1
00:00:06,320 --> 00:00:11,499
[Music]

2
00:00:16,080 --> 00:00:19,600
hello everyone and welcome back from

3
00:00:17,920 --> 00:00:21,600
lunch hopefully the catering was as

4
00:00:19,600 --> 00:00:24,560
awesome as the catering was here

5
00:00:21,600 --> 00:00:26,160
um i would like to introduce simon uh

6
00:00:24,560 --> 00:00:28,080
simon will be discussing with us the

7
00:00:26,160 --> 00:00:31,199
architecture and software that they

8
00:00:28,080 --> 00:00:33,760
chose um in making their api public

9
00:00:31,199 --> 00:00:36,000
simon is a creative technologist at acme

10
00:00:33,760 --> 00:00:38,399
your museum of screen culture which is

11
00:00:36,000 --> 00:00:41,200
in melbourne a software developer

12
00:00:38,399 --> 00:00:44,239
maintainer at a new internationalist

13
00:00:41,200 --> 00:00:46,879
magazine and an open source uh software

14
00:00:44,239 --> 00:00:48,800
and hardware enthusiast uh he helped set

15
00:00:46,879 --> 00:00:51,280
up the mod which is the museum of

16
00:00:48,800 --> 00:00:54,239
discovery and hacker space in adelaide

17
00:00:51,280 --> 00:00:55,440
so on to you simon thank you

18
00:00:54,239 --> 00:00:56,399
thank you

19
00:00:55,440 --> 00:01:01,120
um

20
00:00:56,399 --> 00:01:01,120
i will set my screen sharing going

21
00:01:04,960 --> 00:01:07,439
okay

22
00:01:07,680 --> 00:01:11,840
so hi everybody uh first i'd like to

23
00:01:10,000 --> 00:01:13,920
acknowledge the ghana people as the

24
00:01:11,840 --> 00:01:16,000
traditional owners of the unseated lands

25
00:01:13,920 --> 00:01:18,880
that i'm speaking from today uh always

26
00:01:16,000 --> 00:01:21,439
was always would be aboriginal land

27
00:01:18,880 --> 00:01:24,000
and for those who haven't been to acne

28
00:01:21,439 --> 00:01:26,400
we meet your museum of screen culture in

29
00:01:24,000 --> 00:01:28,560
federation square melbourne victoria

30
00:01:26,400 --> 00:01:30,960
we usually have about 1.5 million

31
00:01:28,560 --> 00:01:32,720
visitors a year and our new experience

32
00:01:30,960 --> 00:01:34,400
the story of the moving image takes you

33
00:01:32,720 --> 00:01:38,720
on an interactive journey of screen

34
00:01:34,400 --> 00:01:40,560
culture in films video games tv and art

35
00:01:38,720 --> 00:01:41,600
it's also free so you should all come

36
00:01:40,560 --> 00:01:44,079
and visit

37
00:01:41,600 --> 00:01:46,320
um last year i talked about how we built

38
00:01:44,079 --> 00:01:49,520
the digital infrastructure and private

39
00:01:46,320 --> 00:01:52,840
apis that power acme uh including how we

40
00:01:49,520 --> 00:01:55,200
deploy about 400 raspberry pi's and dell

41
00:01:52,840 --> 00:01:56,799
microcomputers uh and this year i'm back

42
00:01:55,200 --> 00:01:59,520
to talk about the work that our team has

43
00:01:56,799 --> 00:02:01,439
done releasing our first public api

44
00:01:59,520 --> 00:02:02,740
uh you can see it in your browser at

45
00:02:01,439 --> 00:02:05,200
api.acme.net.edu

46
00:02:02,740 --> 00:02:06,840
[Music]

47
00:02:05,200 --> 00:02:09,759
that's a lovely little screenshot up

48
00:02:06,840 --> 00:02:13,120
here um so

49
00:02:09,759 --> 00:02:15,440
how uh why did we build it um

50
00:02:13,120 --> 00:02:17,920
we received public funding

51
00:02:15,440 --> 00:02:20,480
at acme so we figured that the data that

52
00:02:17,920 --> 00:02:23,120
we make should be public too

53
00:02:20,480 --> 00:02:26,239
and what do i mean by the data we make

54
00:02:23,120 --> 00:02:28,400
um well museum collections might include

55
00:02:26,239 --> 00:02:30,400
uh works that are held on behalf of

56
00:02:28,400 --> 00:02:33,040
communities whose intellectual property

57
00:02:30,400 --> 00:02:34,480
is owned by the creators until it enters

58
00:02:33,040 --> 00:02:36,400
the public domain

59
00:02:34,480 --> 00:02:38,560
um and also scholarly

60
00:02:36,400 --> 00:02:40,480
scholarly descriptive metadata about

61
00:02:38,560 --> 00:02:42,720
those works that museums create

62
00:02:40,480 --> 00:02:45,280
themselves so we're allowed to release

63
00:02:42,720 --> 00:02:47,920
the descriptive metadata um but not the

64
00:02:45,280 --> 00:02:51,120
works in most cases um

65
00:02:47,920 --> 00:02:53,440
uh cxo seb chan uh describes this really

66
00:02:51,120 --> 00:02:55,040
well so i'll just pop in a bit

67
00:02:53,440 --> 00:02:57,519
from him so he

68
00:02:55,040 --> 00:02:59,040
writes uh acme might hold a copy of mad

69
00:02:57,519 --> 00:03:01,280
max but

70
00:02:59,040 --> 00:03:03,040
owns no ip to release even when it

71
00:03:01,280 --> 00:03:05,440
digitizes that copy because the

72
00:03:03,040 --> 00:03:07,519
underlying work is still in copyright

73
00:03:05,440 --> 00:03:09,680
however acme owns all of the ip around

74
00:03:07,519 --> 00:03:11,200
what it says about mad max and can

75
00:03:09,680 --> 00:03:12,319
release that without seeking additional

76
00:03:11,200 --> 00:03:15,360
permissions

77
00:03:12,319 --> 00:03:17,200
so other museums hold things whose

78
00:03:15,360 --> 00:03:19,440
underlying ip is now in the public

79
00:03:17,200 --> 00:03:20,959
domain because they are so old

80
00:03:19,440 --> 00:03:22,080
this allows them to release the work

81
00:03:20,959 --> 00:03:24,560
itself

82
00:03:22,080 --> 00:03:26,560
acme's collection is very young

83
00:03:24,560 --> 00:03:28,799
and doesn't have this option

84
00:03:26,560 --> 00:03:30,959
we do have some works that were

85
00:03:28,799 --> 00:03:32,879
produced as government films

86
00:03:30,959 --> 00:03:34,640
and so they fall under the crown

87
00:03:32,879 --> 00:03:36,000
copyright as well as other works that

88
00:03:34,640 --> 00:03:38,080
are called orphan works because they

89
00:03:36,000 --> 00:03:39,760
have no known copyright owner

90
00:03:38,080 --> 00:03:41,680
uh there are a third set of works whose

91
00:03:39,760 --> 00:03:43,840
creators have given us limited rights to

92
00:03:41,680 --> 00:03:46,879
release on other platforms controlled

93
00:03:43,840 --> 00:03:50,239
and administered by acme and these are

94
00:03:46,879 --> 00:03:50,239
what are on our youtube channel

95
00:03:50,480 --> 00:03:55,040
um another reason we wanted to build it

96
00:03:52,640 --> 00:03:57,280
is for regular updates so our last

97
00:03:55,040 --> 00:03:58,720
starter release was five years ago and

98
00:03:57,280 --> 00:04:02,799
we'd really like to release more

99
00:03:58,720 --> 00:04:05,599
frequently so nightly if possible

100
00:04:02,799 --> 00:04:08,319
uh we wanted to build for accessibility

101
00:04:05,599 --> 00:04:09,599
so when we build the public labels api

102
00:04:08,319 --> 00:04:11,439
um

103
00:04:09,599 --> 00:04:13,439
that one is what we're going to use so

104
00:04:11,439 --> 00:04:15,200
that visitors can translate and read all

105
00:04:13,439 --> 00:04:18,160
of the labels that appear at acme using

106
00:04:15,200 --> 00:04:18,160
their own screen readers

107
00:04:18,239 --> 00:04:22,880
and we also wanted to be able to create

108
00:04:20,880 --> 00:04:24,479
more rich data links

109
00:04:22,880 --> 00:04:26,479
so that we can match collections data

110
00:04:24,479 --> 00:04:28,639
with other institutions so this might

111
00:04:26,479 --> 00:04:30,720
include ids from other film collections

112
00:04:28,639 --> 00:04:33,600
like the dff in germany or else data

113
00:04:30,720 --> 00:04:36,000
sources like wikitata

114
00:04:33,600 --> 00:04:38,000
and we also love experimentation so we

115
00:04:36,000 --> 00:04:40,560
wanted to build applications that extend

116
00:04:38,000 --> 00:04:42,639
and transform our data

117
00:04:40,560 --> 00:04:44,800
and uh public api is super useful for

118
00:04:42,639 --> 00:04:44,800
that

119
00:04:45,440 --> 00:04:49,840
so uh before we built it we needed to

120
00:04:48,160 --> 00:04:50,800
answer who we were really building it

121
00:04:49,840 --> 00:04:52,960
for

122
00:04:50,800 --> 00:04:55,280
um primarily we're building it for

123
00:04:52,960 --> 00:04:57,199
researchers who might use a study

124
00:04:55,280 --> 00:04:59,040
representation and demographics in our

125
00:04:57,199 --> 00:05:02,160
data

126
00:04:59,040 --> 00:05:03,600
we also thought we should build it for

127
00:05:02,160 --> 00:05:05,520
technologists

128
00:05:03,600 --> 00:05:07,759
and take that into account so envisage

129
00:05:05,520 --> 00:05:10,160
them using it in hackathons and to build

130
00:05:07,759 --> 00:05:12,320
visualizations

131
00:05:10,160 --> 00:05:14,560
we also need it for internal use

132
00:05:12,320 --> 00:05:16,639
so we've got three teams technical teams

133
00:05:14,560 --> 00:05:18,560
at acme the audio visual team the

134
00:05:16,639 --> 00:05:20,479
information technology team and also the

135
00:05:18,560 --> 00:05:21,840
team that i belong to experience in

136
00:05:20,479 --> 00:05:23,120
digital

137
00:05:21,840 --> 00:05:24,880
and we've all got creative people in

138
00:05:23,120 --> 00:05:26,800
there who love writing code so the

139
00:05:24,880 --> 00:05:29,360
public api allows us to develop

140
00:05:26,800 --> 00:05:32,320
applications with our collections data

141
00:05:29,360 --> 00:05:34,960
without the fear of bringing down the

142
00:05:32,320 --> 00:05:36,800
private museums api um because that

143
00:05:34,960 --> 00:05:39,680
would suck

144
00:05:36,800 --> 00:05:41,759
um and we also built it for you so we

145
00:05:39,680 --> 00:05:43,440
would love to hear what you think of it

146
00:05:41,759 --> 00:05:47,039
if there's anything we can improve and

147
00:05:43,440 --> 00:05:49,280
most importantly what you build with it

148
00:05:47,039 --> 00:05:53,840
um here are a couple of examples from

149
00:05:49,280 --> 00:05:55,280
rmit students who used our 2016 metadata

150
00:05:53,840 --> 00:05:59,120
so here they explore gender

151
00:05:55,280 --> 00:06:00,800
representation uh on the left and also

152
00:05:59,120 --> 00:06:03,039
the country of origin of works in our

153
00:06:00,800 --> 00:06:05,520
collection on the right

154
00:06:03,039 --> 00:06:06,960
they generated these images using the

155
00:06:05,520 --> 00:06:08,160
metadata spreadsheet that we really

156
00:06:06,960 --> 00:06:09,919
expect them

157
00:06:08,160 --> 00:06:13,199
but with the public api these could now

158
00:06:09,919 --> 00:06:13,199
be live and interactive

159
00:06:15,440 --> 00:06:20,479
so this is a big dependency map of xos

160
00:06:19,280 --> 00:06:21,919
that we built

161
00:06:20,479 --> 00:06:23,199
last year that i presented last year at

162
00:06:21,919 --> 00:06:24,880
the conference

163
00:06:23,199 --> 00:06:26,880
and all of the applications it serves in

164
00:06:24,880 --> 00:06:28,479
our museum and external facing websites

165
00:06:26,880 --> 00:06:30,560
um don't worry if you can't read it it's

166
00:06:28,479 --> 00:06:31,919
just to give you an idea of all the

167
00:06:30,560 --> 00:06:34,000
things

168
00:06:31,919 --> 00:06:35,039
so as you can see a lot depends on the

169
00:06:34,000 --> 00:06:36,720
xos

170
00:06:35,039 --> 00:06:39,120
uptime and performance

171
00:06:36,720 --> 00:06:40,960
and hence the need for a public ati api

172
00:06:39,120 --> 00:06:43,199
not to put any extra strain on it during

173
00:06:40,960 --> 00:06:46,080
museum opening hours

174
00:06:43,199 --> 00:06:49,360
um the private apis it exposes currently

175
00:06:46,080 --> 00:06:52,080
uh the taps when people tap their lens

176
00:06:49,360 --> 00:06:55,280
uh a lenses api for the lens itself for

177
00:06:52,080 --> 00:06:58,560
constellations works creators playlists

178
00:06:55,280 --> 00:07:00,800
monitor devices labels images and videos

179
00:06:58,560 --> 00:07:02,880
so over time we'll release all these as

180
00:07:00,800 --> 00:07:04,720
open apis

181
00:07:02,880 --> 00:07:06,720
but first we wanted to focus on the

182
00:07:04,720 --> 00:07:09,280
collection works api which is the one

183
00:07:06,720 --> 00:07:11,120
i'm talking about now

184
00:07:09,280 --> 00:07:12,240
so how do we build it

185
00:07:11,120 --> 00:07:14,720
um

186
00:07:12,240 --> 00:07:16,960
so we had xos which is our museum

187
00:07:14,720 --> 00:07:17,840
operating system

188
00:07:16,960 --> 00:07:20,800
and

189
00:07:17,840 --> 00:07:22,560
that was really lovely and so what we

190
00:07:20,800 --> 00:07:23,919
wanted to build

191
00:07:22,560 --> 00:07:26,400
was

192
00:07:23,919 --> 00:07:28,880
a glorified offline cache that was

193
00:07:26,400 --> 00:07:32,400
serving json from the file system

194
00:07:28,880 --> 00:07:34,560
um and so the public code repository of

195
00:07:32,400 --> 00:07:36,960
the public api that we've released

196
00:07:34,560 --> 00:07:38,479
includes the api server software itself

197
00:07:36,960 --> 00:07:41,039
as well as the

198
00:07:38,479 --> 00:07:43,680
collections metadata in both json and

199
00:07:41,039 --> 00:07:47,039
tsv file formats

200
00:07:43,680 --> 00:07:50,160
so most of the rights filtering is done

201
00:07:47,039 --> 00:07:51,680
in xos and that inherits the logic from

202
00:07:50,160 --> 00:07:54,080
our collections

203
00:07:51,680 --> 00:07:56,960
cms which is called vernon

204
00:07:54,080 --> 00:07:58,639
but some of the last minute tasks

205
00:07:56,960 --> 00:08:00,400
or we had some last minute risk adverse

206
00:07:58,639 --> 00:08:01,759
decisions by the organization which

207
00:08:00,400 --> 00:08:04,319
meant we had to do a little bit more

208
00:08:01,759 --> 00:08:05,599
rights filtering on the public api so

209
00:08:04,319 --> 00:08:07,199
um

210
00:08:05,599 --> 00:08:08,960
we've got some logic that we're going to

211
00:08:07,199 --> 00:08:11,120
roll back into xos

212
00:08:08,960 --> 00:08:13,280
after we've built the business decision

213
00:08:11,120 --> 00:08:14,960
fields into vernon so it'll get cleaner

214
00:08:13,280 --> 00:08:16,800
over time which is great

215
00:08:14,960 --> 00:08:18,639
um but having the api code in the

216
00:08:16,800 --> 00:08:20,479
metadata on the one repo

217
00:08:18,639 --> 00:08:22,240
uh means that researchers can get clone

218
00:08:20,479 --> 00:08:23,440
our repo and have all of the data ready

219
00:08:22,240 --> 00:08:25,840
to go

220
00:08:23,440 --> 00:08:28,319
plus it also isolates our private museum

221
00:08:25,840 --> 00:08:29,599
apis completely from the public so that

222
00:08:28,319 --> 00:08:32,879
they don't have any chance of

223
00:08:29,599 --> 00:08:32,879
accidentally breaking things either

224
00:08:33,919 --> 00:08:39,360
so what we used was alpine linux docker

225
00:08:37,680 --> 00:08:41,760
container

226
00:08:39,360 --> 00:08:42,959
the main reasons are it's small secure

227
00:08:41,760 --> 00:08:45,200
fast

228
00:08:42,959 --> 00:08:46,080
you can do repeatable builds and use the

229
00:08:45,200 --> 00:08:47,760
same

230
00:08:46,080 --> 00:08:50,480
image for development and production

231
00:08:47,760 --> 00:08:52,720
which is really nice

232
00:08:50,480 --> 00:08:55,200
in terms of the underlying software

233
00:08:52,720 --> 00:08:56,800
the majority of our code acme is written

234
00:08:55,200 --> 00:08:58,880
in python so the dream result was a

235
00:08:56,800 --> 00:09:00,560
python based framework with minimal

236
00:08:58,880 --> 00:09:02,160
dependencies that was quick and easy to

237
00:09:00,560 --> 00:09:06,080
prototype and develop

238
00:09:02,160 --> 00:09:06,080
so flask was really good choice for that

239
00:09:06,959 --> 00:09:11,760
in terms of search we went with um

240
00:09:09,519 --> 00:09:13,920
elastic cloud in production

241
00:09:11,760 --> 00:09:15,760
uh and a matched elasticsearch docker

242
00:09:13,920 --> 00:09:17,680
image for development so the same

243
00:09:15,760 --> 00:09:20,640
re-indexing code runs on staging and

244
00:09:17,680 --> 00:09:22,720
development which allows us for easy

245
00:09:20,640 --> 00:09:25,120
updates and testing and bug fixing which

246
00:09:22,720 --> 00:09:25,120
is great

247
00:09:25,920 --> 00:09:31,279
in terms of the infrastructure

248
00:09:28,480 --> 00:09:32,640
the acme ict team has a background with

249
00:09:31,279 --> 00:09:34,320
microsoft

250
00:09:32,640 --> 00:09:37,040
so we use terraform to build the

251
00:09:34,320 --> 00:09:39,440
infrastructure on azure services

252
00:09:37,040 --> 00:09:41,680
making use of horizontal auto scaling to

253
00:09:39,440 --> 00:09:45,040
handle temporary load increases if the

254
00:09:41,680 --> 00:09:46,880
api cpus average 70 or higher for a

255
00:09:45,040 --> 00:09:47,680
significant amount of time

256
00:09:46,880 --> 00:09:49,120
um

257
00:09:47,680 --> 00:09:50,720
we're finding terraform pretty good

258
00:09:49,120 --> 00:09:52,160
because it makes spinning up new servers

259
00:09:50,720 --> 00:09:54,080
quite easy

260
00:09:52,160 --> 00:09:57,600
though it adds quite a bit more tech

261
00:09:54,080 --> 00:09:57,600
debt to our small team of two dabs

262
00:09:58,720 --> 00:10:01,519
in terms of

263
00:10:00,080 --> 00:10:02,959
nightly updates

264
00:10:01,519 --> 00:10:05,279
to the api

265
00:10:02,959 --> 00:10:07,680
um while building a separate repo for

266
00:10:05,279 --> 00:10:10,320
the api updater i realized that a cron

267
00:10:07,680 --> 00:10:13,040
job in the actual public api code repo

268
00:10:10,320 --> 00:10:16,399
could do the same job so now cronjob

269
00:10:13,040 --> 00:10:19,360
runs the same code which calls our xos

270
00:10:16,399 --> 00:10:20,720
private works api uh it downloads the

271
00:10:19,360 --> 00:10:23,440
json blobs

272
00:10:20,720 --> 00:10:26,160
and saves them to the file system

273
00:10:23,440 --> 00:10:28,079
and then we've got another little bit of

274
00:10:26,160 --> 00:10:30,079
code that removes the image and videos

275
00:10:28,079 --> 00:10:32,480
that we don't have rights to and once

276
00:10:30,079 --> 00:10:34,000
it's finished it just checks for the git

277
00:10:32,480 --> 00:10:36,160
changes

278
00:10:34,000 --> 00:10:38,079
if it finds them then it pushes those

279
00:10:36,160 --> 00:10:40,240
changes back to

280
00:10:38,079 --> 00:10:42,720
the code the github repo

281
00:10:40,240 --> 00:10:44,640
which triggers a github action to run

282
00:10:42,720 --> 00:10:47,120
the linting and tests before deploying

283
00:10:44,640 --> 00:10:50,399
the changes to staging on top of itself

284
00:10:47,120 --> 00:10:53,600
which is quite cool so that's how

285
00:10:50,399 --> 00:10:56,000
it updates the code repo by itself

286
00:10:53,600 --> 00:10:58,160
um we have an environment variable for

287
00:10:56,000 --> 00:11:00,720
toggling updates via terraform

288
00:10:58,160 --> 00:11:02,640
um in a config map so this makes it

289
00:11:00,720 --> 00:11:05,200
really easy to turn off any updates if

290
00:11:02,640 --> 00:11:07,040
um without any code changes the api repo

291
00:11:05,200 --> 00:11:09,040
just in case we have any problems

292
00:11:07,040 --> 00:11:12,160
and we also have environment variables

293
00:11:09,040 --> 00:11:14,000
for including images and videos so

294
00:11:12,160 --> 00:11:15,680
over time as we've got all the rights

295
00:11:14,000 --> 00:11:17,360
ready uh we'll add them to the public

296
00:11:15,680 --> 00:11:20,320
api just via a

297
00:11:17,360 --> 00:11:23,519
terraform toggle which is cool

298
00:11:20,320 --> 00:11:27,279
um in production uh we trigger manual

299
00:11:23,519 --> 00:11:28,720
releases uh weekly and this is done by

300
00:11:27,279 --> 00:11:30,240
github

301
00:11:28,720 --> 00:11:31,600
after we've done some data validation on

302
00:11:30,240 --> 00:11:33,120
staging just to make sure we haven't

303
00:11:31,600 --> 00:11:35,680
broken anything

304
00:11:33,120 --> 00:11:38,480
and then we use unicorn server to serve

305
00:11:35,680 --> 00:11:38,480
the json to you

306
00:11:39,600 --> 00:11:42,160
in development

307
00:11:40,800 --> 00:11:44,079
we provide an environment variable to

308
00:11:42,160 --> 00:11:46,399
switch between the flask server while

309
00:11:44,079 --> 00:11:47,760
building new features so you can do

310
00:11:46,399 --> 00:11:50,160
hot reloading

311
00:11:47,760 --> 00:11:51,600
and also the genocon server for testing

312
00:11:50,160 --> 00:11:52,720
as well just if you want to test the

313
00:11:51,600 --> 00:11:55,200
production

314
00:11:52,720 --> 00:11:57,760
server on your laptop too

315
00:11:55,200 --> 00:11:59,839
um so i thought i'd give a quick demo of

316
00:11:57,760 --> 00:12:02,079
how all of that runs using docker

317
00:11:59,839 --> 00:12:02,800
compose

318
00:12:02,079 --> 00:12:05,680
so

319
00:12:02,800 --> 00:12:08,160
um we pull a clone of the repo

320
00:12:05,680 --> 00:12:10,560
we see the insert development folder and

321
00:12:08,160 --> 00:12:13,760
then simply run docker composer

322
00:12:10,560 --> 00:12:17,279
and that starts the api search container

323
00:12:13,760 --> 00:12:20,320
with elasticsearch and also the api

324
00:12:17,279 --> 00:12:22,639
itself so here it is running

325
00:12:20,320 --> 00:12:24,639
um if we switch over to our browser and

326
00:12:22,639 --> 00:12:26,399
head to port 8081

327
00:12:24,639 --> 00:12:28,959
uh there you can see it

328
00:12:26,399 --> 00:12:31,680
um so you can also go to works and see

329
00:12:28,959 --> 00:12:32,800
an index page of all the 42 000

330
00:12:31,680 --> 00:12:35,120
um

331
00:12:32,800 --> 00:12:36,320
files there and you can also paste in

332
00:12:35,120 --> 00:12:40,720
the id

333
00:12:36,320 --> 00:12:42,240
and grab like a one object uh returned

334
00:12:40,720 --> 00:12:43,760
and this is just showing that the the

335
00:12:42,240 --> 00:12:46,160
search works locally as well so that's

336
00:12:43,760 --> 00:12:48,880
elasticsearch returning the results

337
00:12:46,160 --> 00:12:48,880
of a search too

338
00:12:49,279 --> 00:12:56,480
so um what were the results um so

339
00:12:52,880 --> 00:12:58,880
without any optimization to unicorn yet

340
00:12:56,480 --> 00:12:59,680
um this is the speed that we got coming

341
00:12:58,880 --> 00:13:01,760
back

342
00:12:59,680 --> 00:13:03,920
for all of our requests so when they're

343
00:13:01,760 --> 00:13:06,800
about three requests a second it's about

344
00:13:03,920 --> 00:13:08,240
158 milliseconds going up to 18 requests

345
00:13:06,800 --> 00:13:10,959
a second about

346
00:13:08,240 --> 00:13:14,480
you know half a second so um that's

347
00:13:10,959 --> 00:13:17,360
running pretty well for 100 a month

348
00:13:14,480 --> 00:13:20,800
um and this is on the standard azure

349
00:13:17,360 --> 00:13:23,519
d2v2 nodes

350
00:13:20,800 --> 00:13:25,440
in terms of time to build this

351
00:13:23,519 --> 00:13:27,600
the first commit we made just doing the

352
00:13:25,440 --> 00:13:29,519
rough prototypes using flask was on the

353
00:13:27,600 --> 00:13:32,079
5th of july

354
00:13:29,519 --> 00:13:33,760
and then on the 23rd of september we had

355
00:13:32,079 --> 00:13:35,279
our infrastructure running and

356
00:13:33,760 --> 00:13:36,320
everything running in production which

357
00:13:35,279 --> 00:13:37,920
was cool

358
00:13:36,320 --> 00:13:40,320
so that gave us a lot of time to do the

359
00:13:37,920 --> 00:13:42,480
final touches for the 11th of november

360
00:13:40,320 --> 00:13:43,360
so about four months

361
00:13:42,480 --> 00:13:46,000
um

362
00:13:43,360 --> 00:13:48,320
in terms of dependencies and security um

363
00:13:46,000 --> 00:13:50,480
because of the limited number of alpine

364
00:13:48,320 --> 00:13:52,240
and python dependencies we don't pin

365
00:13:50,480 --> 00:13:54,320
dependencies at all so they update

366
00:13:52,240 --> 00:13:56,399
nightly with every build giving us a

367
00:13:54,320 --> 00:13:58,160
pretty reasonable security posture which

368
00:13:56,399 --> 00:13:59,120
is nice

369
00:13:58,160 --> 00:14:01,440
um

370
00:13:59,120 --> 00:14:03,440
and we've documented everything over at

371
00:14:01,440 --> 00:14:05,040
acme dot ju

372
00:14:03,440 --> 00:14:07,120
api so

373
00:14:05,040 --> 00:14:08,560
hopefully the document is documentation

374
00:14:07,120 --> 00:14:09,360
is easy to read but

375
00:14:08,560 --> 00:14:11,519
um

376
00:14:09,360 --> 00:14:14,240
have a go let me know if it needs any

377
00:14:11,519 --> 00:14:14,240
updates or anything

378
00:14:15,040 --> 00:14:20,720
um we've also got a blog post um with a

379
00:14:18,720 --> 00:14:23,600
bit more history and some prototypes so

380
00:14:20,720 --> 00:14:24,560
you can pop there at labs.acme.nb

381
00:14:23,600 --> 00:14:26,720
um

382
00:14:24,560 --> 00:14:28,160
and in terms of the prototypes um

383
00:14:26,720 --> 00:14:29,760
because we found the majority of bugs

384
00:14:28,160 --> 00:14:31,839
early in our testing

385
00:14:29,760 --> 00:14:34,000
because we built the private apis

386
00:14:31,839 --> 00:14:36,160
that was great but using the public api

387
00:14:34,000 --> 00:14:38,800
ourselves for prototypes helped uncover

388
00:14:36,160 --> 00:14:40,720
a few more which was really useful

389
00:14:38,800 --> 00:14:41,839
so both the prototype applications we

390
00:14:40,720 --> 00:14:44,160
built

391
00:14:41,839 --> 00:14:45,760
run in jupyter notebooks which means

392
00:14:44,160 --> 00:14:47,600
they can be run for free on google

393
00:14:45,760 --> 00:14:49,120
collab online so you don't need to run

394
00:14:47,600 --> 00:14:49,839
any of this code on your own computer if

395
00:14:49,120 --> 00:14:51,680
you're

396
00:14:49,839 --> 00:14:54,320
um ultra paranoid

397
00:14:51,680 --> 00:14:56,720
or just usually paranoid

398
00:14:54,320 --> 00:14:58,079
um so the first one we built was machine

399
00:14:56,720 --> 00:14:59,120
dreaming

400
00:14:58,079 --> 00:15:02,720
this one

401
00:14:59,120 --> 00:15:04,399
generates images from our works metadata

402
00:15:02,720 --> 00:15:06,320
so we thought it might be useful seeing

403
00:15:04,399 --> 00:15:08,880
that we didn't actually release the

404
00:15:06,320 --> 00:15:10,720
images and videos for launch

405
00:15:08,880 --> 00:15:13,519
and the second one

406
00:15:10,720 --> 00:15:15,519
matches wiki data biographies to

407
00:15:13,519 --> 00:15:18,399
acme creators

408
00:15:15,519 --> 00:15:20,959
so there so that's quite cool too

409
00:15:18,399 --> 00:15:22,560
um then since we've released it we've

410
00:15:20,959 --> 00:15:26,240
had some public use which is really

411
00:15:22,560 --> 00:15:28,240
really nice um so paul from nfsa which

412
00:15:26,240 --> 00:15:31,040
is the national film and sound archive

413
00:15:28,240 --> 00:15:34,720
of australia uh he used the api to match

414
00:15:31,040 --> 00:15:36,639
acme works to wikidata entities so um

415
00:15:34,720 --> 00:15:38,160
we started matching craters he's

416
00:15:36,639 --> 00:15:39,519
matching works now and it's really

417
00:15:38,160 --> 00:15:40,480
really lovely

418
00:15:39,519 --> 00:15:43,279
um

419
00:15:40,480 --> 00:15:45,120
the other use was by aaron from san

420
00:15:43,279 --> 00:15:47,279
francisco museum

421
00:15:45,120 --> 00:15:49,199
he matched acme accession numbers to

422
00:15:47,279 --> 00:15:51,440
website urls which is going to be really

423
00:15:49,199 --> 00:15:53,600
handy for accessible accessibility

424
00:15:51,440 --> 00:15:57,120
labels so he's working on a project so

425
00:15:53,600 --> 00:15:59,839
that all museums can share this code um

426
00:15:57,120 --> 00:16:01,440
which will read a museum label and take

427
00:15:59,839 --> 00:16:03,519
you directly to the website which is

428
00:16:01,440 --> 00:16:04,800
very cool

429
00:16:03,519 --> 00:16:06,720
uh the other thing we learned about just

430
00:16:04,800 --> 00:16:07,839
recently is that wikimedia australia

431
00:16:06,720 --> 00:16:09,600
have launched a thousand dollar

432
00:16:07,839 --> 00:16:11,279
fellowships um

433
00:16:09,600 --> 00:16:13,440
and they're featuring the acme public

434
00:16:11,279 --> 00:16:14,240
api as an example project

435
00:16:13,440 --> 00:16:15,279
so

436
00:16:14,240 --> 00:16:16,800
if you're into data you should

437
00:16:15,279 --> 00:16:19,519
definitely apply for that

438
00:16:16,800 --> 00:16:19,519
looks really fun

439
00:16:20,079 --> 00:16:26,639
so um here you can see the first machine

440
00:16:23,440 --> 00:16:29,279
dreaming prototype um that we created so

441
00:16:26,639 --> 00:16:30,959
this is using the metadata from um the

442
00:16:29,279 --> 00:16:32,240
matrix

443
00:16:30,959 --> 00:16:33,680
and so

444
00:16:32,240 --> 00:16:35,600
i don't know if you know i'll give you a

445
00:16:33,680 --> 00:16:37,199
little bit of history about gans so gans

446
00:16:35,600 --> 00:16:38,720
are machine learning models that pit one

447
00:16:37,199 --> 00:16:40,560
network that generates the output

448
00:16:38,720 --> 00:16:42,240
against another that classifies that

449
00:16:40,560 --> 00:16:44,480
output

450
00:16:42,240 --> 00:16:46,639
so they're used a lot in generative

451
00:16:44,480 --> 00:16:49,199
music texts and images

452
00:16:46,639 --> 00:16:51,040
and vqgan plus clip which is

453
00:16:49,199 --> 00:16:53,199
what we're using here

454
00:16:51,040 --> 00:16:54,639
that combines

455
00:16:53,199 --> 00:16:57,360
machine learning models that generate

456
00:16:54,639 --> 00:16:59,759
the image from text

457
00:16:57,360 --> 00:17:03,199
so that's we use that to

458
00:16:59,759 --> 00:17:04,720
pass the acme metadata in this case

459
00:17:03,199 --> 00:17:06,480
it's split

460
00:17:04,720 --> 00:17:09,120
so we pass in the title first and then

461
00:17:06,480 --> 00:17:10,959
the um brief description

462
00:17:09,120 --> 00:17:13,280
uh all split by a maximum number of

463
00:17:10,959 --> 00:17:15,760
characters which it can handle uh and it

464
00:17:13,280 --> 00:17:16,720
generates that image

465
00:17:15,760 --> 00:17:18,240
so

466
00:17:16,720 --> 00:17:19,919
as you probably know machine learning

467
00:17:18,240 --> 00:17:22,240
models carry the biases which they're

468
00:17:19,919 --> 00:17:24,400
trained on and so it looks likely that

469
00:17:22,240 --> 00:17:26,559
there's probably a matrix poster in the

470
00:17:24,400 --> 00:17:28,559
training data that we use given how

471
00:17:26,559 --> 00:17:30,000
similarly that looks

472
00:17:28,559 --> 00:17:32,160
um

473
00:17:30,000 --> 00:17:33,600
likewise the strength image from our mad

474
00:17:32,160 --> 00:17:35,280
max metadata

475
00:17:33,600 --> 00:17:37,120
um it looks like there might have been

476
00:17:35,280 --> 00:17:39,600
some film stills in the training data as

477
00:17:37,120 --> 00:17:41,039
well given how accurate

478
00:17:39,600 --> 00:17:42,640
uh that looks to some of the characters

479
00:17:41,039 --> 00:17:45,360
in there

480
00:17:42,640 --> 00:17:47,600
and so it's showing like every frame is

481
00:17:45,360 --> 00:17:49,679
one more step um

482
00:17:47,600 --> 00:17:52,240
that it thinks

483
00:17:49,679 --> 00:17:54,080
uh is closer to the image that it's

484
00:17:52,240 --> 00:17:56,080
dreamt up based on that metadata so

485
00:17:54,080 --> 00:17:57,200
there's about 200 steps per

486
00:17:56,080 --> 00:18:00,960
um

487
00:17:57,200 --> 00:18:03,679
in our example that we used

488
00:18:00,960 --> 00:18:04,640
uh here we fed the video game of the

489
00:18:03,679 --> 00:18:06,640
hobbit

490
00:18:04,640 --> 00:18:08,320
um but clearly the training data had an

491
00:18:06,640 --> 00:18:10,880
image of the book in there so

492
00:18:08,320 --> 00:18:12,840
um it kind of dropped up the book rather

493
00:18:10,880 --> 00:18:15,440
than a video

494
00:18:12,840 --> 00:18:16,240
game which is not surprising saying it's

495
00:18:15,440 --> 00:18:17,760
a

496
00:18:16,240 --> 00:18:20,160
spectrum

497
00:18:17,760 --> 00:18:21,120
game so yeah

498
00:18:20,160 --> 00:18:23,039
um

499
00:18:21,120 --> 00:18:24,320
the last one here um the training set

500
00:18:23,039 --> 00:18:26,559
also appears to know that outback

501
00:18:24,320 --> 00:18:28,799
australia is aboriginal land too which

502
00:18:26,559 --> 00:18:30,799
is very nice so if you'd like to try

503
00:18:28,799 --> 00:18:32,799
some others as i mentioned our notebooks

504
00:18:30,799 --> 00:18:34,640
run for free on google collab servers

505
00:18:32,799 --> 00:18:35,520
and there's a link to it in our acme git

506
00:18:34,640 --> 00:18:38,480
lab

507
00:18:35,520 --> 00:18:39,919
um from the blog post api documentation

508
00:18:38,480 --> 00:18:41,840
and also all of these links in the

509
00:18:39,919 --> 00:18:45,120
slides work so

510
00:18:41,840 --> 00:18:47,520
you can download them after my talk

511
00:18:45,120 --> 00:18:50,000
so in terms of challenges

512
00:18:47,520 --> 00:18:53,440
um so we wanted to release everything as

513
00:18:50,000 --> 00:18:56,640
creative commons zero licenses

514
00:18:53,440 --> 00:18:57,440
but yeah we couldn't for many reasons

515
00:18:56,640 --> 00:19:00,559
um

516
00:18:57,440 --> 00:19:02,080
seb chan uh the ct cxo uh has some

517
00:19:00,559 --> 00:19:04,320
really nice words about this so he

518
00:19:02,080 --> 00:19:05,679
writes uh this is not hard because of

519
00:19:04,320 --> 00:19:07,840
technical reasons but because of

520
00:19:05,679 --> 00:19:10,320
copyright laws ethical consideration of

521
00:19:07,840 --> 00:19:12,720
creators rats outside of copyright laws

522
00:19:10,320 --> 00:19:14,320
such as indigenous ip so should

523
00:19:12,720 --> 00:19:15,440
indigenous works ever enter the public

524
00:19:14,320 --> 00:19:17,919
domain

525
00:19:15,440 --> 00:19:20,480
and other also other moral rights

526
00:19:17,919 --> 00:19:22,480
acme and no museum has

527
00:19:20,480 --> 00:19:24,799
resources to manually research

528
00:19:22,480 --> 00:19:26,880
contact clear and re-license every work

529
00:19:24,799 --> 00:19:28,880
in its collection for all purposes

530
00:19:26,880 --> 00:19:30,720
so some works we have in our collection

531
00:19:28,880 --> 00:19:33,280
even need us to contact the creator or

532
00:19:30,720 --> 00:19:34,960
their estate to show in an exhibition

533
00:19:33,280 --> 00:19:37,360
other works were acquired before the web

534
00:19:34,960 --> 00:19:40,000
existed and thus don't have flexibility

535
00:19:37,360 --> 00:19:42,160
without re-contacting the rights holder

536
00:19:40,000 --> 00:19:44,000
so we had to remo remove a large portion

537
00:19:42,160 --> 00:19:46,960
of our collection from our public api

538
00:19:44,000 --> 00:19:46,960
for this first release

539
00:19:47,280 --> 00:19:54,160
um the other challenge is expectations

540
00:19:51,600 --> 00:19:56,480
um instead of right again museum data

541
00:19:54,160 --> 00:19:58,320
isn't very useful um without being

542
00:19:56,480 --> 00:20:00,960
combined with other data sources there's

543
00:19:58,320 --> 00:20:02,000
no easy open gov model for reuse to

544
00:20:00,960 --> 00:20:04,080
apply

545
00:20:02,000 --> 00:20:07,760
uh there is no when is the next bus

546
00:20:04,080 --> 00:20:09,760
arriving app to make acme has tried to

547
00:20:07,760 --> 00:20:11,200
build the scaffolding in the data to

548
00:20:09,760 --> 00:20:13,360
make it easier to connect with other

549
00:20:11,200 --> 00:20:15,360
public and private data sets to make the

550
00:20:13,360 --> 00:20:16,799
data we hold more useful but understand

551
00:20:15,360 --> 00:20:18,799
that this is not in the general

552
00:20:16,799 --> 00:20:21,200
operations of the museum so needs

553
00:20:18,799 --> 00:20:22,080
automation and tolerance for error for

554
00:20:21,200 --> 00:20:23,760
example

555
00:20:22,080 --> 00:20:25,080
uh when we're pulling in metadata from

556
00:20:23,760 --> 00:20:28,159
imdb or

557
00:20:25,080 --> 00:20:30,080
tdtmdb or wikidata ids

558
00:20:28,159 --> 00:20:32,960
that's not part of acme's cataloging

559
00:20:30,080 --> 00:20:34,640
process but added later so they're

560
00:20:32,960 --> 00:20:37,679
essential for making acne's data to be

561
00:20:34,640 --> 00:20:39,200
able to be work with um but sometimes

562
00:20:37,679 --> 00:20:41,600
there are some errors in that auto

563
00:20:39,200 --> 00:20:41,600
linking

564
00:20:42,320 --> 00:20:46,400
um data creation is slow so even though

565
00:20:44,880 --> 00:20:49,280
we've got

566
00:20:46,400 --> 00:20:52,480
an automated uh transcoding setup at

567
00:20:49,280 --> 00:20:55,120
acme now um digitizing a film from the

568
00:20:52,480 --> 00:20:59,039
analog film takes about four times the

569
00:20:55,120 --> 00:21:02,799
runtime of that footage in star flavor

570
00:20:59,039 --> 00:21:04,640
um so seb also writes on that cataloging

571
00:21:02,799 --> 00:21:06,960
a film takes significant research time

572
00:21:04,640 --> 00:21:08,559
especially if the film is unique to us

573
00:21:06,960 --> 00:21:10,400
along with some or all of the people

574
00:21:08,559 --> 00:21:12,720
involved are featured in the film

575
00:21:10,400 --> 00:21:14,480
because data creation is slow we also

576
00:21:12,720 --> 00:21:16,240
have records that are very old and

577
00:21:14,480 --> 00:21:17,600
featured feature language that was

578
00:21:16,240 --> 00:21:19,360
common

579
00:21:17,600 --> 00:21:20,799
in common use in previous decades that

580
00:21:19,360 --> 00:21:23,039
may be considered offensive or

581
00:21:20,799 --> 00:21:24,960
inappropriate in modern time remember

582
00:21:23,039 --> 00:21:27,440
that acme was the state film archive all

583
00:21:24,960 --> 00:21:28,720
the way back to the 1940s this means

584
00:21:27,440 --> 00:21:30,880
that there is

585
00:21:28,720 --> 00:21:33,120
internal concern about what previous

586
00:21:30,880 --> 00:21:35,280
language might say about us now and this

587
00:21:33,120 --> 00:21:37,679
creates more work to fix the data these

588
00:21:35,280 --> 00:21:40,720
are human cultural issues that manifest

589
00:21:37,679 --> 00:21:40,720
in risk aversion

590
00:21:41,440 --> 00:21:46,000
another challenge was just the sheer

591
00:21:43,200 --> 00:21:48,799
update time so updating 45 000 records

592
00:21:46,000 --> 00:21:50,960
and doing some uh changes to them takes

593
00:21:48,799 --> 00:21:54,000
about four hours so we run that update

594
00:21:50,960 --> 00:21:56,640
overnight which is great um but if we

595
00:21:54,000 --> 00:21:58,159
need to do any live fixes then yeah it

596
00:21:56,640 --> 00:22:00,080
takes about four hours so we've got a

597
00:21:58,159 --> 00:22:02,320
bit of a turnaround time there

598
00:22:00,080 --> 00:22:03,600
uh and also the

599
00:22:02,320 --> 00:22:06,080
uh

600
00:22:03,600 --> 00:22:08,480
we get the nightly get diffs on 45 000

601
00:22:06,080 --> 00:22:10,080
files uh sometimes fails on the pods

602
00:22:08,480 --> 00:22:11,360
because they've got a limited amount of

603
00:22:10,080 --> 00:22:14,000
memory so

604
00:22:11,360 --> 00:22:16,159
um we've got a bit of a struggle keeping

605
00:22:14,000 --> 00:22:20,559
the resources really low and then making

606
00:22:16,159 --> 00:22:20,559
sure that the diffs don't kill the pods

607
00:22:20,720 --> 00:22:24,320
uh thank you um i'd also like to

608
00:22:22,720 --> 00:22:26,000
acknowledge the excellent work done by

609
00:22:24,320 --> 00:22:27,679
our original team who built the private

610
00:22:26,000 --> 00:22:30,640
xos apis so

611
00:22:27,679 --> 00:22:34,159
uh ali haberfield andrew sarong benjamin

612
00:22:30,640 --> 00:22:37,600
ladd david amores uh greg turner the cto

613
00:22:34,159 --> 00:22:39,200
at the time uh pipshi uh and myself and

614
00:22:37,600 --> 00:22:41,120
also want to acknowledge our new

615
00:22:39,200 --> 00:22:43,280
developer sam mayer for his help through

616
00:22:41,120 --> 00:22:45,840
this process which has been excellent

617
00:22:43,280 --> 00:22:48,240
um also a shout out to our bosses so sub

618
00:22:45,840 --> 00:22:51,600
chan magazine guy lucy patterson

619
00:22:48,240 --> 00:22:53,840
brilliant uh francesco romini uh who's

620
00:22:51,600 --> 00:22:55,520
my direct boss uh who puts up with me a

621
00:22:53,840 --> 00:22:57,600
lot which is great

622
00:22:55,520 --> 00:22:59,799
um if you'd like to see the slides

623
00:22:57,600 --> 00:23:01,440
there's a link to them there so s i g

624
00:22:59,799 --> 00:23:02,640
hmon.com

625
00:23:01,440 --> 00:23:03,840
acme

626
00:23:02,640 --> 00:23:08,280
um

627
00:23:03,840 --> 00:23:08,280
and that's it any questions

628
00:23:08,400 --> 00:23:12,240
thank you simon we've got a few

629
00:23:10,559 --> 00:23:15,840
questions for you

630
00:23:12,240 --> 00:23:18,320
um are you asking users to sign in or up

631
00:23:15,840 --> 00:23:20,640
uh so you can get some usage stats at

632
00:23:18,320 --> 00:23:25,200
all or are you just providing http

633
00:23:20,640 --> 00:23:26,480
access no it's totally open so um we

634
00:23:25,200 --> 00:23:28,880
thought about

635
00:23:26,480 --> 00:23:31,039
doing api keys and that sort of thing

636
00:23:28,880 --> 00:23:33,360
um so at the moment we've got absolutely

637
00:23:31,039 --> 00:23:35,520
no analytics on people using it other

638
00:23:33,360 --> 00:23:37,360
than how much cpu load

639
00:23:35,520 --> 00:23:39,520
is on our pods at the moment so we'll

640
00:23:37,360 --> 00:23:42,880
keep it like that until we need to worry

641
00:23:39,520 --> 00:23:45,679
about costs or anything

642
00:23:42,880 --> 00:23:47,520
um is there a skew to the data that is

643
00:23:45,679 --> 00:23:49,919
being looked up anything interesting or

644
00:23:47,520 --> 00:23:52,400
surprising or any new apis you've been

645
00:23:49,919 --> 00:23:52,400
asked for

646
00:23:52,480 --> 00:23:56,799
um we haven't actually had time to look

647
00:23:54,640 --> 00:23:59,039
through the data ourselves yet so we've

648
00:23:56,799 --> 00:24:00,880
just been building this thing um so i

649
00:23:59,039 --> 00:24:03,039
suspect we

650
00:24:00,880 --> 00:24:05,120
as we work with different universities

651
00:24:03,039 --> 00:24:07,200
who are going to do projects with it um

652
00:24:05,120 --> 00:24:09,279
we'll find out a lot of cool things

653
00:24:07,200 --> 00:24:10,240
um sorry what was the second after the

654
00:24:09,279 --> 00:24:12,799
question

655
00:24:10,240 --> 00:24:15,039
um is there any any new apis you've been

656
00:24:12,799 --> 00:24:17,039
asked for

657
00:24:15,039 --> 00:24:20,000
nothing we've been asked for yet but i

658
00:24:17,039 --> 00:24:21,039
think um our creators api and our labels

659
00:24:20,000 --> 00:24:22,960
api

660
00:24:21,039 --> 00:24:24,480
are probably the logical next ones but

661
00:24:22,960 --> 00:24:26,720
yeah if anyone else wants any of our

662
00:24:24,480 --> 00:24:28,960
other private apis i mentioned opened up

663
00:24:26,720 --> 00:24:30,320
then definitely email uh because that

664
00:24:28,960 --> 00:24:32,320
always helps me

665
00:24:30,320 --> 00:24:34,320
get the bosses to prioritize it so it's

666
00:24:32,320 --> 00:24:35,600
gonna be really useful

667
00:24:34,320 --> 00:24:37,440
um

668
00:24:35,600 --> 00:24:39,919
another question here i'm curious as to

669
00:24:37,440 --> 00:24:41,520
how you support and build relationships

670
00:24:39,919 --> 00:24:43,679
with users how do you approach

671
00:24:41,520 --> 00:24:47,840
documentation community and building

672
00:24:43,679 --> 00:24:50,480
awareness of your data availability

673
00:24:47,840 --> 00:24:54,640
um yeah so we've got uptime monitors to

674
00:24:50,480 --> 00:24:55,600
make sure our api is um up and um

675
00:24:54,640 --> 00:24:56,880
available

676
00:24:55,600 --> 00:24:58,559
but seeing it's brand new we really

677
00:24:56,880 --> 00:25:00,640
haven't

678
00:24:58,559 --> 00:25:03,520
had that many problems yet so if you can

679
00:25:00,640 --> 00:25:04,880
crash it that'd be great and we will uh

680
00:25:03,520 --> 00:25:06,720
please get in touch if you do and tell

681
00:25:04,880 --> 00:25:08,080
us how you crushed it so we can we can

682
00:25:06,720 --> 00:25:12,559
make it more stable

683
00:25:08,080 --> 00:25:12,559
uh in terms of building community um

684
00:25:12,640 --> 00:25:16,799
that's a good one uh giving talks like

685
00:25:14,480 --> 00:25:19,600
this um we're hoping to connect with

686
00:25:16,799 --> 00:25:20,640
other organizations um the state library

687
00:25:19,600 --> 00:25:22,320
and melbourne are doing some really

688
00:25:20,640 --> 00:25:25,360
interesting things with machine learning

689
00:25:22,320 --> 00:25:28,000
at the moment and auto categorization so

690
00:25:25,360 --> 00:25:29,600
it'll be really great to work with

691
00:25:28,000 --> 00:25:30,640
people who are doing

692
00:25:29,600 --> 00:25:32,240
tagging

693
00:25:30,640 --> 00:25:34,000
and categorization

694
00:25:32,240 --> 00:25:36,159
because that's what we need help with in

695
00:25:34,000 --> 00:25:38,480
our collection

696
00:25:36,159 --> 00:25:40,480
um have you had any data sovereignty

697
00:25:38,480 --> 00:25:42,640
issues being raised around the use of

698
00:25:40,480 --> 00:25:43,679
commercial cloud services like aws or

699
00:25:42,640 --> 00:25:45,200
google

700
00:25:43,679 --> 00:25:47,919
as you've worked through getting your

701
00:25:45,200 --> 00:25:49,760
public api up and running

702
00:25:47,919 --> 00:25:52,640
uh i haven't yet that's probably a good

703
00:25:49,760 --> 00:25:54,000
one for seb chan to answer so um yeah

704
00:25:52,640 --> 00:25:56,400
ping him

705
00:25:54,000 --> 00:25:59,760
chan on twitter and ask any questions

706
00:25:56,400 --> 00:26:02,240
like that he's he's a good one for those

707
00:25:59,760 --> 00:26:04,000
wonderful and one last question have you

708
00:26:02,240 --> 00:26:06,799
had any issues with lawyers not

709
00:26:04,000 --> 00:26:08,880
understanding floss software licenses

710
00:26:06,799 --> 00:26:11,520
i.e coming from all software must be

711
00:26:08,880 --> 00:26:13,840
licensed or else it's stolen model

712
00:26:11,520 --> 00:26:15,840
or insisting that a signed license or

713
00:26:13,840 --> 00:26:17,919
non-disclosure agreements

714
00:26:15,840 --> 00:26:19,520
but then adding exceptions

715
00:26:17,919 --> 00:26:21,919
that sort of thing

716
00:26:19,520 --> 00:26:23,360
uh we haven't yet touched wood so i'm

717
00:26:21,919 --> 00:26:24,960
sure

718
00:26:23,360 --> 00:26:27,760
down the track

719
00:26:24,960 --> 00:26:30,559
as we try and release images and videos

720
00:26:27,760 --> 00:26:32,000
and a lot of other external content then

721
00:26:30,559 --> 00:26:33,679
um yeah we're going to run into a lot

722
00:26:32,000 --> 00:26:35,600
but i think we took a pretty cautious

723
00:26:33,679 --> 00:26:38,240
approach with this first

724
00:26:35,600 --> 00:26:39,760
dump of data where it's all just stuff

725
00:26:38,240 --> 00:26:42,240
that we've created

726
00:26:39,760 --> 00:26:42,240
ourselves

727
00:26:42,400 --> 00:26:46,880
wonderful well thank you very much um

728
00:26:45,000 --> 00:26:48,880
[Music]

729
00:26:46,880 --> 00:26:50,799
oh sorry i keep going you said but you

730
00:26:48,880 --> 00:26:53,520
keep mentioning said that's probably why

731
00:26:50,799 --> 00:26:56,000
um simon will be available in the go

732
00:26:53,520 --> 00:26:58,080
glam chat functionality which is to the

733
00:26:56,000 --> 00:27:00,400
left hand side of your screen but we

734
00:26:58,080 --> 00:27:02,960
will be moving into

735
00:27:00,400 --> 00:27:05,360
katherine's session which is at 2 p.m in

736
00:27:02,960 --> 00:27:08,799
the video chat session called go glam

737
00:27:05,360 --> 00:27:11,919
dash shovel ready glamor graduates

738
00:27:08,799 --> 00:27:13,440
so i'll see you all there at uh 2pm

739
00:27:11,919 --> 00:27:15,279
thank you simon

740
00:27:13,440 --> 00:27:18,279
thanks so much

741
00:27:15,279 --> 00:27:18,279
thanks