

618
00:33:48,938 --> 00:33:52,220
where our cameras come in. So
what our cameras do is they,

619
00:33:52,220 --> 00:33:55,671
they capture, new image data.
And that image data then can be

620
00:33:55,671 --> 00:33:59,179
used to be labeled. And then,
with that labeled data, then you

621
00:33:59,179 --> 00:34:02,744
can then retrain the model. And
what that does is it creates an

622
00:34:02,744 --> 00:34:05,799
ecosystem where you can
iteratively improve your model

623
00:34:05,799 --> 00:34:09,364
and start to get something that
maybe works 99% of the time, or

624
00:34:09,364 --> 00:34:12,985
99 plus percent of the time. And
then over time, it becomes more

625
00:34:12,985 --> 00:34:16,323
and more proprietary, you know,
custom built for your exact

626
00:34:16,323 --> 00:34:19,831
purpose because you're capturing
and feeding it in data that's

627
00:34:19,831 --> 00:34:23,000
from your actual, you know, you
know, business use case.

628
00:34:23,000 --> 00:34:26,306
Audrow Nash: Yeah. Very
interesting. It's cool to hear

629
00:34:26,306 --> 00:34:29,980
that you guys are. Using unity
for generating quite a bit of

630
00:34:29,980 --> 00:34:33,714
data. For, labeling data. Once
you start, once you're getting

631
00:34:33,714 --> 00:34:37,571
your own data, do you have any
workflows for that? Like, do you

632
00:34:37,571 --> 00:34:41,367
do you make it easy for people
to do it or they. Do, they just

633
00:34:41,367 --> 00:34:45,224
have to figure out how to grab
it within the relevant times and

634
00:34:45,224 --> 00:34:49,020
then label it. Or I imagine this
is a vertical you'll start to

635
00:34:49,020 --> 00:34:52,449
get to. But are you are you
working on it yet? Or how is

636
00:34:52,449 --> 00:34:53,000
labeling?

637
00:34:53,000 --> 00:34:56,212
Bradley Dillon: Yeah, we are so
it's a combination of things.

638
00:34:56,212 --> 00:34:59,582
The first is, is that we have an
in-house labeling team. They're

639
00:34:59,582 --> 00:35:02,953
actually located in Tunisia. So
in Tunisia and Tunis. Rehab. Oh,

640
00:35:02,953 --> 00:35:06,007
one of our machine learning
engineers. And we have a small

641
00:35:06,007 --> 00:35:08,903
group of folks that, help to
help to do the manual data

642
00:35:08,903 --> 00:35:11,799
labeling that's needed. So what
we do for those type of

643
00:35:11,799 --> 00:35:14,696
opportunities with our clients
is we put together, some

644
00:35:14,696 --> 00:35:17,539
training information over
exactly what we're trying to

645
00:35:17,539 --> 00:35:20,646
label in each image. And then
what we do is they, you know,

646
00:35:20,646 --> 00:35:23,490
they go out and they're
labeling, you know, thousands,

647
00:35:23,490 --> 00:35:26,649
tens of thousands of images. And
then we have mechanisms for

648
00:35:26,649 --> 00:35:29,440
being able to, you know,
manually and with technology

649
00:35:29,440 --> 00:35:32,389
audit, you know, to make sure
that we're labeling things

650
00:35:32,389 --> 00:35:35,707
right. If you don't label things
correctly, then you don't know

651
00:35:35,707 --> 00:35:39,024
the truth. And if you don't know
that, then you can't you can't

652
00:35:39,024 --> 00:35:42,395
create something that's going to
be useful. So that's that's why

653
00:35:42,395 --> 00:35:45,607
labeling of course, is is, you
know, is so critical. And then

654
00:35:45,607 --> 00:35:48,609
the other part that we use is
that when synthetic data is

655
00:35:48,609 --> 00:35:51,873
created and also this technology
is, is in some cases used for

656
00:35:51,873 --> 00:35:54,980
real world data, you can auto
label it. And so that's where

657
00:35:54,980 --> 00:35:58,193
you kind of tell the computer,
hey, this is the type of thing

658
00:35:58,193 --> 00:36:01,352
that I'm trying to label when
it's in unity. It's very clear

659
00:36:01,352 --> 00:36:04,406
because in unity, of course,
when the scenes build, you've

660
00:36:04,406 --> 00:36:07,513
labeled everything and it can
capture it and then it's real

661
00:36:07,513 --> 00:36:10,726
world data. You know, you in
some cases it can be work really

662
00:36:10,726 --> 00:36:13,990
well, you can say, okay, here's
how you want to auto label it.

663
00:36:13,990 --> 00:36:17,361
And in some cases you have to do
something that's pretty manual.

664
00:36:17,361 --> 00:36:20,573
So you put that all together and
it makes it so that for, you

665
00:36:20,573 --> 00:36:23,890
know, customers that need, you
know, tens of thousands, in some

666
00:36:23,890 --> 00:36:27,050
cases hundreds of thousands or
millions of images, you know,

667
00:36:27,050 --> 00:36:30,052
proprietary images that are
labeled for them, that that's

668
00:36:30,052 --> 00:36:32,000
how we go about helping them do
that.

669
00:36:32,000 --> 00:36:36,252
Audrow Nash: Cool. I did not
realize. So from what it sounds

670
00:36:36,252 --> 00:36:40,575
like to me, you guys are a lot
more service. Based than I had

671
00:36:40,575 --> 00:36:45,039
realized for this. Like, like.
Helping with data generation and

672
00:36:45,039 --> 00:36:49,079
simulation. And then also
helping with labeling with your

673
00:36:49,079 --> 00:36:53,472
labeling team. Is that a large
part of your revenue or is it a

674
00:36:53,472 --> 00:36:57,937
growing part? Because that that
seems very interesting, because

675
00:36:57,937 --> 00:37:02,331
it's like almost you could sell
the devices at cost. Say, and.

676
00:37:02,331 --> 00:37:06,583
Then have teams that help with.
Services that customers will

677
00:37:06,583 --> 00:37:10,622
need, like data generation and
labeling for this kind of.

678
00:37:10,622 --> 00:37:14,520
Thing. Is it a large part of
your business model? Is it

679
00:37:14,520 --> 00:37:17,000
growing? How do you think about
it?

680
00:37:17,000 --> 00:37:20,033
Bradley Dillon: So it is a
significant portion of our

681
00:37:20,033 --> 00:37:23,581
business model. And the way that
we think about it is, is that

682
00:37:23,581 --> 00:37:26,957
we're we're trying to get our
clients to get to where they,

683
00:37:26,957 --> 00:37:30,505
you know, have a solution that
they can deploy scale. And when

684
00:37:30,505 --> 00:37:33,938
we see that there's gaps in the
market, we work to fill them

685
00:37:33,938 --> 00:37:37,315
because, you know, we want to
get our help. Our clients get

686
00:37:37,315 --> 00:37:40,691
there as fast as possible. And
it creates this nice virtual

687
00:37:40,691 --> 00:37:44,067
virtuous cycle. And what that
cycle is, is that clients buy

688
00:37:44,067 --> 00:37:47,615
hardware. Then they're able to,
you know, use our services, to

689
00:37:47,615 --> 00:37:51,220
be able to get it so that that
hardware can do something useful

690
00:37:51,220 --> 00:37:54,711
for them both on the software
side and, you know, they pay us

691
00:37:54,711 --> 00:37:58,373
for that. And then in turn, when
that software works, well, then

692
00:37:58,373 --> 00:38:01,578
they go and they buy more
hardware. And so that's like a

693
00:38:01,578 --> 00:38:05,183
nice, very complimentary part of
our part of our revenue model.

694
00:38:05,183 --> 00:38:08,617
The majority of our revenue
today is from hardware sales. So

695
00:38:08,617 --> 00:38:12,222
okay, 80% something along those
lines. But we do also have, you

696
00:38:12,222 --> 00:38:15,541
know, a growing, you know,
software development portion as

697
00:38:15,541 --> 00:38:18,860
well. And so in some cases,
those are software tools where

698
00:38:18,860 --> 00:38:22,522
the clients are able to use the
tools on their own. And in other

699
00:38:22,522 --> 00:38:26,070
cases they're using those tools.
But we have, you know, teams.

700
00:38:26,070 --> 00:38:29,332
Yeah, we kind of think of as
like the enterprise title of

701
00:38:29,332 --> 00:38:32,937
that kind of software tool where
we have teams that are helping

702
00:38:32,937 --> 00:38:36,485
that kind of get get to their,
you know, moment where they can

703
00:38:36,485 --> 00:38:37,000
scale up.

704
00:38:37,000 --> 00:38:40,797
Audrow Nash: Do you imagine so
in say you've been around for

705
00:38:40,797 --> 00:38:44,532
six years, do you imagine in
four years from now that. This

706
00:38:44,532 --> 00:38:48,203
is like 80. Percent of your
revenue or, do you think it'll

707
00:38:48,203 --> 00:38:52,000
always stay a kind of small
thing or what would you imagine?

708
00:38:52,000 --> 00:38:57,000
Bradley Dillon: So I'll first
say we don't know for sure.

709
00:38:57,000 --> 00:38:58,000
Audrow Nash: Is it a relatively
new thing?

710
00:38:58,000 --> 00:39:01,330
Bradley Dillon: No, it's it's
not a it's not a relatively new

711
00:39:01,330 --> 00:39:04,550
thing. And it's something that
we talk a lot about. So I'll

712
00:39:04,550 --> 00:39:07,607
first say that we're here
flexible, on whatever is going

713
00:39:07,607 --> 00:39:10,936
to be the best path for us as an
organization. And, you know,

714
00:39:10,936 --> 00:39:14,375
we're we're exploring different
options of what that could look

715
00:39:14,375 --> 00:39:17,650
like. You know, one thing that's
out there, like you already

716
00:39:17,650 --> 00:39:20,979
mentioned, is you could, you
know, you could potentially sell

717
00:39:20,979 --> 00:39:24,418
the hardware at a lower price or
at no price. And then you have

718
00:39:24,418 --> 00:39:27,584
it kind of bundled as a, you
know, call it a platform as a

719
00:39:27,584 --> 00:39:31,077
service or something along those
lines. So that's something that

720
00:39:31,077 --> 00:39:34,461
we think would be a model that
could be a really enticing, you

721
00:39:34,461 --> 00:39:37,900
know, growth opportunity for us,
for certain clients where that

722
00:39:37,900 --> 00:39:41,284
fits into their operation. And
what it does is it removes kind

723
00:39:41,284 --> 00:39:44,614
of the, the, the way that they
can scale up with folks on us.

724
00:39:44,614 --> 00:39:47,834
It moves it from being, you
know, a CapEx to it to an OpEx.

725
00:39:47,834 --> 00:39:51,000
And that can really work well
for certain business models.

726
00:39:51,000 --> 00:39:52,000
Audrow Nash: So we think so too.
Yeah.

727
00:39:52,000 --> 00:39:55,744
Bradley Dillon: So we we are
open to exploring that that more

728
00:39:55,744 --> 00:39:59,550
you know, as, as we, as we grow,
I will say that we do believe

729
00:39:59,550 --> 00:40:03,110
that the software kind of
subscription portion is going to

730
00:40:03,110 --> 00:40:06,977
be a major contributor, to our
revenue long term. We don't know

731
00:40:06,977 --> 00:40:10,844
what the exact mixes, you know,
when we're, you know, ten scale

732
00:40:10,844 --> 00:40:14,527
right now. Is it, is it half, is
it, is it 20%. You know, we

733
00:40:14,527 --> 00:40:16,000
don't we don't know yet.

734
00:40:16,000 --> 00:40:17,000
Audrow Nash: It's hard to say.

735
00:40:17,000 --> 00:40:19,838
Bradley Dillon: Yeah. It's.
Yeah. And we're but you know we

736
00:40:19,838 --> 00:40:22,820
we ultimately believe that, you
know our value proposition is,

737
00:40:22,820 --> 00:40:25,754
is that, you know we have great
hardware that that, you know,

738
00:40:25,754 --> 00:40:28,833
helps make great that goes along
with great software that can do

739
00:40:28,833 --> 00:40:31,767
something useful. And so we're
kind of really we're trying to

740
00:40:31,767 --> 00:40:34,605
do everything we can to just
stay customer obsessed. And so

741
00:40:34,605 --> 00:40:37,395
like the reason why we've done
the things that I mentioned

742
00:40:37,395 --> 00:40:40,473
before on like data labeling and
some of these services, because

743
00:40:40,473 --> 00:40:43,311
we know it's a pain point for
our customers, we know we can

744
00:40:43,311 --> 00:40:46,342
solve it. And so we kind of work
backwards from that and tackle

745
00:40:46,342 --> 00:40:49,420
it. So we'll kind of remain kind
of with with our customers kind

746
00:40:49,420 --> 00:40:52,258
of, you know, hopefully get
guidance guiding the way for us

747
00:40:52,258 --> 00:40:55,000
to be able to, to navigate what
the exact mix of that is.

748
00:40:55,000 --> 00:40:58,685
Audrow Nash: Yeah. I think
that's really clever. And I

749
00:40:58,685 --> 00:41:02,506
think that that seems like a
very. Good model to me. And

750
00:41:02,506 --> 00:41:06,122
yeah, I guess the market will
pull you in the correct

751
00:41:06,122 --> 00:41:10,080
direction. So you allocate
accordingly. But. It seems like

752
00:41:10,080 --> 00:41:13,764
a nice thing because I'm
thinking of, say, startups. A

753
00:41:13,764 --> 00:41:18,131
thing that's hard with. Startups
is getting people that are very

754
00:41:18,131 --> 00:41:22,225
sophisticated, like, like super
experts in some narrow thing

755
00:41:22,225 --> 00:41:26,456
that your startup needs. And so
a lot of times it's. Easier to

756
00:41:26,456 --> 00:41:30,686
just hire for that kind of hire
out, those specific roles. And

757
00:41:30,686 --> 00:41:34,575
so it's interesting. It's like,
you guys already have the

758
00:41:34,575 --> 00:41:38,738
expertise. You're putting the
pipelines in place that make it

759
00:41:38,738 --> 00:41:42,627
so you are very efficient at
data labeling. You have your

760
00:41:42,627 --> 00:41:46,516
team allocated and you already
know how all your software

761
00:41:46,516 --> 00:41:50,474
works, so that it's very easy
for you to get the data. And

762
00:41:50,474 --> 00:41:54,773
then they don't have to like you
can help them bootstrap and go

763
00:41:54,773 --> 00:41:59,003
from 80% working to 99% or 99.9
and a few other nines working.

764
00:41:59,003 --> 00:42:03,029
And then that allows them to
deploy more and then they just

765
00:42:03,029 --> 00:42:07,123
buy more. And the in-house part
of. Training. Or the part of

766
00:42:07,123 --> 00:42:11,149
training their model has just
been completely outsourced to

767
00:42:11,149 --> 00:42:15,447
you guys, which would be really
cool. So I don't know, I'm just

768
00:42:15,447 --> 00:42:19,200
thinking that's a really cool
model, and you could take

769
00:42:19,200 --> 00:42:23,158
equity. In some of these.
Companies if they're very early,

770
00:42:23,158 --> 00:42:25,000
which would be a fun thing.

771
00:42:25,000 --> 00:42:28,128
Bradley Dillon: Yeah, you're
exactly right. I think these

772
00:42:28,128 --> 00:42:31,475
organizations are looking at,
you know, a classical trade off

773
00:42:31,475 --> 00:42:34,822
of like, do we make versus buy
and when they're, when they're

774
00:42:34,822 --> 00:42:37,895
looking at that, we like to
think that that usually when

775
00:42:37,895 --> 00:42:40,858
they see, you know, that they're
relatively, you know,

776
00:42:40,858 --> 00:42:44,205
competitive, you know, economic
price of our cameras and what

777
00:42:44,205 --> 00:42:47,498
that the platform can provide
that they say, yeah, this is a

778
00:42:47,498 --> 00:42:50,845
no brainer. There's other things
that maybe we want to try to

779
00:42:50,845 --> 00:42:54,302
make ourselves. But in this case
where we should just we should

780
00:42:54,302 --> 00:42:57,759
just buy these buy these devices
and the other thing is we want

781
00:42:57,759 --> 00:43:01,161
to we really want to control our
own destiny with our clients.

782
00:43:01,161 --> 00:43:04,618
And so when in a world where we
say, oh, we've solved this area

783
00:43:04,618 --> 00:43:07,800
here, over on this side, but we
have this other pain point

784
00:43:07,800 --> 00:43:11,367
that's blocking our clients from
getting further along. And so by

785
00:43:11,367 --> 00:43:14,605
being able to kind of control
those aspects, it makes it so

786
00:43:14,605 --> 00:43:18,007
that, you know, we we have the
capability to. Yeah, to get the

787
00:43:18,007 --> 00:43:21,409
client all the way to the finish
line. And that at the current

788
00:43:21,409 --> 00:43:24,591
stage of where, you know, the
intersection of robotics and

789
00:43:24,591 --> 00:43:27,939
artificial intelligence is I
think that's really important to

790
00:43:27,939 --> 00:43:31,450
us. You have a lot of customers
today that are really, you know,

791
00:43:31,450 --> 00:43:34,359
they're really pursuing,
investments in this area and

792
00:43:34,359 --> 00:43:37,706
exploring a lot of stuff. And
what it means is you have a lot

793
00:43:37,706 --> 00:43:41,218
of pilots and, you know, is this
possible? Is that possible? And

794
00:43:41,218 --> 00:43:44,784
we're we're really obsessed with
trying to be able to have really

795
00:43:44,784 --> 00:43:48,296
high conversion rate from those
pilots to to actually, you know,

796
00:43:48,296 --> 00:43:51,588
going at scale. Of course, some
of that is selfish. Like, we

797
00:43:51,588 --> 00:43:54,936
know we want to be able to grow
our revenue in a big way. But

798
00:43:54,936 --> 00:43:58,228
it's also like about what's
rewarding, it's not rewarding to

799
00:43:58,228 --> 00:44:01,465
work on pilots all the time.
What's really rewarding is you

800
00:44:01,465 --> 00:44:04,099
actually get to see that
innovation, use it, use

801
00:44:04,099 --> 00:44:07,392
intensely at scale. And so for
our engineers, definitely the

802
00:44:07,392 --> 00:44:10,519
engineers I mentioned, the most
exciting part is when you

803
00:44:10,519 --> 00:44:13,976
actually get to go, oh my gosh,
like, look, look, look at this.

804
00:44:13,976 --> 00:44:17,378
That's being used out there in
the world. So that's, you know,

805
00:44:17,378 --> 00:44:20,616
we're we're we're really kind of
obsessed with that part of

806
00:44:20,616 --> 00:44:21,000
things.

807
00:44:21,000 --> 00:44:24,776
Audrow Nash: Totally. Yeah. For
so like, I mean, I've been at

808
00:44:24,776 --> 00:44:28,551
Open Robotics and then we were
acquired by intrinsic. But why

809
00:44:28,551 --> 00:44:32,574
also Open. Robotics does ROS And
I think that was really exciting

810
00:44:32,574 --> 00:44:36,226
is when iRobot announced that
they were using ROS 2. And so

811
00:44:36,226 --> 00:44:40,064
now ROS two is on a whole bunch
of robots all around. And that

812
00:44:40,064 --> 00:44:43,716
was just like the biggest win
for our team for this kind of

813
00:44:43,716 --> 00:44:47,306
thing. So I can imagine it's
like, yeah, when when you see

814
00:44:47,306 --> 00:44:50,957
people using what you're making
really productively, that's

815
00:44:50,957 --> 00:44:53,000
super, super cool and
energizing.

816
00:44:53,000 --> 00:44:55,855
Bradley Dillon: Yeah,
absolutely. And I'll say that,

817
00:44:55,855 --> 00:44:59,259
you know, the world is the robot
change in such an interesting

818
00:44:59,259 --> 00:45:02,718
way on our journey. Just just to
kind of add in here, I kind of

819
00:45:02,718 --> 00:45:05,903
think of it as, you know, 2022
is kind of this like before

820
00:45:05,903 --> 00:45:09,143
Christ, after Christ moment for
artificial intelligence and

821
00:45:09,143 --> 00:45:12,602
before 2020, you know, 2023 is
things kind of really scaled up.

822
00:45:12,602 --> 00:45:15,841
Their organizations were not
asking themselves, how can we,

823
00:45:15,841 --> 00:45:18,696
you know, how can we use
artificial intelligence and

824
00:45:18,696 --> 00:45:22,046
robotics? They weren't doing
that from the top very often. It

825
00:45:22,046 --> 00:45:25,395
was coming very bottom. Some
engineers were saying, you know,

826
00:45:25,395 --> 00:45:28,744
baking their leadership. Hey
there. You know, there's so much

827
00:45:28,744 --> 00:45:32,039
that's possible here. Let's lean
in and invest. And now that

828
00:45:32,039 --> 00:45:35,553
we're in this kind of like after
Christ moment, excuse that. You

829
00:45:35,553 --> 00:45:38,847
know, the religious analogy is
that. Yeah, organizations are

830
00:45:38,847 --> 00:45:42,032
now doing it from a very top
down perspective. And they're

831
00:45:42,032 --> 00:45:45,546
asking themselves by their their
leadership teams or executives,

832
00:45:45,546 --> 00:45:48,565
their board of directors or
saying are investing enough

833
00:45:48,565 --> 00:45:52,025
here? Are we moving fast enough?
How can we do more? How can we

834
00:45:52,025 --> 00:45:55,429
try things? And so that's been
really cool in the conversation

835
00:45:55,429 --> 00:45:58,229
that a lot more of our
discussions today where, you

836
00:45:58,229 --> 00:46:01,578
know, circa 2022, it was like
100% versus engineers. That was

837
00:46:01,578 --> 00:46:04,763
all we talked to. You work
sales. It's now become honestly

838
00:46:04,763 --> 00:46:08,167
like a 5050 split where you have
you have a lot of people that

839
00:46:08,167 --> 00:46:11,407
are, you know, more more on the
leadership top-down side of

840
00:46:11,407 --> 00:46:14,756
things. Now the Bottoms Up is
still so excited about it, too.

841
00:46:14,756 --> 00:46:18,215
They're driving it too. So it's
been interesting to see the way

842
00:46:18,215 --> 00:46:21,509
the world's the world's change
there. And that really became

843
00:46:21,509 --> 00:46:24,694
solidified in 2024. And that
momentum is really continuing

844
00:46:24,694 --> 00:46:27,000
into 2025 which is really
exciting to see.

845
00:46:27,000 --> 00:46:31,331
Audrow Nash: How do you think it
will. So okay 2022 2023. Three

846
00:46:31,331 --> 00:46:35,388
2024 2025. How do you think
it'll go. For the rest of like,

847
00:46:35,388 --> 00:46:39,306
how do you predict that that
trend will continue? Or like

848
00:46:39,306 --> 00:46:43,500
where do you think it'll go. So
like 2026 2027. Like where do

849
00:46:43,500 --> 00:46:47,556
you what do you see it going?
Maybe 2030. All these things.

850
00:46:47,556 --> 00:46:49,000
Like what do you bet.

851
00:46:49,000 --> 00:46:52,423
Bradley Dillon: Yeah. It's a
great question how we we think

852
00:46:52,423 --> 00:46:56,078
about this a lot. You know, I, I
think that. The example that I

853
00:46:56,078 --> 00:46:59,617
think is most, tangible to sort
of think about is probably in

854
00:46:59,617 --> 00:47:02,923
humanoid robotics and with
humanoids, if you think about,

855
00:47:02,923 --> 00:47:06,694
you know, what's the, what's the
total addressable market for it?

856
00:47:06,694 --> 00:47:10,233
And the reality is, like in a
place like the United States, I

857
00:47:10,233 --> 00:47:13,308
think I saw a research study
that said, you know, the

858
00:47:13,308 --> 00:47:16,847
addressable market should be
like 400 million of them. And so

859
00:47:16,847 --> 00:47:20,328
I think humanoids, what they
would enable is, is that if you

860
00:47:20,328 --> 00:47:23,983
have a humanoid that can do you
know, pretty much anything that

861
00:47:23,983 --> 00:47:27,580
a human can do, which is, you
know, starting to seem feasible,

862
00:47:27,580 --> 00:47:31,060
you know, call it by the end of
this decade, then you really

863
00:47:31,060 --> 00:47:34,425
gets at this kind of moment
that's sort of similar to like

864
00:47:34,425 --> 00:47:38,080
the computer revolution of all
of a sudden everybody had a, you

865
00:47:38,080 --> 00:47:41,619
know, a laptop on every desk. I
think they said Microsoft was

866
00:47:41,619 --> 00:47:45,274
founded with the idea that, you
know, Bill Gates and Paul Allen

867
00:47:45,274 --> 00:47:48,697
had is just a laptop and every
desk that was there and that

868
00:47:48,697 --> 00:47:52,178
that became a reality. Right.
And so I think with humanoids,

869
00:47:52,178 --> 00:47:55,426
you have this, this possibility
that it's a, you know, a

870
00:47:55,426 --> 00:47:58,965
humanoid in every home and a
humanoid in every workplace. And

871
00:47:58,965 --> 00:48:02,156
so I think that that's where you
can kind of see things

872
00:48:02,156 --> 00:48:05,695
projecting out to is, I think,
embodied AI and where you have

873
00:48:05,695 --> 00:48:09,350
you have these kind of this Uber
multimodal model that can kind

874
00:48:09,350 --> 00:48:12,657
of just have this whole world
view as kind of world of it

875
00:48:12,657 --> 00:48:16,370
could do, you know, behavior and
vision and, you know, you know,

876
00:48:16,370 --> 00:48:19,967
sound and, you know, voice and
you know, all of the above. And

877
00:48:19,967 --> 00:48:22,983
I think that's where the
intersection of all of this

878
00:48:22,983 --> 00:48:26,000
technology is kind of converging
in an exciting way.

879
00:48:26,000 --> 00:48:30,184
Audrow Nash: Yeah. It is really
exciting. It's very interesting

880
00:48:30,184 --> 00:48:33,837
to me. So like. Being in
robotics over the last like 15

881
00:48:33,837 --> 00:48:37,954
years or. So. And it's just wild
how much things have changed.

882
00:48:37,954 --> 00:48:41,740
It's like what I learned in
school has like it's just not

883
00:48:41,740 --> 00:48:45,924
the way we do things anymore for
a lot of things, which is just

884
00:48:45,924 --> 00:48:49,843
absolutely crazy. So it's moving
fast and it's changing and

885
00:48:49,843 --> 00:48:53,894
better hardware like you guys
are making makes it much easier

886
00:48:53,894 --> 00:48:58,211
to do things like the barrier to
entry is decreasing. The ability

887
00:48:58,211 --> 00:49:01,000
to do stuff is increasing. It's
just wild.

888
00:49:01,000 --> 00:49:04,578
Bradley Dillon: Yeah, it's it's
really exciting. 20 years ago,

889
00:49:04,578 --> 00:49:08,041
during my undergrad, I studied
mechanical engineering and my

890
00:49:08,041 --> 00:49:11,676
kind of, focused kind of all the
extra classes I could take was

891
00:49:11,676 --> 00:49:14,966
in mechatronics and robotics.
And so we built robots that

892
00:49:14,966 --> 00:49:18,255
played dodgeball against each
other. We built robots that

893
00:49:18,255 --> 00:49:21,487
played Capture the Flag. You
know, we did things so, you

894
00:49:21,487 --> 00:49:25,008
know, the inverted pendulum and
all this cool stuff. And, you

895
00:49:25,008 --> 00:49:28,470
know, it's funny, you know,
thinking back to that, you know,

896
00:49:28,470 --> 00:49:32,106
20, 20 years ago, where we come
to today and, you know, getting

897
00:49:32,106 --> 00:49:35,626
to see, for example, like Waymo
cars scale up with autonomous

898
00:49:35,626 --> 00:49:39,147
driving and thinking back to,
you know, this autonomous robot

899
00:49:39,147 --> 00:49:42,667
that like, you know, you pointed
a sensor at the table and it

900
00:49:42,667 --> 00:49:46,187
could sense, you know, black
tape and, you know, that was how

901
00:49:46,187 --> 00:49:49,939
it could go. It was kind of like
back then the idea that, oh, you

902
00:49:49,939 --> 00:49:53,401
could translate this all the way
up to say, having a totally

903
00:49:53,401 --> 00:49:56,633
autonomous, you know, 5,000
pound car barreling down the

904
00:49:56,633 --> 00:50:00,269
road at 70 miles an hour, it's
it's it's really exciting to see

905
00:50:00,269 --> 00:50:02,000
the see the progress for sure.

906
00:50:02,000 --> 00:50:06,419
Audrow Nash: So, one thing that
was pretty exciting to me. With

907
00:50:06,419 --> 00:50:10,417
ChatGPT was their. Vision
capabilities. So like you could

908
00:50:10,417 --> 00:50:14,415
put a picture in. And it will
have. Fairly. Sophisticated

909
00:50:14,415 --> 00:50:18,694
scene understanding, at least by
what it's able to write. For

910
00:50:18,694 --> 00:50:22,973
this kind of thing. So like I
really like that kind of thing.

911
00:50:22,973 --> 00:50:27,252
You guys, being that you have
cameras and compute, I imagine,

912
00:50:27,252 --> 00:50:31,460
are well-placed to do that kind
of thing and have it locally

913
00:50:31,460 --> 00:50:35,388
because you have powerful
compute. Tell me about some of

914
00:50:35,388 --> 00:50:39,527
these newer AI techniques. If
you're leveraging that all or

915
00:50:39,527 --> 00:50:41,000
enabling them at all.

916
00:50:41,000 --> 00:50:44,399
Bradley Dillon: This is an area
that we've been doing a lot of

917
00:50:44,399 --> 00:50:47,797
exploratory research, talking,
talking to some different folks

918
00:50:47,797 --> 00:50:50,867
and, you know, evaluating it
internally. Some of the use

919
00:50:50,867 --> 00:50:54,211
cases that we think are pretty
interesting is, is that if you

920
00:50:54,211 --> 00:50:57,390
have a, you know, you have a
video, a bunch of sensor data

921
00:50:57,390 --> 00:51:00,789
that's streaming off of a device
that, you know, with, with a,

922
00:51:00,789 --> 00:51:03,859
you know, with an API and or a
simple, you know, natural

923
00:51:03,859 --> 00:51:07,093
language interface. You could
maybe ask you a question, you

924
00:51:07,093 --> 00:51:10,327
could say, like, describe what
you're seeing right now. You

925
00:51:10,327 --> 00:51:13,342
could potentially say, hey, I
want to can you send me a

926
00:51:13,342 --> 00:51:16,740
notification every time you see
something like this? You know,

927
00:51:16,740 --> 00:51:20,194
you you see somebody fall down
now? Right now, if you wanted to

928
00:51:20,194 --> 00:51:23,154
create, say, a system that
detects when somebody fell,

929
00:51:23,154 --> 00:51:26,333
maybe you'd have to create a
model or something like that.

930
00:51:26,333 --> 00:51:29,896
But behind the tools you make it
so that maybe you just say, hey,

931
00:51:29,896 --> 00:51:33,240
just let me know. The last time
in this video stream that you

932
00:51:33,240 --> 00:51:36,529
saw somebody fall down. And so
it makes it so that you could

933
00:51:36,529 --> 00:51:40,092
grab real time data, but it also
makes it so you could search the

934
00:51:40,092 --> 00:51:43,601
data too. So you could say, hey,
I've got, you know, 24 hours of

935
00:51:43,601 --> 00:51:46,944
video saved. Maybe it's, maybe
you're a construction site and

936
00:51:46,944 --> 00:51:50,288
you're the safety manager and
you say, can you, can you point

937
00:51:50,288 --> 00:51:53,577
me to the points in the last 24
hours when, we saw something

938
00:51:53,577 --> 00:51:56,866
that was dangerous, that that
was happening, something where

939
00:51:56,866 --> 00:51:59,991
somebody wasn't wearing their
safety equipment, you know,

940
00:51:59,991 --> 00:52:02,951
something like that. And so I
think that that's really

941
00:52:02,951 --> 00:52:06,075
interesting for us. We do
believe with the, you know, the

942
00:52:06,075 --> 00:52:09,364
cloud application layer that we
have that we can enable that

943
00:52:09,364 --> 00:52:12,818
type of technology pretty well.
And we're kind of exploring the

944
00:52:12,818 --> 00:52:16,052
ways that that can be, I guess
we'll call it AI is a really

945
00:52:16,052 --> 00:52:19,286
meaningful step in the journey
to really democratizing, you

946
00:52:19,286 --> 00:52:22,575
know, computer vision, making it
so that something like that

947
00:52:22,575 --> 00:52:25,864
would be pretty difficult for an
engineer to do classically.

948
00:52:25,864 --> 00:52:29,153
Definitely. Where it could be
something where. Yeah. I mean,

949
00:52:29,153 --> 00:52:32,552
any anybody that's even not, not
a coder would be able to, you

950
00:52:32,552 --> 00:52:36,115
know, grab some, you know, grab,
grab some cameras and just start

951
00:52:36,115 --> 00:52:39,349
to start doing some natural
language prompts and, you know,

952
00:52:39,349 --> 00:52:42,748
find out what's going on and set
things up. So it's definitely

953
00:52:42,748 --> 00:52:45,817
exciting, something we're
exploring. There's some really

954
00:52:45,817 --> 00:52:49,106
promising new companies and
startups that are tackling this.

955
00:52:49,106 --> 00:52:52,395
I know, like a lot of, latest
crop of startups over the last

956
00:52:52,395 --> 00:52:55,191
couple of years have been
founded around, you know,

957
00:52:55,191 --> 00:52:57,000
working on this type of use
case.

958
00:52:57,000 --> 00:53:00,862
Audrow Nash: Cool. Yeah. What a
nice thing that is. I mean, I

959
00:53:00,862 --> 00:53:04,723
haven't thought. Of it, but the
ability to search data really

960
00:53:04,723 --> 00:53:08,711
quick, like search video really
quickly would be incredible. So

961
00:53:08,711 --> 00:53:12,636
if you could say, hey, take me
to the moment this is happening

962
00:53:12,636 --> 00:53:16,751
and it goes, oh yeah, yeah, here
you go. That would be awesome. I

963
00:53:16,751 --> 00:53:20,739
wonder, like one of the things
that I thought was. Really cool,

964
00:53:20,739 --> 00:53:24,601
and I built a little prototype
of this for some side projects

965
00:53:24,601 --> 00:53:28,399
was you take an image and then
you have it select from a few

966
00:53:28,399 --> 00:53:32,008
app options. So it's kind of
like. Image. Does all of its

967
00:53:32,008 --> 00:53:35,869
thinking. And then it selects
like one of a few options which

968
00:53:35,869 --> 00:53:39,667
might be like drive a robot
forward, turn around, pick up an

969
00:53:39,667 --> 00:53:43,212
object, whatever. I feel like
this kind of thing is very

970
00:53:43,212 --> 00:53:46,504
useful to just for like
prototyping stuff. Or even I

971
00:53:46,504 --> 00:53:50,303
mean, probably. They're real
simple applications that people

972
00:53:50,303 --> 00:53:53,658
could have, or even more
sophisticated ones, that you

973
00:53:53,658 --> 00:53:56,000
could do with this kind of
framework.

974
00:53:56,000 --> 00:53:58,741
Bradley Dillon: Yeah,
absolutely. I know that, you

975
00:53:58,741 --> 00:54:02,030
know, one really cool video
that's out there to say the, the

976
00:54:02,030 --> 00:54:05,264
figure AI humanoid and, you
know, it's looking at the table

977
00:54:05,264 --> 00:54:08,553
and it says, you know, what do
you think you should do next?

978
00:54:08,553 --> 00:54:11,678
And there's like a dirty dish
and it's like, oh, I should

979
00:54:11,678 --> 00:54:15,186
probably put the dirty dish and
put it in the dishwasher. And so

980
00:54:15,186 --> 00:54:18,147
I think that's really
interesting when you think about

981
00:54:18,147 --> 00:54:21,545
kind of these behavior models,
for robotics companies that are

982
00:54:21,545 --> 00:54:24,889
working on kind of prototyping
what's, what's possible. And I

983
00:54:24,889 --> 00:54:28,288
think that's where that, that
real time understanding of video

984
00:54:28,288 --> 00:54:31,577
can, can be really useful. So
that you'd be able to practice

985
00:54:31,577 --> 00:54:34,921
different prompts based on the
scene that you're seeing. What

986
00:54:34,921 --> 00:54:38,320
are the three next steps that
you could see that you'd want to

987
00:54:38,320 --> 00:54:41,444
take. And you know, a way that
you can really start to to

988
00:54:41,444 --> 00:54:44,514
learn. And what it really
tackles is, is that, you know,

989
00:54:44,514 --> 00:54:47,529
robotics have been used for
decades in, in really think

990
00:54:47,529 --> 00:54:51,038
scaled environments, but they're
very controlled. Right. So this

991
00:54:51,038 --> 00:54:54,491
could be a, you know, say like,
automobile manufacturing plant.

992
00:54:54,491 --> 00:54:57,725
It's doing the same tasks a
million times over and over. It

993
00:54:57,725 --> 00:55:01,069
doesn't even need vision, right?
It's doing it on a precision

994
00:55:01,069 --> 00:55:04,468
basis. You're taking this item,
you're putting it here, things

995
00:55:04,468 --> 00:55:07,976
like that. And when you go into
these unstructured environments,

996
00:55:07,976 --> 00:55:11,156
that's where it's of course
really challenging because the

997
00:55:11,156 --> 00:55:14,609
scene is always changing. It's
it's navigating something that's

998
00:55:14,609 --> 00:55:17,624
never happened before. It's a
it's a scene that's never

999
00:55:17,624 --> 00:55:20,749
existed in all of history
because it's completely unique.

1000
00:55:20,749 --> 00:55:24,038
And that's and that's where I
think that, yeah. Like I said,

1001
00:55:24,038 --> 00:55:26,998
the real time, real time
understanding could be really

1002
00:55:26,998 --> 00:55:30,013
interesting. I'm kind of
prompting those next steps and

1003
00:55:30,013 --> 00:55:31,000
and what have you.

1004
00:55:31,000 --> 00:55:35,764
Audrow Nash: Do you, I don't
really know. Well, but maybe.

1005
00:55:35,764 --> 00:55:40,281
Because you guys have been
looking into it. Do you know

1006
00:55:40,281 --> 00:55:45,455
much about how it is to do these
scene understanding tasks with

1007
00:55:45,455 --> 00:55:49,480
video for this? Like I'm
imagining. Trying to use

1008
00:55:49,480 --> 00:55:54,572
Transformers or something. Some
AI technology on it was trying

1009
00:55:54,572 --> 00:55:59,171
to find features that are
relevant and look at them over

1010
00:55:59,171 --> 00:56:03,853
time. It sounds really hard to
me, but, it's also kind of

1011
00:56:03,853 --> 00:56:08,863
magic. Do you do you have any,
like, thoughts on how feasible

1012
00:56:08,863 --> 00:56:13,955
it is? Because right now, like
OpenAI's chat GPT. I think just

1013
00:56:13,955 --> 00:56:18,965
as images, maybe it does videos,
but, What what's kind of the

1014
00:56:18,965 --> 00:56:23,565
state of the art in language
models with. Video or video

1015
00:56:23,565 --> 00:56:28,000
models or something? Video,
large models or something?

1016
00:56:28,000 --> 00:56:31,612
Bradley Dillon: You know, I use
ChatGPT a lot. It's really cool.

1017
00:56:31,612 --> 00:56:34,943
I got these hikes yesterday. I
was on a hike and I saw this

1018
00:56:34,943 --> 00:56:38,442
footprint and I didn't. Yeah,
animal print. I didn't recognize

1019
00:56:38,442 --> 00:56:41,942
it. I just took a picture and I
said, hey, ChatGPT. What? What

1020
00:56:41,942 --> 00:56:45,554
do you animal do you think this
is? And it looked at and it said

1021
00:56:45,554 --> 00:56:48,997
described it and it said, we
think this is a deer and this is

1022
00:56:48,997 --> 00:56:52,440
why like, oh, it's so it's so
cool. And it, you know, it does

1023
00:56:52,440 --> 00:56:55,714
do it does do videos. Well there
it does. Well, what we've

1024
00:56:55,714 --> 00:56:58,819
learned is that there's
different or I believe it does.

1025
00:56:58,819 --> 00:57:01,810
Video of what we've learned is
that there's different

1026
00:57:01,810 --> 00:57:05,310
techniques that the different
companies are using to try to be

1027
00:57:05,310 --> 00:57:08,809
able to do it. The real question
mark for us is whether we can

1028
00:57:08,809 --> 00:57:11,914
now or in the future, have
hardware or could run on the

1029
00:57:11,914 --> 00:57:15,357
edge and I think that what that
could entail is, is that, you

1030
00:57:15,357 --> 00:57:18,856
know, right now the artificial
intelligence, you know, purpose

1031
00:57:18,856 --> 00:57:21,735
built harbor that we have is
primarily around doing

1032
00:57:21,735 --> 00:57:25,122
inference. So, you know, you
have a model and it's just kind

1033
00:57:25,122 --> 00:57:28,508
of inferring and operating, the
what we've seen at seen, but

1034
00:57:28,508 --> 00:57:32,008
what we potentially could do is
we could have, you know, let's

1035
00:57:32,008 --> 00:57:35,564
also enable like Chennai on the
edge. And so that's where it's,

1036
00:57:35,564 --> 00:57:38,668
you know, you know, you know,
generating something. And

1037
00:57:38,668 --> 00:57:41,604
there's, Alec as it stands
today, we're probably for

1038
00:57:41,604 --> 00:57:44,708
enabling something like this.
It's probably going to be

1039
00:57:44,708 --> 00:57:47,925
enabled with, say, something
that's done on the cloud. So

1040
00:57:47,925 --> 00:57:51,256
that's really the kind of a
question that we have as we try

1041
00:57:51,256 --> 00:57:54,812
to navigate this is, is that is
this something where we want to

1042
00:57:54,812 --> 00:57:58,368
do that? Now when I use a when I
did that, the example I shared

1043
00:57:58,368 --> 00:58:01,698
with ChatGPT that took, I think,
you know, 30 or 45 seconds

1044
00:58:01,698 --> 00:58:05,254
because you know that that's you
know, that that prompt is that

1045
00:58:05,254 --> 00:58:08,753
it is run in the cloud or it's
processed somewhere. So kind of

1046
00:58:08,753 --> 00:58:12,366
the question is, is like whether
it's a, you know, we're talking

1047
00:58:12,366 --> 00:58:15,922
about like my latest smartphone
or we're talking about like our

1048
00:58:15,922 --> 00:58:19,026
latest devices, is is it
possible to be able to do that

1049
00:58:19,026 --> 00:58:22,131
on the edge? Because that's
where our value proposition

1050
00:58:22,131 --> 00:58:25,800
comes up, because then you could
have it be lower latency. That's

1051
00:58:25,800 --> 00:58:29,299
where latency really matters,
right? Is what what's a way that

1052
00:58:29,299 --> 00:58:32,968
you could do that in like a very
real time basis versus something

1053
00:58:32,968 --> 00:58:35,000
that maybe would have more of a
lag.

1054
00:58:35,000 --> 00:58:38,737
Audrow Nash: Yeah I see so
focusing on the edge, it's not,

1055
00:58:38,737 --> 00:58:42,602
it's not clear if it's in scope
or. When it will be in scope

1056
00:58:42,602 --> 00:58:46,532
because you never know. I mean
some of the models. Seem to be

1057
00:58:46,532 --> 00:58:50,527
getting lighter and lighter, but
you never know if it'll be as

1058
00:58:50,527 --> 00:58:54,006
sophisticated or if it'll be
able to do vision related

1059
00:58:54,006 --> 00:58:58,000
things. I don't know how large
any of these vision models are.

1060
00:58:58,000 --> 00:59:01,159
Bradley Dillon: Yeah, it's
exactly right. And so for for

1061
00:59:01,159 --> 00:59:04,374
us, it's it's all about like,
what do we do best? And you

1062
00:59:04,374 --> 00:59:07,872
know, for us it makes sense to
enable this sort of technology.

1063
00:59:07,872 --> 00:59:11,256
If we can, we can ultimately,
you know, maybe not initially,

1064
00:59:11,256 --> 00:59:14,641
but we know that we have a path
or we can enable that on the

1065
00:59:14,641 --> 00:59:17,687
edge. Then that's where it
becomes something that gets

1066
00:59:17,687 --> 00:59:21,185
becomes really compelling for
us. And so that's the that's the

1067
00:59:21,185 --> 00:59:24,569
main lens that we're kind of
evaluating it through. And then

1068
00:59:24,569 --> 00:59:28,123
putting that aside like the holy
cow wow factor is so high that

1069
00:59:28,123 --> 00:59:31,000
test testing it out us is, you
know, well worth it.

1070
00:59:31,000 --> 00:59:35,594
Audrow Nash: Yeah. Hell yeah. So
one thing that's interesting to

1071
00:59:35,594 --> 00:59:39,830
me, especially with thinking
about you guys having more of,

1072
00:59:39,830 --> 00:59:44,209
like service or where you're
working with a lot of customers.

1073
00:59:44,209 --> 00:59:48,516
Do you ever, like, I imagine you
come across a lot of, like,

1074
00:59:48,516 --> 00:59:52,608
really. Good use cases that
you're very well suited. For.

1075
00:59:52,608 --> 00:59:56,772
It's an interesting thing
because I imagine if I was in a.

1076
00:59:56,772 --> 01:00:01,223
Similar position to you, I would
be distracted by all the cool

1077
01:00:01,223 --> 01:00:05,602
applications that are coming up.
And I would want to go run a

1078
01:00:05,602 --> 01:00:09,407
startup or have something
related to each like of the

1079
01:00:09,407 --> 01:00:14,001
like, really promising problems
that you come across. How do you

1080
01:00:14,001 --> 01:00:17,806
think about this? And, I know
you're working with the

1081
01:00:17,806 --> 01:00:22,185
customers, but, I don't know.
Just tell me about all the cool

1082
01:00:22,185 --> 01:00:26,564
domains. That you're coming in
touch with. And, how you think

1083
01:00:26,564 --> 01:00:28,000
about it, I suppose.

1084
01:00:28,000 --> 01:00:32,063
Bradley Dillon: So we have a lot
of very spirited debates on this

1085
01:00:32,063 --> 01:00:36,000
topic, as you can imagine. And
you would would it really? Yeah.

1086
01:00:36,000 --> 01:00:38,943
Audrow Nash: We have the
expertise. You have the

1087
01:00:38,943 --> 01:00:42,255
technology, you have the
manufacturing. Sometimes it's

1088
01:00:42,255 --> 01:00:45,750
like just if you were to run.
With it, you're in the best

1089
01:00:45,750 --> 01:00:49,000
position to capitalize on it is
what I would imagine.

1090
01:00:49,000 --> 01:00:52,652
Bradley Dillon: Yeah. That's
right. And so what it boils down

1091
01:00:52,652 --> 01:00:56,364
to is, you know what what what's
our focus? What are the areas

1092
01:00:56,364 --> 01:01:00,196
where we think that we have, you
know, an unfair advantage to be

1093
01:01:00,196 --> 01:01:03,608
able to pursue them and scale up
with our clients and the

1094
01:01:03,608 --> 01:01:07,080
industries that we work on
today. I'll kind of put them in

1095
01:01:07,080 --> 01:01:10,732
a few different buckets. So one
of them is, you know, kind of

1096
01:01:10,732 --> 01:01:14,325
transportation logistics and
warehousing. And another one is

1097
01:01:14,325 --> 01:01:17,558
industrial equipment and
agriculture, like I mentioned

1098
01:01:17,558 --> 01:01:21,210
earlier. And then another area
is like retail and quick serve

1099
01:01:21,210 --> 01:01:24,802
restaurants. And then kind of
some final buckets is call it,

1100
01:01:24,802 --> 01:01:28,334
you know, construction and
manufacturing and at any time in

1101
01:01:28,334 --> 01:01:32,106
our journey we probably could it
could just have already picked

1102
01:01:32,106 --> 01:01:35,518
one of those and just said,
we're going to ignore all the

1103
01:01:35,518 --> 01:01:38,811
rest, and we're just going to
get stuck in the narrowly

1104
01:01:38,811 --> 01:01:42,523
focused all in on this. And I
will say that, like, the coolest

1105
01:01:42,523 --> 01:01:46,055
part of the job is, is getting
to learn across all of these

1106
01:01:46,055 --> 01:01:49,468
different cases. So, like,
selfishly, I'd be a little sad

1107
01:01:49,468 --> 01:01:53,060
if we did it all in one area,
but very agnostic to to to it.

1108
01:01:53,060 --> 01:01:56,832
And really it's about like if we
think that our innovations can

1109
01:01:56,832 --> 01:02:00,244
be used most intensely in one
area, then, then we want to

1110
01:02:00,244 --> 01:02:03,836
focus there. So we do have some
spots where we are trying to

1111
01:02:03,836 --> 01:02:07,009
kind of lean more into, because
we're seeing that our

1112
01:02:07,009 --> 01:02:10,661
investments are paying off in
the way that we want. And as we

1113
01:02:10,661 --> 01:02:14,493
look ahead, we're really trying
to make it so that, you know, we

1114
01:02:14,493 --> 01:02:18,265
we probably have, you know, I'd
say probably no more than three

1115
01:02:18,265 --> 01:02:22,036
kind of concentrated bets. And,
and areas that we think that we

1116
01:02:22,036 --> 01:02:25,748
can, you know, continue to get
the most traction. And so a lot

1117
01:02:25,748 --> 01:02:29,460
of the first six years of our
journey is kind of been figuring

1118
01:02:29,460 --> 01:02:33,172
out these different industries,
figuring out how they operate,

1119
01:02:33,172 --> 01:02:36,944
you know, figuring out how our
technology can help. And then as

1120
01:02:36,944 --> 01:02:40,596
you look ahead to, you know,
knock on wood, hopefully call it

1121
01:02:40,596 --> 01:02:44,308
our next six years, it's it's
really probably, been a bit more

1122
01:02:44,308 --> 01:02:47,840
narrowed saying, you know,
here's the like I said, probably

1123
01:02:47,840 --> 01:02:51,133
three, maybe like 4 or 5 match,
but probably just a few

1124
01:02:51,133 --> 01:02:54,785
industries. We're like, yeah,
this is where we want to go all

1125
01:02:54,785 --> 01:02:57,000
and and focus the technology to
Mars.

1126
01:02:57,000 --> 01:02:59,929
Audrow Nash: What are those.
What are those areas for you.

1127
01:02:59,929 --> 01:03:02,000
The three areas you want to make
bets on.

1128
01:03:02,000 --> 01:03:05,514
Bradley Dillon: Well I, you know
I share with you kind of the,

1129
01:03:05,514 --> 01:03:08,688
the main industries that we're
in today and I think well

1130
01:03:08,688 --> 01:03:12,031
without a doubt would be, you
know, be kind of one of those

1131
01:03:12,031 --> 01:03:15,319
for sure. Or some of those for
sure. So I'd say any one of

1132
01:03:15,319 --> 01:03:18,492
them, to be honest. Like we
could go deep on it. I could

1133
01:03:18,492 --> 01:03:22,176
explain why. I think it could be
like a very bullish case of like

1134
01:03:22,176 --> 01:03:25,690
why we where we could go after
it. And I guess I'll, you know,

1135
01:03:25,690 --> 01:03:29,034
answer today is saying that
like, we don't we we don't have

1136
01:03:29,034 --> 01:03:32,661
three that we've picked yet. We
were starting to have, you know,

1137
01:03:32,661 --> 01:03:36,288
really strong hypotheses on why,
why, why we think some might be

1138
01:03:36,288 --> 01:03:39,689
better than the others. But
don't have enough to, don't have

1139
01:03:39,689 --> 01:03:42,863
enough, you know, data and
conviction to day to say, oh,

1140
01:03:42,863 --> 01:03:46,433
this is the, you know, this is
what we know for sure. And to go

1141
01:03:46,433 --> 01:03:47,000
all in on.

1142
01:03:47,000 --> 01:03:51,075
Audrow Nash: Yeah. It's an
interesting thing because I mean

1143
01:03:51,075 --> 01:03:55,426
you guys are well positioned. I
just like the, the thing that I

1144
01:03:55,426 --> 01:03:59,432
would think is it's like a lot
of companies. They'll build

1145
01:03:59,432 --> 01:04:03,299
something general on their way
to finding something very

1146
01:04:03,299 --> 01:04:07,443
specific. And then, then they go
hard at the really specific

1147
01:04:07,443 --> 01:04:11,656
application where you have lots
of growth potential. And then

1148
01:04:11,656 --> 01:04:15,800
that growth gets to fund to the
more general again. Which is

1149
01:04:15,800 --> 01:04:19,944
quite cool. So I don't know. I
would love to I would love to

1150
01:04:19,944 --> 01:04:24,226
hear. You make a bullish case on
maybe 1 or 2 of the different

1151
01:04:24,226 --> 01:04:28,439
areas that are somewhat like
agriculture chair or, one of the

1152
01:04:28,439 --> 01:04:32,721
other ones that you were saying
maybe manufacturing or I don't

1153
01:04:32,721 --> 01:04:35,000
know, I just hear the case on
it.

1154
01:04:35,000 --> 01:04:38,418
Bradley Dillon: Yeah,
absolutely. Well, I'd say that

1155
01:04:38,418 --> 01:04:42,559
the, the areas that we have, our
biggest customers were we were

1156
01:04:42,559 --> 01:04:46,371
having the biggest revenue
gains. You know, one of them is

1157
01:04:46,371 --> 01:04:49,000
in, security actually. And. Oh,
I'm not.

1158
01:04:49,000 --> 01:04:51,000
Audrow Nash: Surprised by that,
actually. Okay.

1159
01:04:51,000 --> 01:04:53,175
Bradley Dillon: Yeah, it's an
interesting space and talk a

1160
01:04:53,175 --> 01:04:54,000
little bit more about.

1161
01:04:54,000 --> 01:04:58,299
Audrow Nash: Because also like,
home security systems, I did a

1162
01:04:58,299 --> 01:05:02,529
deep dive, maybe a year or two
ago, and they really suck. And

1163
01:05:02,529 --> 01:05:06,343
there's a lot of bad data
privacy concerns with most of

1164
01:05:06,343 --> 01:05:10,712
the leading ones, it seems. And
so I, have half been wanting to

1165
01:05:10,712 --> 01:05:15,011
just build my own for this kind
of thing. Like, how cool would

1166
01:05:15,011 --> 01:05:19,033
it be? You just have a camera
pointing to something and it

1167
01:05:19,033 --> 01:05:23,263
says, did I leave the garage
door open? Did I leave the stove

1168
01:05:23,263 --> 01:05:27,423
on? And you have it answer this
kind of thing? But okay, I'd

1169
01:05:27,423 --> 01:05:31,168
love to hear more about
security. Yeah. This one seems

1170
01:05:31,168 --> 01:05:32,000
great to me.

1171
01:05:32,000 --> 01:05:35,408
Bradley Dillon: Yeah, it's
exciting for sure. So I'll say

1172
01:05:35,408 --> 01:05:38,576
that I'll. I'll talk about
security in a second. The,

1173
01:05:38,576 --> 01:05:42,343
another area where we're getting
a lot of traction is it's just

1174
01:05:42,343 --> 01:05:46,169
kind of related to, automobiles.
And so. Yeah, you know, related

1175
01:05:46,169 --> 01:05:49,995
to automobiles. We've done some
stuff with, like, AI, dash cams.

1176
01:05:49,995 --> 01:05:53,523
We've done some stuff on, you
know, calibrating some of the

1177
01:05:53,523 --> 01:05:57,229
vehicle systems, themselves. And
in some of them on, you know,

1178
01:05:57,229 --> 01:06:00,458
monitoring, observing the
vehicles, like license plate

1179
01:06:00,458 --> 01:06:04,105
recognition. So that's another
huge area for sure. And then I

1180
01:06:04,105 --> 01:06:07,572
would say that, you know, I do
think that agricultural and

1181
01:06:07,572 --> 01:06:11,279
industrial equipment space has
also been been really promising

1182
01:06:11,279 --> 01:06:14,986
for us, the area that we have
not had as much traction, but we

1183
01:06:14,986 --> 01:06:18,214
actually are pretty bullish on
is like the quick serve

1184
01:06:18,214 --> 01:06:21,980
restaurant retail case. And to
go back to the beginning, so the

1185
01:06:21,980 --> 01:06:25,448
security, we've traditionally
thought that security was an

1186
01:06:25,448 --> 01:06:28,915
industry to potentially avoid.
And in the sense that we're

1187
01:06:28,915 --> 01:06:32,323
like, okay, what can we do
that's special here? That, you

1188
01:06:32,323 --> 01:06:36,149
know, say home security systems
or legacy systems can't provide.

1189
01:06:36,149 --> 01:06:39,199
And yeah, we've we found some,
some really exciting

1190
01:06:39,199 --> 01:06:42,188
opportunities where we're
helping and the kind of,

1191
01:06:42,188 --> 01:06:45,894
security screening that would
happen at things like elementary

1192
01:06:45,894 --> 01:06:49,422
schools, warehouses, stadiums,
things like that. So this is

1193
01:06:49,422 --> 01:06:52,471
really high volume, high
throughput security where,

1194
01:06:52,471 --> 01:06:55,998
basically magnetic technology is
then fuzed with, you know,

1195
01:06:55,998 --> 01:06:59,705
vision technology to make it so
that if a stream of people are

1196
01:06:59,705 --> 01:07:03,172
walking in somewhere, you can,
you know, the weapon can be

1197
01:07:03,172 --> 01:07:06,580
detected and tagged to an
individual, and that person can

1198
01:07:06,580 --> 01:07:09,868
be, can, can be grabbed. You're
the systems are kind of

1199
01:07:09,868 --> 01:07:13,276
necessary when, if you want to
get like 100,000 people in

1200
01:07:13,276 --> 01:07:17,102
somewhere really quickly or you
want to be able to get, say, you

1201
01:07:17,102 --> 01:07:20,809
know, a couple thousand people
in school on time every day and

1202
01:07:20,809 --> 01:07:24,217
you want to screen them. You
can't use like a traditional

1203
01:07:24,217 --> 01:07:27,923
metal detector. It's just too
slow. So that's where like these

1204
01:07:27,923 --> 01:07:31,391
kind of higher throughput in
some cases is more performant

1205
01:07:31,391 --> 01:07:34,739
because you can be kind of
sensitive to certain types of

1206
01:07:34,739 --> 01:07:38,565
weapons. That's that's where we
where that's where we've kind of

1207
01:07:38,565 --> 01:07:40,000
seen with that security.

1208
01:07:40,000 --> 01:07:44,529
Audrow Nash: Interesting. Okay.
Very cool case. For this kind of

1209
01:07:44,529 --> 01:07:48,491
thing. But not enough. Not
compelling enough to make you

1210
01:07:48,491 --> 01:07:53,020
want to jump ship not jump ship,
but like pivot a little bit. In

1211
01:07:53,020 --> 01:07:57,407
this direction I guess, because
then it's finding you like you

1212
01:07:57,407 --> 01:08:01,441
guys have been on the specific
problem of making your own

1213
01:08:01,441 --> 01:08:05,828
hardware work really well and
then, adding compute into it. So

1214
01:08:05,828 --> 01:08:10,144
those have all been difficult
lifts, I imagine, for the team.

1215
01:08:10,144 --> 01:08:13,895
And so this would be like
acquiring domain expertise,

1216
01:08:13,895 --> 01:08:17,716
finding a good problem, finding
a good market and then

1217
01:08:17,716 --> 01:08:22,174
everything else on top of that.
So makes sense, but still cool.

1218
01:08:22,174 --> 01:08:26,207
I wonder if eventually you guys
get bored ever with where

1219
01:08:26,207 --> 01:08:30,382
you're. Where you're at, Go
pivot and make start ups around

1220
01:08:30,382 --> 01:08:33,000
it or something for specific
domains.

1221
01:08:33,000 --> 01:08:35,990
Bradley Dillon: Yeah. We, you
know, we've we've seriously

1222
01:08:35,990 --> 01:08:39,189
thought about, you know, having
a, you know, kind of creating

1223
01:08:39,189 --> 01:08:42,126
sister companies that would
tackle these specific cases.

1224
01:08:42,126 --> 01:08:45,115
What's interesting is you get
really industry specific is

1225
01:08:45,115 --> 01:08:48,000
yeah, it can kind of become a
whole company on its own.

1226
01:08:48,000 --> 01:08:49,000
Audrow Nash: It is. Yeah.

1227
01:08:49,000 --> 01:08:52,175
Bradley Dillon: And so an
example area could be say, the

1228
01:08:52,175 --> 01:08:55,690
military. And so in the United
States, of course, there's been

1229
01:08:55,690 --> 01:08:58,638
a lot of investment kind of
actually venture capital

1230
01:08:58,638 --> 01:09:01,643
investment, which hasn't
historically happened that's

1231
01:09:01,643 --> 01:09:05,044
gone into, you know, new, new
generations of technology past

1232
01:09:05,044 --> 01:09:08,446
the military. And yeah, it's
it's of course, pretty special.

1233
01:09:08,446 --> 01:09:12,074
Is that technology? It requires
a lot of like, domain expertise.

1234
01:09:12,074 --> 01:09:15,192
And so, you know, the other
things are these things are

1235
01:09:15,192 --> 01:09:18,594
interesting, for sure. So, I
mean, who knows what what we'll

1236
01:09:18,594 --> 01:09:21,825
we'll see how it plays out. It's
going to be fun to watch

1237
01:09:21,825 --> 01:09:25,000
whatever whatever ends up
happening that we do for sure.

1238
01:09:25,000 --> 01:09:28,758
Audrow Nash: Yeah. You guys are
in such an exciting spot. With

1239
01:09:28,758 --> 01:09:32,636
all of this. What do you think.
So I mean we've talked a little.

1240
01:09:32,636 --> 01:09:36,333
Bit about what the future. Is.
What do you think. Your future

1241
01:09:36,333 --> 01:09:40,212
is like? Look, Sony's future is,
when you think of the next like

1242
01:09:40,212 --> 01:09:43,000
3 to 5 years, where do you hope
you guys grow?

1243
01:09:43,000 --> 01:09:46,875
Bradley Dillon: So we are, you
know, six years deep into this,

1244
01:09:46,875 --> 01:09:50,688
like I said, but fortunately, we
have we have a team of folks

1245
01:09:50,688 --> 01:09:54,125
that are, you know, young,
hungry, excited, excited to,

1246
01:09:54,125 --> 01:09:58,125
excited for the future. So, you
know, one of our core values is,

1247
01:09:58,125 --> 01:10:01,813
is is kind of the tagline,
creating the future. And I think

1248
01:10:01,813 --> 01:10:05,750
there's this there's this great
quote I read in some, some book

1249
01:10:05,750 --> 01:10:08,938
about the history of
semiconductors that, you know,

1250
01:10:08,938 --> 01:10:13,000
said that the best way to create
that is to predict the future is

1251
01:10:13,000 --> 01:10:17,000
to create it. And so definitely,
you know, we definitely kind of

1252
01:10:17,000 --> 01:10:20,750
believe in that. We're we're
very we're very heads down and,

1253
01:10:20,750 --> 01:10:24,375
and just this intense building
phase. And so, you know, we

1254
01:10:24,375 --> 01:10:28,063
think as we as we scale up,
we're, we're excited to be more

1255
01:10:28,063 --> 01:10:31,750
than anything, like I said,
doing having an increasing kind

1256
01:10:31,750 --> 01:10:35,563
of, you know, plethora of use
cases that have gone to really,

1257
01:10:35,563 --> 01:10:39,500
really big scale. And so, so I
think as you what's so cool, for

1258
01:10:39,500 --> 01:10:43,188
example, is like we recently
were at CES, the trade show in

1259
01:10:43,188 --> 01:10:47,250
Las Vegas, and there were dozens
of companies in booths that were

1260
01:10:47,250 --> 01:10:51,000
using our cameras for various
things. And, and a lot of them

1261
01:10:51,000 --> 01:10:54,750
there are folks who we never
even heard of before. And yeah,

1262
01:10:54,750 --> 01:10:58,688
you know, that type of thing is
really rewarding. So I think as

1263
01:10:58,688 --> 01:11:02,563
a company, we, you know, we do
aspire to become the, you know,

1264
01:11:02,563 --> 01:11:05,938
the de facto, provider of
artificial intelligence, you

1265
01:11:05,938 --> 01:11:09,688
know, 3D camera devices. And
we've really position ourselves

1266
01:11:09,688 --> 01:11:13,750
to be the, you know, the leading
independent American provider of

1267
01:11:13,750 --> 01:11:17,500
that, which we're really, you
know, really proud of. And so,

1268
01:11:17,500 --> 01:11:21,063
you know, we just we see the
future is us just wanting to

1269
01:11:21,063 --> 01:11:24,938
build on that success, become,
like I said, the, you know, the

1270
01:11:24,938 --> 01:11:28,875
the go to spot where folks can,
you know, from start to finish,

1271
01:11:28,875 --> 01:11:32,688
be able to, you know, tackle
these, you know, computer vision

1272
01:11:32,688 --> 01:11:36,500
problems and yeah, I mean, I
don't you know, it's not I don't

1273
01:11:36,500 --> 01:11:39,813
think that none of us is in a
hurry to get, you know,

1274
01:11:39,813 --> 01:11:43,625
whatever, whatever the end of
this journey is, we're having a

1275
01:11:43,625 --> 01:11:47,563
lot of fun, and we're just. We.
So I'm really excited every day

1276
01:11:47,563 --> 01:11:48,000
for it.

1277
01:11:48,000 --> 01:11:51,669
Audrow Nash: Yeah. I mean, the
fact that you guys have grown

1278
01:11:51,669 --> 01:11:55,032
very organically. Like the
Kickstarter campaigns in the

1279
01:11:55,032 --> 01:11:58,884
beginning and then friends and
family funding and bootstrapping

1280
01:11:58,884 --> 01:12:02,553
to where you are with a few
rounds of like angel investment.

1281
01:12:02,553 --> 01:12:06,038
It's like it sounds like a fun
journey. Like I don't know

1282
01:12:06,038 --> 01:12:09,524
better than, like we took a
bunch of advanced investment.

1283
01:12:09,524 --> 01:12:13,070
We're under huge pressure to
scale as quickly as possible.

1284
01:12:13,070 --> 01:12:16,617
And, the runway is rapidly
burning up. This kind of thing,

1285
01:12:16,617 --> 01:12:20,530
like, it just, I don't know. And
then you're seeing all of these

1286
01:12:20,530 --> 01:12:24,199
really cool use cases. You're
getting a lot of feedback from

1287
01:12:24,199 --> 01:12:27,623
the community where they're
enjoying your sensors. I can

1288
01:12:27,623 --> 01:12:31,108
imagine it's been a very fun
ride so far, and I expect it

1289
01:12:31,108 --> 01:12:34,594
will probably keep being a
really great one, too. I think

1290
01:12:34,594 --> 01:12:36,000
you're well positioned.

1291
01:12:36,000 --> 01:12:39,486
Bradley Dillon: Yeah. Thank you.
Yeah. I mean, we're, you know,

1292
01:12:39,486 --> 01:12:42,971
in some cases where we haven't
done something that's, you know,

1293
01:12:42,971 --> 01:12:46,512
necessarily traditional and how
and how, you know, we're able to

1294
01:12:46,512 --> 01:12:49,887
scale the company. But yeah,
we're very fortunate that that's

1295
01:12:49,887 --> 01:12:53,262
worked well the way that the way
that we have and and yeah, I

1296
01:12:53,262 --> 01:12:56,748
mean, it's yeah, it's a, it's a,
it's a cool. Yeah. It's a cool

1297
01:12:56,748 --> 01:13:00,178
spot to be. I mean, I will say
that the startup journey is not

1298
01:13:00,178 --> 01:13:03,609
for the faint of heart. It has.
I mean, it is literally like a

1299
01:13:03,609 --> 01:13:06,652
roller coaster there. It's
frequently a Monday and it's

1300
01:13:06,652 --> 01:13:10,193
like, gosh, this is never going
to work. What are we doing? This

1301
01:13:10,193 --> 01:13:13,678
is, this is this is just this is
not working out. And then, you

1302
01:13:13,678 --> 01:13:16,777
know, Tuesday comes around,
we're like, oh my gosh, this

1303
01:13:16,777 --> 01:13:20,318
thing's going to the moon. This
is so exciting. Oh my gosh. It's

1304
01:13:20,318 --> 01:13:23,693
like gravel on gravel on tight.
This is a rocketship. So it's

1305
01:13:23,693 --> 01:13:27,178
yeah yeah it's the ups and downs
are intense. But you know I, I

1306
01:13:27,178 --> 01:13:30,000
personally wouldn't have it any
other way for sure.

1307
01:13:30,000 --> 01:13:33,529
Audrow Nash: Oh yeah. Do you
have any specific stories around

1308
01:13:33,529 --> 01:13:37,000
the roller coaster. Like I'd
love to hear anything for that.

1309
01:13:37,000 --> 01:13:40,777
Bradley Dillon: Oh gosh. Whoosh.
Yeah. There's. Yeah, there's so

1310
01:13:40,777 --> 01:13:44,378
many for sure. I mean I'd say
like this week is Oh yeah. It's

1311
01:13:44,378 --> 01:13:48,214
not like the, the best examples,
but I'll share it like it's just

1312
01:13:48,214 --> 01:13:51,873
this week because, you know, it
gives you the idea that. Yeah,

1313
01:13:51,873 --> 01:13:55,297
for me because hopefully I've
already forgot what the last

1314
01:13:55,297 --> 01:13:58,838
ones are, because if I kept
remembering vividly to about it.

1315
01:13:58,838 --> 01:14:02,320
So we, we actually, we actually
won a very large order this

1316
01:14:02,320 --> 01:14:05,802
week. And so it was, let's see,
it was a few million dollar

1317
01:14:05,802 --> 01:14:08,989
order, that one. And I'm
delivering all the units this

1318
01:14:08,989 --> 01:14:12,472
year. And it should we should
scale up a lot from that from

1319
01:14:12,472 --> 01:14:15,777
from here. And it's with, you
know, a global Fortune 100

1320
01:14:15,777 --> 01:14:19,259
company. Can't say the name,
but, you know, something we're

1321
01:14:19,259 --> 01:14:22,859
really proud of, you're working
on for a long time. So that's

1322
01:14:22,859 --> 01:14:26,460
like. Oh, yeah. Congrats for
exciting. That's the big win for

1323
01:14:26,460 --> 01:14:30,178
the week. The big loss is we had
a, we had an I won't say which

1324
01:14:30,178 --> 01:14:33,896
government entity, but we had an
unexpected tax bill. And, it's

1325
01:14:33,896 --> 01:14:37,320
funny, you know, we operate
around the world. We ship into

1326
01:14:37,320 --> 01:14:40,920
112 countries. You know, we sell
product in all 50 states. We

1327
01:14:40,920 --> 01:14:44,166
have employees located in
several states throughout the

1328
01:14:44,166 --> 01:14:47,648
United States. We have employees
and contractors in a dozen

1329
01:14:47,648 --> 01:14:51,308
countries around the world. And
as a startup, there are, there

1330
01:14:51,308 --> 01:14:54,672
are different government groups
that, you know, we didn't

1331
01:14:54,672 --> 01:14:58,036
realize that we, you know,
needed to be needed to do this

1332
01:14:58,036 --> 01:15:01,341
or that way. So we had an
unexpected one this week where

1333
01:15:01,341 --> 01:15:05,000
we have this tax bill and it's
five we'll pay. The tax bill is

1334
01:15:05,000 --> 01:15:08,424
not not the end of the world.
But it was it was, you know,

1335
01:15:08,424 --> 01:15:12,201
unexpected for us. And you know,
of course, when you're, startup

1336
01:15:12,201 --> 01:15:15,860
where, you know, you're not yet
profitable and you're not left

1337
01:15:15,860 --> 01:15:19,402
yet, you know, positive cash
flow, you're wanting to be able

1338
01:15:19,402 --> 01:15:23,120
to reinvest every single dollar
that you possibly can into, you

1339
01:15:23,120 --> 01:15:26,602
know, hiring more people. So
it's always disappointing when

1340
01:15:26,602 --> 01:15:30,143
you go, oh, shoot, we have to
pay this tax over here that we

1341
01:15:30,143 --> 01:15:33,626
didn't realize we had to. So we
can't hire somebody new. So

1342
01:15:33,626 --> 01:15:37,285
that's, that's that was kind of
the low point. But luckily the

1343
01:15:37,285 --> 01:15:40,000
high point overwhelmed the low
point for sure.

1344
01:15:40,000 --> 01:15:43,000
Audrow Nash: Oh yeah. Yeah.
Overall the week was a win.

1345
01:15:43,000 --> 01:15:45,960
Bradley Dillon: Yeah. Exactly.
Exactly. And yeah like I'll I'll

1346
01:15:45,960 --> 01:15:48,826
send it I'll set a tier, you
know, pour one out for the extra

1347
01:15:48,826 --> 01:15:51,268
employee. We'd like to have
hired and just hope that

1348
01:15:51,268 --> 01:15:53,946
whatever, whatever government
entity we have to give that

1349
01:15:53,946 --> 01:15:56,718
money to, that they are going to
spend it better than us. I

1350
01:15:56,718 --> 01:15:57,000
guess.

1351
01:15:57,000 --> 01:16:03,279
Audrow Nash: What a thing. Yeah.
Taxes. So I'd like to just go a

1352
01:16:03,279 --> 01:16:09,362
little bit more. Like, I don't
know, kind of random. Questions

1353
01:16:09,362 --> 01:16:15,642
for different things. Now. What
are your thoughts on vision like

1354
01:16:15,642 --> 01:16:21,725
cameras. Versus lidar or kind of
which application? Yeah, both

1355
01:16:21,725 --> 01:16:23,000
are good for.

1356
01:16:23,000 --> 01:16:26,216
Bradley Dillon: Yeah. So
ultimately we're actually

1357
01:16:26,216 --> 01:16:30,205
agnostic to either technology.
We are not trying to make a bet

1358
01:16:30,205 --> 01:16:34,000
that says like it's all vision
all the way or all wired or.

1359
01:16:34,000 --> 01:16:35,000
Audrow Nash: All, but yeah,
totally.

1360
01:16:35,000 --> 01:16:38,608
Bradley Dillon: And so we're
we're more you know, as I shared

1361
01:16:38,608 --> 01:16:42,156
with you, we do have our first
you know, lidar based camera,

1362
01:16:42,156 --> 01:16:45,290
which is a time of flight
solution. And, you know, of

1363
01:16:45,290 --> 01:16:48,957
course, also our devices have a
lot of cameras in top of other

1364
01:16:48,957 --> 01:16:52,328
sensors. We understand the
arguments for, for both cases,

1365
01:16:52,328 --> 01:16:55,699
you know, we talk internally
about the example being like

1366
01:16:55,699 --> 01:16:59,011
Tesla who's gone all in with
vision and their thesis is,

1367
01:16:59,011 --> 01:17:02,796
well, if a human can drive with
vision then, you know, why can't

1368
01:17:02,796 --> 01:17:06,285
a computer drive with vision.
Whereas others like say Waymo

1369
01:17:06,285 --> 01:17:09,952
that sky you know, so hey, we
want every sensor we can get and

1370
01:17:09,952 --> 01:17:13,559
that's going to include lidar.
What we see with our customers

1371
01:17:13,559 --> 01:17:17,167
is, is that there's some cases
where a lighter based solution

1372
01:17:17,167 --> 01:17:20,715
is better. And, you know, you
go, great, okay. We can we can

1373
01:17:20,715 --> 01:17:24,027
support that. And then we see
some kind of vision can be

1374
01:17:24,027 --> 01:17:27,457
better. I do think it'll be
really interesting is, is that

1375
01:17:27,457 --> 01:17:31,005
the economies of scale with
cameras have made it so that you

1376
01:17:31,005 --> 01:17:34,376
have really high performant,
really low cost cameras. And

1377
01:17:34,376 --> 01:17:38,043
like you, you mentioned there's
like say, you know, a thousand

1378
01:17:38,043 --> 01:17:41,591
megapixel camera out there. You
know, that's not the type of

1379
01:17:41,591 --> 01:17:44,548
thing that's needed for most
commercial use cases.

1380
01:17:44,548 --> 01:17:47,919
Conversely, with lidar, you
know, the prices have stayed,

1381
01:17:47,919 --> 01:17:51,527
you know, stayed pretty high.
But you do see that coming down

1382
01:17:51,527 --> 01:17:55,253
really dramatically. So I think
over the next five years that's

1383
01:17:55,253 --> 01:17:58,801
going to be the biggest, change.
And I think that's going to

1384
01:17:58,801 --> 01:18:02,468
change the dynamics. You know,
this lidar versus vision thing.

1385
01:18:02,468 --> 01:18:06,134
And I would be surprised if you
know, once the cost comes down

1386
01:18:06,134 --> 01:18:09,624
low enough that, you know, say,
company would end up adding

1387
01:18:09,624 --> 01:18:13,290
lidar anyway when you know, they
can get it to get it at a low

1388
01:18:13,290 --> 01:18:14,000
enough cost.

1389
01:18:14,000 --> 01:18:18,221
Audrow Nash: Interesting. Yeah.
That's I wonder I wonder if so

1390
01:18:18,221 --> 01:18:21,898
it say like Waymo. Deploys
massively then that'll help

1391
01:18:21,898 --> 01:18:26,119
light our get huge economies of
scale. Or some other companies

1392
01:18:26,119 --> 01:18:30,000
say and then that might drive
down the cost tremendously.

1393
01:18:30,000 --> 01:18:31,000
Bradley Dillon: Yeah. We're
seeing.

1394
01:18:31,000 --> 01:18:32,000
Audrow Nash: That would be
really.

1395
01:18:31,000 --> 01:18:34,316
Bradley Dillon: Cool. Yeah.
We're seeing some new market

1396
01:18:34,316 --> 01:18:37,750
entrants that are, you know,
have it scaled up yet but are

1397
01:18:37,750 --> 01:18:41,362
saying that they have, you know,
new, new technology, new new

1398
01:18:41,362 --> 01:18:45,092
techniques that make it so that
they can do it a lot cheaper on

1399
01:18:45,092 --> 01:18:48,882
a night out there. So I think, I
think that space is going to be

1400
01:18:48,882 --> 01:18:51,783
very deflationary, and
deflationary, even for the

1401
01:18:51,783 --> 01:18:55,158
normal scales that electronics
have. Which electronics of

1402
01:18:55,158 --> 01:18:58,000
course, is, is has been very
deflationary space.

1403
01:18:58,000 --> 01:19:02,526
Audrow Nash: Definitely. Let's
see. So another one, another

1404
01:19:02,526 --> 01:19:07,281
random segue. How you mentioned
at the beginning 2022 and like

1405
01:19:07,281 --> 01:19:11,807
all the chip craziness, Tell me
in the time of Luxonis into

1406
01:19:11,807 --> 01:19:16,639
current day, and then also, I'd
love to hear your prediction of

1407
01:19:16,639 --> 01:19:20,705
how it will be. How has
procurement of the chips. And

1408
01:19:20,705 --> 01:19:25,000
different electrical components
you need been over time?

1409
01:19:25,000 --> 01:19:28,724
Bradley Dillon: It I'd like to
say it's come easy. It has it.

1410
01:19:28,724 --> 01:19:32,326
There's there's still times
where there's still times where

1411
01:19:32,326 --> 01:19:35,745
in a, you know, a bill of
materials that we have for one

1412
01:19:35,745 --> 01:19:39,469
of our devices, which is, excuse
me, depending on the device,

1413
01:19:39,469 --> 01:19:43,254
maybe 50 to 150 items, there's
still sometimes items that trip

1414
01:19:43,254 --> 01:19:47,100
us up. So every time when we're
looking at our production plans

1415
01:19:47,100 --> 01:19:50,825
over the next upcoming period,
we're still having to identify

1416
01:19:50,825 --> 01:19:54,304
high risk items. And we're
needing to, you know, purchase

1417
01:19:54,304 --> 01:19:57,968
ahead on them. And then we do
have some components that have

1418
01:19:57,968 --> 01:20:01,753
painfully long lead times. So
when one of our components for a

1419
01:20:01,753 --> 01:20:05,599
series for of hardware, it has a
seven month lead time. Oh, and

1420
01:20:05,599 --> 01:20:09,201
when you think of the seven
month lead time, you know, once

1421
01:20:09,201 --> 01:20:12,803
we have all the components it
takes you know, call it about

1422
01:20:12,803 --> 01:20:16,161
eight weeks to, you know,
manufacture and assemble them

1423
01:20:16,161 --> 01:20:19,946
and get them ready to ship to
customers. So that means is that

1424
01:20:19,946 --> 01:20:23,487
we're sitting here today and we
have to predict what's our

1425
01:20:23,487 --> 01:20:27,394
demand going to be like for this
device in nine months. And, you

1426
01:20:27,394 --> 01:20:30,996
know, that is that is not easy
to do. Yeah. You have to you

1427
01:20:30,996 --> 01:20:34,781
know, you have to just follow
your instincts and hope and pray

1428
01:20:34,781 --> 01:20:38,689
that it's going to work out. So
yeah, it continues. It continues

1429
01:20:38,689 --> 01:20:42,474
to be pretty challenging. And
then as we look ahead, I think I

1430
01:20:42,474 --> 01:20:46,198
think what's interesting, you
know, coming, for example, most

1431
01:20:46,198 --> 01:20:49,983
recently at CES where I start
Consumer Electronics Show in Las

1432
01:20:49,983 --> 01:20:53,707
Vegas, where you have all of the
different chip manufacturers

1433
01:20:53,707 --> 01:20:57,126
that are showing off this latest
and latest and greatest

1434
01:20:57,126 --> 01:21:00,972
hardware. You have a lot of them
that are kind of more floating

1435
01:21:00,972 --> 01:21:04,818
ideas, like, hey, here's a chip
that could do this, or here's a

1436
01:21:04,818 --> 01:21:08,665
chip that could do that, and you
don't know yet. Like, what are

1437
01:21:08,665 --> 01:21:12,145
the ones that they're actually
going to be supporting. At

1438
01:21:12,145 --> 01:21:15,808
scale. So I think as you look
ahead, it's going to be really

1439
01:21:15,808 --> 01:21:19,593
interesting to see how the pace
of, AI hardware progresses and

1440
01:21:19,593 --> 01:21:22,951
to, to see, you know, what are
some of these new market

1441
01:21:22,951 --> 01:21:26,614
entrants that are going to kind
of disrupt things outside of

1442
01:21:26,614 --> 01:21:30,338
like kind of the way you see big
chip, providers and what are

1443
01:21:30,338 --> 01:21:33,451
the, is it is it going to
continue to be this, like

1444
01:21:33,451 --> 01:21:36,870
dizzying array of different
potential options or they're

1445
01:21:36,870 --> 01:21:40,716
going to is it going to get us
back a bit more narrow? In terms

1446
01:21:40,716 --> 01:21:44,502
of what's available. So that'll
be, I think, as we look ahead,

1447
01:21:44,502 --> 01:21:48,226
will be interesting to see, as
you have just a lot of players

1448
01:21:48,226 --> 01:21:51,767
working, a lot of different
stuff and just so much I mean,

1449
01:21:51,767 --> 01:21:55,613
things are moving so fast there.
You back to you honestly can't

1450
01:21:55,613 --> 01:21:59,276
keep can't keep off every, you
know, every few weeks there's

1451
01:21:59,276 --> 01:22:03,000
some crazy new announcement on
on some new piece of hardware.

1452
01:22:03,000 --> 01:22:08,224
Audrow Nash: Yeah for sure. Very
interesting. You mentioned high

1453
01:22:08,224 --> 01:22:12,878
risk items. Is there any like
theme or pattern around the

1454
01:22:12,878 --> 01:22:18,102
items that are. Typically high.
Risk, in your opinion or in your

1455
01:22:18,102 --> 01:22:19,000
experience?

1456
01:22:19,000 --> 01:22:22,293
Bradley Dillon: It sometimes
seems random. So a component

1457
01:22:22,293 --> 01:22:25,644
that trip to said recently was
an LED and a light emitting

1458
01:22:25,644 --> 01:22:29,053
diode. And it was a to us. Would
we pick the design? It was

1459
01:22:29,053 --> 01:22:32,577
available component. It seemed
pretty standard. It turned out

1460
01:22:32,577 --> 01:22:36,274
that there's basically only one
manufacturer that was making it.

1461
01:22:36,274 --> 01:22:39,741
They had an issue. And, you
know, boom, it caused, you know,

1462
01:22:39,741 --> 01:22:43,438
I think it was like a four week
delay or something like that. So

1463
01:22:43,438 --> 01:22:46,962
sometimes it's really random.
It's, you know, it's our job to

1464
01:22:46,962 --> 01:22:50,371
try to not let those type of
situations come out of nowhere

1465
01:22:50,371 --> 01:22:53,722
on us. I will say that the
things that were tripping folks

1466
01:22:53,722 --> 01:22:57,419
up, you know, back in 21 and 22,
things like memory and storage,

1467
01:22:57,419 --> 01:23:00,886
that area has gotten stable.
Where there's there's there's a

1468
01:23:00,886 --> 01:23:04,467
lot that's available and, and
yeah, I think for us on the high

1469
01:23:04,467 --> 01:23:07,645
risk components, it's really
more a function of who the

1470
01:23:07,645 --> 01:23:10,591
actual like upstream
manufacturer is. What are the,

1471
01:23:10,591 --> 01:23:14,058
you know, like what are the
available substitutes for it. So

1472
01:23:14,058 --> 01:23:17,351
what tripped us up with the LED
was, is that there wasn't

1473
01:23:17,351 --> 01:23:20,991
actually any substitutes. And so
other areas of our components,

1474
01:23:20,991 --> 01:23:24,457
and a lot of times what we do is
we actually create multiple

1475
01:23:24,457 --> 01:23:28,039
designs of our product that can
handle different variants of a

1476
01:23:28,039 --> 01:23:31,447
very similar, components. So
that way over time if one goes

1477
01:23:31,447 --> 01:23:35,029
out of stock, then boom, we can
just we can just go in and you

1478
01:23:35,029 --> 01:23:38,669
could swap in the other one. So
it's really about ones where we

1479
01:23:38,669 --> 01:23:42,309
go. There's no other option for
this component. We have to have

1480
01:23:42,309 --> 01:23:45,371
it. And then you know how how
many options do we have

1481
01:23:45,371 --> 01:23:48,837
available for it. And if it's
very few or just one then it's

1482
01:23:48,837 --> 01:23:52,246
flag. Say shoot. And we you know
basically been our finance

1483
01:23:52,246 --> 01:23:55,770
department says, okay, you need
how much cash to de-risk this

1484
01:23:55,770 --> 01:23:59,410
and you have to make a business
decision of, yeah, you're going

1485
01:23:59,410 --> 01:24:02,761
to go ahead and lay out some
cash so that these components

1486
01:24:02,761 --> 01:24:06,516
can sit on a warehouse somewhere
so that you, you know, make sure

1487
01:24:06,516 --> 01:24:09,000
you have reliable supply for
your customer.

1488
01:24:09,000 --> 01:24:12,896
Audrow Nash: Yeah. That's wild.
What a thing where you have to

1489
01:24:12,896 --> 01:24:16,478
have it instead of like just in
time, you end up having a

1490
01:24:16,478 --> 01:24:20,374
warehouse full of components.
You need to de-risk. And that's,

1491
01:24:20,374 --> 01:24:24,208
it's an interesting time that
and I think a lot of businesses

1492
01:24:24,208 --> 01:24:28,041
are doing that more generally to
make sure that they can keep

1493
01:24:28,041 --> 01:24:31,372
delivering even if, components
are not being reliably

1494
01:24:31,372 --> 01:24:32,000
available.

1495
01:24:32,000 --> 01:24:35,270
Bradley Dillon: Yeah. That's
really interesting. Yeah. And

1496
01:24:35,270 --> 01:24:38,596
I'd say from a procurement
standpoint and just sort of, you

1497
01:24:38,596 --> 01:24:41,696
know, hardware in general, you
have to pick what you're

1498
01:24:41,696 --> 01:24:44,910
prioritizing. And so for
Luxonis, there's two things that

1499
01:24:44,910 --> 01:24:48,349
are top priority. It's quality
and it's having supply. And so

1500
01:24:48,349 --> 01:24:51,844
we're at a stage as a company
that not having supply is not an

1501
01:24:51,844 --> 01:24:55,395
option. We have to be stock with
our components. And quality is

1502
01:24:55,395 --> 01:24:59,060
the other piece where, you know,
the stuff that we deliver has to

1503
01:24:59,060 --> 01:25:02,217
be really high quality. The
things that we sacrifice for

1504
01:25:02,217 --> 01:25:05,825
that is sometimes we pay higher
cost, and that's where sometimes

1505
01:25:05,825 --> 01:25:09,038
we procure on the spot market
rather than something being

1506
01:25:09,038 --> 01:25:12,421
planned, for example. And we
tolerate a higher cost, so long

1507
01:25:12,421 --> 01:25:15,972
as we can make sure that we're,
you know, have it having have a

1508
01:25:15,972 --> 01:25:19,298
great quality. And we, we,
actually have a reliable supply.

1509
01:25:19,298 --> 01:25:22,906
And we also sometimes we have to
tolerate things being not quite

1510
01:25:22,906 --> 01:25:26,232
as fast, so long as, you know,
they're not, they're not too

1511
01:25:26,232 --> 01:25:29,164
much delayed. So it's
interesting, you know, at some

1512
01:25:29,164 --> 01:25:32,490
point later in the journey,
we'll actually care about cost,

1513
01:25:32,490 --> 01:25:35,985
but that that time has actually
not come yet. For us to try to

1514
01:25:35,985 --> 01:25:37,000
optimize for that.

1515
01:25:37,000 --> 01:25:40,580
Audrow Nash: Yeah. And I hope
you. Build in a healthy margin

1516
01:25:40,580 --> 01:25:44,160
for everything too. So you have
a nice buffer, like of let's

1517
01:25:44,160 --> 01:25:47,920
say, you say you have to pay a
little more for some components.

1518
01:25:47,920 --> 01:25:51,739
I hope that's priced in for this
kind of thing too. But yeah, it

1519
01:25:51,739 --> 01:25:55,200
makes sense. You guys are. In
very much a scaling like get

1520
01:25:55,200 --> 01:25:58,660
everyone using it, refine the
ecosystem, thing so they can

1521
01:25:58,660 --> 01:26:01,883
make lots of good stuff with
what you're making, in my

1522
01:26:01,883 --> 01:26:05,582
opinion, or in my estimation of
where you guys are. Yeah. It's

1523
01:26:05,582 --> 01:26:06,000
coming.

1524
01:26:06,000 --> 01:26:09,108
Bradley Dillon: Yeah. We do, we
do leave room for it. So, for

1525
01:26:09,108 --> 01:26:11,707
example, I had a call today
where, yeah, one of our

1526
01:26:11,707 --> 01:26:14,561
engineers said, hey, there's
what we think can be a more

1527
01:26:14,561 --> 01:26:17,822
performant component on the the
bill of materials. It's going to

1528
01:26:17,822 --> 01:26:20,777
add like $5 a cost with this
product. Should we do it. And

1529
01:26:20,777 --> 01:26:23,631
the answer is like yes, yes.
It's going to like our it's

1530
01:26:23,631 --> 01:26:26,790
going to help our customer win.
So like the answer is yes. And

1531
01:26:26,790 --> 01:26:30,051
you know, well, what we do first
is we say okay, let's implement

1532
01:26:30,051 --> 01:26:33,108
that and then we'll go later,
you know later, later on we'll

1533
01:26:33,108 --> 01:26:36,420
try to figure out how to make it
so that it doesn't actually cost

1534
01:26:36,420 --> 01:26:38,000
us, you know, five extra bucks.

1535
01:26:38,000 --> 01:26:42,607
Audrow Nash: So you mentioned
earlier that you like adding

1536
01:26:42,607 --> 01:26:47,453
sensors. To your, your systems.
What do you think will be the

1537
01:26:47,453 --> 01:26:52,220
next like, do you have other
sensors that you're considering

1538
01:26:52,220 --> 01:26:55,000
adding. And what. Are some of
them.

1539
01:26:55,000 --> 01:26:58,223
Bradley Dillon: We've been
exploring some options. So

1540
01:26:58,223 --> 01:27:01,992
things that we've discussed or
things like that based cameras.

1541
01:27:01,992 --> 01:27:05,580
There things like the other
things like structured light in

1542
01:27:05,580 --> 01:27:08,741
other cases, they're about
taking one of our current

1543
01:27:08,741 --> 01:27:12,572
sensors and just making it a lot
more performant or capable. So

1544
01:27:12,572 --> 01:27:16,098
it could be having like, you
know, a higher resolution. It

1545
01:27:16,098 --> 01:27:19,807
could be that maybe it has a,
you know, a bigger, you know, a

1546
01:27:19,807 --> 01:27:23,455
bigger range of things that it
can do. And yeah, we don't we

1547
01:27:23,455 --> 01:27:27,164
actually don't have we're just
in the exploratory phase right

1548
01:27:27,164 --> 01:27:30,812
now because we've been focused
on getting the, you know, the

1549
01:27:30,812 --> 01:27:34,582
thermal, option and the, the,
you know, the light, our time of

1550
01:27:34,582 --> 01:27:38,291
light option out there. And so,
yeah, we haven't set our next

1551
01:27:38,291 --> 01:27:42,000
plans yet, but that's some of
the stuff that we're exploring.

1552
01:27:42,000 --> 01:27:45,059
Audrow Nash: Yeah. Actually tell
me a bit about the thermal

1553
01:27:45,059 --> 01:27:48,326
camera because that opens up a
lot of different. Applications I

1554
01:27:48,326 --> 01:27:49,000
would expect.

1555
01:27:49,000 --> 01:27:52,542
Bradley Dillon: Yeah it does. So
some of the cases that our

1556
01:27:52,542 --> 01:27:55,904
clients seem to be really
excited about some of them are

1557
01:27:55,904 --> 01:27:59,626
in oil and gas. And so it can be
placed somewhere and it could

1558
01:27:59,626 --> 01:28:02,928
detect like a leak, say
something that's, you know, hot

1559
01:28:02,928 --> 01:28:06,711
or cold, you know, that a human
wouldn't be able to see, and it

1560
01:28:06,711 --> 01:28:10,553
can also in that type of a case,
in some other use cases, it can

1561
01:28:10,553 --> 01:28:14,335
see that equipment is operating
at the right temperature. So it

1562
01:28:14,335 --> 01:28:17,757
could see if maybe a component
is overheating the thermal

1563
01:28:17,757 --> 01:28:21,239
sensitivity is I think about
plus or minus two Celsius. So

1564
01:28:21,239 --> 01:28:24,661
you can see it with pretty
pretty good accuracy. And then

1565
01:28:24,661 --> 01:28:28,083
some other things are for
example in the agricultural use

1566
01:28:28,083 --> 01:28:31,685
case, we had a conversation with
the customer that's redoing

1567
01:28:31,685 --> 01:28:35,528
indoor farming. And they're able
to use it to be able to monitor

1568
01:28:35,528 --> 01:28:39,070
their crop. And then there's
been also some interesting use

1569
01:28:39,070 --> 01:28:42,612
cases. We talked earlier about
elderly care. And so there's

1570
01:28:42,612 --> 01:28:46,094
been some over elderly care
where the thermal sensor could

1571
01:28:46,094 --> 01:28:49,696
be added to vision. And you
would be able to, you know, tell

1572
01:28:49,696 --> 01:28:53,418
things, at a higher performance
rate of things like if there's

1573
01:28:53,418 --> 01:28:56,000
been a fall or something along
those lines.

1574
01:28:56,000 --> 01:28:59,839
Audrow Nash: Oh, interesting.
Yeah. Because I think the thing

1575
01:28:59,839 --> 01:29:03,678
with thermal cameras. It makes
setting people very easy cause

1576
01:29:03,678 --> 01:29:07,517
it's super hot outside, like
indoors, say, where you have air

1577
01:29:07,517 --> 01:29:11,231
conditioning because you have
that big differential between

1578
01:29:11,231 --> 01:29:14,000
the person and, the rest of the
environment.

1579
01:29:14,000 --> 01:29:16,689
Bradley Dillon: Yeah. So we
actually at, see you in Las

1580
01:29:16,689 --> 01:29:19,575
Vegas. Like I keep mentioning,
we had a demo of the thermal

1581
01:29:19,575 --> 01:29:22,264
camera and it's yeah, it's
interesting because the it's

1582
01:29:22,264 --> 01:29:25,149
people's skin that just really,
you know, you really see it

1583
01:29:25,149 --> 01:29:28,083
light up because it's so, so
warm. And the, you know, things

1584
01:29:28,083 --> 01:29:31,213
like it's interesting, you know,
you're wearing glasses. So when

1585
01:29:31,213 --> 01:29:34,196
people wear glasses it looks
like they kind of, you know that

1586
01:29:34,196 --> 01:29:36,983
the temperature profile is so
different. So it looks like

1587
01:29:36,983 --> 01:29:40,015
they're wearing goggles. Like it
looks like they're like Robin

1588
01:29:40,015 --> 01:29:43,095
from Batman and Robin. Yep. So
yes, it's Yeah. It's interesting

1589
01:29:43,095 --> 01:29:45,932
there. So when, you know, kind
of, if so thermally like if

1590
01:29:45,932 --> 01:29:48,670
somebody is wearing a short
sleeve shirt and shorts, you

1591
01:29:48,670 --> 01:29:51,506
know, it's actually kind of
easier to detect them than if,

1592
01:29:51,506 --> 01:29:54,000
say, they're wearing, you know,
jeans and a hoodie.

1593
01:29:54,000 --> 01:29:55,000
Audrow Nash: And it has
absolute.

1594
01:29:55,000 --> 01:29:56,000
Bradley Dillon: Yeah. Exactly.

1595
01:29:56,000 --> 01:30:00,288
Audrow Nash: Yeah. Yeah. Very
funny. Yeah. Thermal cameras.

1596
01:30:00,288 --> 01:30:04,648
Are awesome. That's so exciting
that you guys are, making it

1597
01:30:04,648 --> 01:30:08,645
more and more accessible to
people. Because, like, I, I

1598
01:30:08,645 --> 01:30:13,151
used, Leptin. Or was. Something
like that. One of those in the

1599
01:30:13,151 --> 01:30:17,439
past, and. It was. Okay. And I
think you guys have probably

1600
01:30:17,439 --> 01:30:21,000
made it even nicer for
everything. So super cool.

1601
01:30:21,000 --> 01:30:24,558
Bradley Dillon: Yeah, we'd like
to hope so. We've learned some

1602
01:30:24,558 --> 01:30:27,715
stuff along the way. So there's
actually some like Itar

1603
01:30:27,715 --> 01:30:30,814
regulations, which is like
international trade of arms

1604
01:30:30,814 --> 01:30:34,086
control of arms regulation or
something like that. And it

1605
01:30:34,086 --> 01:30:37,816
turns out that, you know, things
like a thermal camera, you can't

1606
01:30:37,816 --> 01:30:40,858
export to certain places. So
we've not had to do this

1607
01:30:40,858 --> 01:30:44,474
normally with our products, but
we've actually had to limit the

1608
01:30:44,474 --> 01:30:47,286
export of it to NATO plus
countries. So it's NATO

1609
01:30:47,286 --> 01:30:50,500
countries plus a few others that
are kind of in the good

1610
01:30:50,500 --> 01:30:53,886
relations. So we've had some
angry customers say like, hey,

1611
01:30:53,886 --> 01:30:57,502
why can't you ship that to our
country? And it's like, no, this

1612
01:30:57,502 --> 01:31:01,060
is a hard no. You know, we've
we, we we consulted with respect

1613
01:31:01,060 --> 01:31:04,504
security. We got an official
decision here and we absolutely

1614
01:31:04,504 --> 01:31:07,833
have to comply. So it's been
interesting for us because we

1615
01:31:07,833 --> 01:31:11,219
haven't normally had that type
of restriction. But you know

1616
01:31:11,219 --> 01:31:14,662
we've learned that yeah it's
it's you know those regulations

1617
01:31:14,662 --> 01:31:18,278
are in there. You know, they're
in place for a reason. So we've

1618
01:31:18,278 --> 01:31:20,000
learned a lot in that process.

1619
01:31:20,000 --> 01:31:23,424
Audrow Nash: Yeah. That's wild.
That's not something you

1620
01:31:23,424 --> 01:31:26,908
normally think about but
interesting. Yeah. You then have

1621
01:31:26,908 --> 01:31:30,026
to comply with them. And I.
Guess it's for military

1622
01:31:30,026 --> 01:31:33,694
applications. Maybe you can. See
people at night or. I don't

1623
01:31:33,694 --> 01:31:34,000
know.

1624
01:31:34,000 --> 01:31:35,000
Bradley Dillon: Why do you think
thermal.

1625
01:31:35,000 --> 01:31:40,906
Audrow Nash: Is not allowed or
has these items, required months

1626
01:31:40,906 --> 01:31:44,000
or. I don't know, limitations
or.

1627
01:31:44,000 --> 01:31:47,328
Bradley Dillon: I don't know, I
think I think you, I think you

1628
01:31:47,328 --> 01:31:50,172
described it correctly that
there's some applications

1629
01:31:50,172 --> 01:31:53,339
related to military or maybe
more nefarious activities that

1630
01:31:53,339 --> 01:31:56,506
that type of technology would
work on. And, you know, we're

1631
01:31:56,506 --> 01:31:59,887
we're not we're not working on
anything that's related to that.

1632
01:31:59,887 --> 01:32:03,000
However, you know, our, our our
device can you know, it's.

1633
01:32:03,000 --> 01:32:04,000
Audrow Nash: Could in theory.
Yeah. Yeah.

1634
01:32:04,000 --> 01:32:07,134
Bradley Dillon: It's an open
it's open playground. Right. So

1635
01:32:07,134 --> 01:32:10,269
you could do a lot of things
with it. So, you know, for, for

1636
01:32:10,269 --> 01:32:12,985
whatever reason those
regulations that were were put

1637
01:32:12,985 --> 01:32:16,067
in place and, you know, it's,
yeah, it applies to kind of a

1638
01:32:16,067 --> 01:32:19,254
lot of that type of technology
that we kind of learned about.

1639
01:32:19,254 --> 01:32:22,388
So there's different sensors
that are out there that we say,

1640
01:32:22,388 --> 01:32:25,313
you know, that those kind of
it's our regulations are in

1641
01:32:25,313 --> 01:32:28,187
place for. And so, you know,
it's interesting that, the

1642
01:32:28,187 --> 01:32:31,582
resolution of the thermal sensor
matters for evaluating that. The

1643
01:32:31,582 --> 01:32:34,507
hertz, the number of frames per
se frequency matter, the

1644
01:32:34,507 --> 01:32:37,694
frequency. So yeah, it's it is
interesting to understand that

1645
01:32:37,694 --> 01:32:40,672
because there's, you know,
depending on those knobs over,

1646
01:32:40,672 --> 01:32:43,493
you know, whether there's
restrictions or not. So, you

1647
01:32:43,493 --> 01:32:46,000
know, it's been interesting to
learn about that.

1648
01:32:46,000 --> 01:32:50,024
Audrow Nash: Definitely. Okay.
Now, so for those who have

1649
01:32:50,024 --> 01:32:54,047
listened or watched and are
curious and wondering. How to

1650
01:32:54,047 --> 01:32:58,494
get started with Luxonis cameras
and devices. Where should they

1651
01:32:58,494 --> 01:33:02,729
go? What kind of resources do
you have for them? Where would

1652
01:33:02,729 --> 01:33:04,000
you point them to?

1653
01:33:04,000 --> 01:33:06,935
Bradley Dillon: Yeah,
absolutely. So our products are

1654
01:33:06,935 --> 01:33:10,368
really easy to purchase, really
low priced. So we have cameras

1655
01:33:10,368 --> 01:33:13,691
starting in $99. And if you go
to shop dot locks on HSN.com,

1656
01:33:13,691 --> 01:33:16,682
you can order them. And like I
said, we shipped in 112

1657
01:33:16,682 --> 01:33:19,838
countries. So around the world
and for, you know, getting

1658
01:33:19,838 --> 01:33:23,161
started with you know, we do
have a lot of, software details

1659
01:33:23,161 --> 01:33:26,705
that are available if you go to
our website works on smart comm,

1660
01:33:26,705 --> 01:33:30,194
you also have like a, you know,
a GitHub as well. And one thing

1661
01:33:30,194 --> 01:33:33,240
that's great to kind of
brainstorm different ideas gets

1662
01:33:33,240 --> 01:33:36,673
some get kind of the creative
juices flowing, you know, figure

1663
01:33:36,673 --> 01:33:39,885
out if you think that that
technology can be used for this

1664
01:33:39,885 --> 01:33:43,485
or that. We actually have a chat
bot. It's, you know, ChatGPT for

1665
01:33:43,485 --> 01:33:46,918
power that has been, that, you
know, passes through all of our

1666
01:33:46,918 --> 01:33:50,462
documentation, all of our forum
posts, and it's it's incredible.

1667
01:33:50,462 --> 01:33:53,840
So it's the chat that works on
this.com and you can say, hey,

1668
01:33:53,840 --> 01:33:56,831
this is the application I'm
working on. You know, what

1669
01:33:56,831 --> 01:34:00,319
camera do you recommend based on
this application? Do you think

1670
01:34:00,319 --> 01:34:03,199
it's possible? Would this be
feasible? Would that be

1671
01:34:03,199 --> 01:34:06,522
feasible? And then later on, if
you're working on developing

1672
01:34:06,522 --> 01:34:09,955
something, if you're running a
do a blocker, you want to learn

1673
01:34:09,955 --> 01:34:13,389
more on something you can ask.
And what's so great is it's way

1674
01:34:13,389 --> 01:34:16,767
more patient than any engineer
ever will be. So the 20th time

1675
01:34:16,767 --> 01:34:19,923
you prompt it, it's just as
plentiful and happy to answer

1676
01:34:19,923 --> 01:34:23,080
your questions as the as the
first. And it's actually way

1677
01:34:23,080 --> 01:34:26,458
more creative. It comes up with
stuff that I never would. And

1678
01:34:26,458 --> 01:34:29,836
what's also nice is you don't
have to be embarrassed that you

1679
01:34:29,836 --> 01:34:33,380
ask it questions that you maybe
would feel weird with the human.

1680
01:34:33,380 --> 01:34:36,814
So for me, I asked some of my
dumbest questions I can think of

1681
01:34:36,814 --> 01:34:40,247
to it because. Because I know
that if I asked it to one of our

1682
01:34:40,247 --> 01:34:43,847
engineers, they would, you know,
they they would maybe be annoyed

1683
01:34:43,847 --> 01:34:47,114
by it. So that's where I want
them to, to, to kind of maybe

1684
01:34:47,114 --> 01:34:48,000
get started too.

1685
01:34:48,000 --> 01:34:52,295
Audrow Nash: That's awesome. And
so, what would you wrapping up

1686
01:34:52,295 --> 01:34:56,659
for all that we've talked about,
what do you hope our listeners.

1687
01:34:56,659 --> 01:35:00,000
And watchers take away from this
whole interview?

1688
01:35:00,000 --> 01:35:03,362
Bradley Dillon: Well, I think
from a broad perspective, I

1689
01:35:03,362 --> 01:35:06,725
think that it's a really
exciting time to get to see this

1690
01:35:06,725 --> 01:35:10,087
technology used across these
different industries and use

1691
01:35:10,087 --> 01:35:13,803
across these these exciting use
cases, getting to see something

1692
01:35:13,803 --> 01:35:17,579
that used to be science fiction
and become reality is, you know,

1693
01:35:17,579 --> 01:35:21,354
really inspiring and makes it so
that it's a, it's an incredible

1694
01:35:21,354 --> 01:35:25,070
it's incredible to be a part of
this would be part of this wave

1695
01:35:25,070 --> 01:35:28,787
and be part of this journey. You
know, we at Luxonis we we love

1696
01:35:28,787 --> 01:35:32,503
we love the community. Yeah. We
love our users. And so if there

1697
01:35:32,503 --> 01:35:35,924
if there's ever anything that
you want to chat about, feel

1698
01:35:35,924 --> 01:35:39,522
free to reach out to me. Reach
out to our community. And, you

1699
01:35:39,522 --> 01:35:42,000
know, I hope everybody enjoyed
the lesson.

1700
01:35:42,000 --> 01:35:45,000
Audrow Nash: Oh yeah. Alright.
Thank you Bradley.

1701
01:35:45,000 --> 01:35:46,000
Bradley Dillon: Thank you.

1702
01:35:46,000 --> 01:35:48,000
Audrow Nash: Bye everyone.
