1
00:00:00,080 --> 00:00:02,680
If you're interested in graph 
databases and curious about 

2
00:00:02,680 --> 00:00:06,040
knowledge graphs and how you can
use both within generative AI, 

3
00:00:06,320 --> 00:00:09,080
this episode is for you. 
Joining me today, I have two 

4
00:00:09,080 --> 00:00:11,800
guests, Bosco Brookmeyer, head 
of engineering over at Every 

5
00:00:11,800 --> 00:00:15,160
Cure and Neil Seyoung, product 
manager over at Neo 4 J. 

6
00:00:15,600 --> 00:00:18,160
We go through the basics. 
What are graph databases? 

7
00:00:18,160 --> 00:00:20,880
What are knowledge graphs, and 
what problems do they solve? 

8
00:00:21,280 --> 00:00:23,280
You might actually have more 
graph problems than you think, 

9
00:00:23,720 --> 00:00:29,160
so enjoy. 
Do you have any experience with 

10
00:00:29,160 --> 00:00:31,960
regards to like rack solutions 
or implementing stuff in 

11
00:00:31,960 --> 00:00:33,680
production with regards to Chen 
AI? 

12
00:00:33,960 --> 00:00:36,240
Oh, it's difficult. 
I mean, that's the reality of it

13
00:00:36,240 --> 00:00:38,920
getting to production is is I 
think currently the biggest 

14
00:00:38,920 --> 00:00:41,800
challenge with Chen AI. 
Yeah, yeah, I think it's quite 

15
00:00:41,800 --> 00:00:44,640
easy to set something up. 
But once you start work, 

16
00:00:44,640 --> 00:00:47,680
especially like when you're I'm 
I'm living in a graph world, 

17
00:00:47,680 --> 00:00:50,080
right. 
When you're converting documents

18
00:00:50,080 --> 00:00:53,280
to a graph, that's something 
that's quite easy to to do a 

19
00:00:53,280 --> 00:00:55,360
first demo, but you're going to 
have to start thinking about 

20
00:00:55,360 --> 00:00:57,360
chunking. 
You have to kind of model, you 

21
00:00:57,360 --> 00:01:00,040
have to do so many things to get
to something that's production 

22
00:01:00,040 --> 00:01:04,040
worthy, and scaling to 10s of 
millions of documents is where 

23
00:01:04,080 --> 00:01:05,720
the challenge is. 
What do you mean when you say 

24
00:01:05,760 --> 00:01:07,880
OK, we're going from a document 
to a graph? 

25
00:01:07,920 --> 00:01:09,760
Like visually, I don't know what
that means. 

26
00:01:09,840 --> 00:01:12,040
Yeah. 
So I mean, you can probably 

27
00:01:12,040 --> 00:01:15,240
speak about this as well, but 
when you're in my world, when 

28
00:01:15,240 --> 00:01:18,600
you're building a knowledge 
graph from unstructured data, 

29
00:01:19,200 --> 00:01:21,800
you're taking a document, you're
converting that into a knowledge

30
00:01:21,800 --> 00:01:23,720
graph, which means that you have
to define a model. 

31
00:01:24,000 --> 00:01:26,240
You take the model which has 
your entities in it. 

32
00:01:26,320 --> 00:01:29,840
For example, you work in a bank.
Your entities would be the bank 

33
00:01:29,840 --> 00:01:33,360
itself, a client, an account. 
These will be your notes in the 

34
00:01:33,360 --> 00:01:35,560
graph. 
And then you use your large 

35
00:01:35,560 --> 00:01:38,720
language model to extract these 
entities and convert that into a

36
00:01:39,000 --> 00:01:40,640
knowledge graph. 
So you can use a large language 

37
00:01:40,640 --> 00:01:43,840
model to extract that. 
So this is something that is 

38
00:01:43,840 --> 00:01:46,160
kind of crucial when you want to
do graph Rag, which I'm sure 

39
00:01:46,160 --> 00:01:49,720
we'll discuss later, to build 
the right knowledge graph. 

40
00:01:49,840 --> 00:01:52,360
And that's where most people 
kind of struggle. 

41
00:01:52,760 --> 00:01:54,680
Maybe you can say something 
about that, because I know 

42
00:01:54,680 --> 00:01:57,560
you've had a lot of very 
ambitious ideas to construct 

43
00:01:57,560 --> 00:02:00,720
knowledge graphs from from data.
Yeah. 

44
00:02:00,720 --> 00:02:06,240
So I mean, my, my team right 
now, we mostly grab existing 

45
00:02:06,240 --> 00:02:08,840
knowledge graphs and they're 
based on certain ontologies and 

46
00:02:08,840 --> 00:02:12,960
we merge all of them together to
try and create one very large 

47
00:02:12,960 --> 00:02:14,880
set. 
Because then you can still 

48
00:02:14,880 --> 00:02:18,280
introduce filtering steps 
between your modeling and the 

49
00:02:18,280 --> 00:02:21,680
graph itself where you can say, 
OK, I'm going to slice away 

50
00:02:21,680 --> 00:02:23,000
stuff. 
I'm going to take away certain 

51
00:02:23,000 --> 00:02:25,720
categories, I'm going to take 
away certain kinds of nodes. 

52
00:02:26,320 --> 00:02:27,760
But you can't add afterwards, 
right? 

53
00:02:27,760 --> 00:02:31,520
So we first throw everything 
together and make sure that it's

54
00:02:31,520 --> 00:02:33,000
all harmonized within the 
ontology. 

55
00:02:33,000 --> 00:02:38,040
And then we do data experiments.
So we say, OK, what if we remove

56
00:02:38,040 --> 00:02:40,080
all of the proteins from the 
knowledge graph? 

57
00:02:40,720 --> 00:02:42,960
Does that up? 
Or like, does that lift up our 

58
00:02:42,960 --> 00:02:45,320
performance? 
Actually the experiment, I told 

59
00:02:45,320 --> 00:02:49,720
you on Friday that it was 
running where you perturbate the

60
00:02:49,720 --> 00:02:52,000
edges. 
So I love how we just go 

61
00:02:52,000 --> 00:02:56,760
straight to the graph. 
So the question was what happens

62
00:02:56,760 --> 00:02:59,120
if you mess with the knowledge 
graph? 

63
00:02:59,120 --> 00:03:03,240
Does it actually impact your 
model performance because the 

64
00:03:03,600 --> 00:03:06,760
the hypothesis you want to 
refute or you really want to 

65
00:03:06,760 --> 00:03:09,040
avoid that? 
You can mess the knowledge graph

66
00:03:09,040 --> 00:03:11,480
up completely, but the model 
still performs the same way. 

67
00:03:12,040 --> 00:03:13,880
If you have that, then it means 
you actually didn't. 

68
00:03:14,000 --> 00:03:16,520
You're not deriving information 
from the knowledge graph. 

69
00:03:16,520 --> 00:03:18,800
Your model isn't performing good
because of the structure. 

70
00:03:20,000 --> 00:03:22,880
Your model is performing good 
because it's memorizing answers 

71
00:03:22,880 --> 00:03:26,960
or because it's just in our case
may be latching onto it, saying 

72
00:03:26,960 --> 00:03:30,960
that certain drugs, which we 
call frequent Flyers, just 

73
00:03:30,960 --> 00:03:35,120
always help. 
So if you inject someone with 

74
00:03:35,520 --> 00:03:38,600
adrenaline, that usually 
resolves most symptoms because 

75
00:03:38,800 --> 00:03:40,440
adrenaline just has that impact 
on the human. 

76
00:03:40,440 --> 00:03:42,960
Body shocks your system. 
But that's not necessarily 

77
00:03:42,960 --> 00:03:44,840
something that helps people in 
term. 

78
00:03:45,840 --> 00:03:48,880
And so that could be memorized 
by the model, but So what? 

79
00:03:48,960 --> 00:03:51,840
What we then found out in the 
experiment is if you mess up the

80
00:03:51,840 --> 00:03:56,960
knowledge graph 99%. 
So I took the target nodes. 

81
00:03:56,960 --> 00:03:59,600
So you always have two nodes 
that are being connected with an

82
00:03:59,600 --> 00:04:01,880
edge and they're directional in 
our case. 

83
00:04:01,880 --> 00:04:07,640
So I took them edges that point 
at that object and I changed 

84
00:04:07,640 --> 00:04:09,400
them to another object of the 
same class. 

85
00:04:09,400 --> 00:04:13,000
So I would say drug treats 
disease, but it would be a 

86
00:04:13,000 --> 00:04:15,080
different disease and that 
wouldn't actually be true, 

87
00:04:15,080 --> 00:04:16,839
right? 
It's false information most of 

88
00:04:16,839 --> 00:04:20,680
the time. 
But I wouldn't say drug treats, 

89
00:04:21,079 --> 00:04:23,160
I don't know. 
We have food types. 

90
00:04:23,240 --> 00:04:25,040
We also have like a food type 
category. 

91
00:04:25,320 --> 00:04:27,800
So you wouldn't say drug treats 
hamburger because of course 

92
00:04:27,800 --> 00:04:30,480
that's silly. 
And so we perturbate all of the 

93
00:04:30,480 --> 00:04:34,240
edges within each category, one 
percent, 2050 and 99. 

94
00:04:35,240 --> 00:04:39,240
And what is interesting to see 
is the model performance doesn't

95
00:04:39,240 --> 00:04:44,040
drop to 0 when you perturbate 
99%, it actually drops about 

96
00:04:44,040 --> 00:04:46,560
half. 
So half of our models 

97
00:04:46,560 --> 00:04:49,360
performance is from, at least 
that's my hypothesis. 

98
00:04:49,920 --> 00:04:52,280
It's from the data. 
So it's actually picking up from

99
00:04:52,280 --> 00:04:55,680
signal from the graph. 
But the other half is also not 

100
00:04:55,680 --> 00:04:59,240
from the data. 
So yeah, that was just an 

101
00:04:59,240 --> 00:05:00,440
experiment. 
Yeah, yeah. 

102
00:05:00,880 --> 00:05:04,000
So you cut down your knowledge 
graph and you have a model that 

103
00:05:04,040 --> 00:05:06,280
uses this as a base for 
reasoning. 

104
00:05:06,280 --> 00:05:09,960
Do I understand that correctly? 
It's actually we open sourced 

105
00:05:09,960 --> 00:05:14,040
the whole thing last week. 
So that's, it's all in the in 

106
00:05:14,040 --> 00:05:18,880
the giant mono repo, but we take
the knowledge graph, we then 

107
00:05:18,880 --> 00:05:21,400
create vector embeddings for 
each node. 

108
00:05:22,000 --> 00:05:24,000
And there's a variety of ways 
you can do embeddings. 

109
00:05:24,440 --> 00:05:26,360
At the moment we do node to vet 
because it's the highest 

110
00:05:26,360 --> 00:05:29,040
performing one. 
But you really want to encode 

111
00:05:29,280 --> 00:05:32,400
the neighborhood and kind of the
topology around a node in the 

112
00:05:32,400 --> 00:05:35,840
number that describes this. 
And then you give those numbers 

113
00:05:35,840 --> 00:05:39,800
to a model because models, of 
course, they always need a, a 

114
00:05:39,800 --> 00:05:43,400
numerical representation. 
And then you try to predict what

115
00:05:43,400 --> 00:05:46,160
is the probability that a 
certain drug treats a certain 

116
00:05:46,160 --> 00:05:48,920
disease. 
So I think it's the class of 

117
00:05:48,920 --> 00:05:50,960
edge prediction essentially. 
Yeah. 

118
00:05:51,440 --> 00:05:52,760
You mentioned ontologies as 
well. 

119
00:05:52,760 --> 00:05:54,960
Can you lay out what what do 
ontologies mean? 

120
00:05:55,120 --> 00:05:57,240
Do you want to do that? 
Because I'm not a big Ontology 

121
00:05:57,240 --> 00:05:59,880
fan. 
Yeah, I mean, they're this, this

122
00:05:59,880 --> 00:06:01,720
is a very scary topic because 
people have different 

123
00:06:01,720 --> 00:06:05,280
definitions for an ontology. 
But in my world, an ontology is 

124
00:06:05,280 --> 00:06:08,720
just a model definition, OK? 
It's just a specification, as he

125
00:06:08,720 --> 00:06:10,400
was saying. 
And this drug treats this 

126
00:06:10,400 --> 00:06:13,400
disease, and this disease is 
related to that disease. 

127
00:06:13,400 --> 00:06:16,560
So in the world of property 
crafts, we call it a model. 

128
00:06:16,560 --> 00:06:18,680
In the RDF world, they like to 
use the word ontology. 

129
00:06:18,680 --> 00:06:21,200
I kind of switched them up 
interchangeably. 

130
00:06:21,200 --> 00:06:24,400
Yeah, what's RDF? 
RDF is a oh, that I will 

131
00:06:24,400 --> 00:06:26,000
redirect that to you. 
That's a. 

132
00:06:26,320 --> 00:06:29,280
So they're they're basically two
big philosophy in the craft 

133
00:06:29,280 --> 00:06:33,800
world on on storing craft data. 
Yeah, one camp says property 

134
00:06:33,800 --> 00:06:37,120
graphs are the way to go and the
other side says RDF is is much 

135
00:06:37,120 --> 00:06:39,200
better way to go. 
OK, so it's a way of storing 

136
00:06:39,200 --> 00:06:40,440
your graph data. 
Yes, it is. 

137
00:06:40,480 --> 00:06:41,200
Yeah. 
Gotcha. 

138
00:06:41,280 --> 00:06:42,520
Yeah. 
What does it stand for then? 

139
00:06:43,280 --> 00:06:45,840
Resource description framework 
OK. 

140
00:06:45,880 --> 00:06:47,720
Yeah, just a different way of 
storing it. 

141
00:06:47,880 --> 00:06:49,960
Yeah, yeah, it's, it's all 
graphs, right? 

142
00:06:49,960 --> 00:06:52,360
It's just a different way of 
storing the data. 

143
00:06:52,360 --> 00:06:53,840
It's a different way of querying
the data. 

144
00:06:55,680 --> 00:06:57,440
Yeah, yeah. 
And graph Graph thinking is not 

145
00:06:57,440 --> 00:07:00,040
something I'm used to. 
I was thinking before the show. 

146
00:07:00,040 --> 00:07:03,680
Have I seen any, any graphs in 
real life like you mentioned the

147
00:07:03,680 --> 00:07:07,800
entities and there I can draw a 
relation to kind of DDD when it 

148
00:07:07,800 --> 00:07:10,880
comes to defining your domains 
within a certain field. 

149
00:07:11,760 --> 00:07:15,040
So that was kind of similar. 
I've recently been really 

150
00:07:15,040 --> 00:07:17,800
structuring my notes and I'm 
trying to be a settle custom 

151
00:07:17,800 --> 00:07:20,640
practitioner. 
And there one of my colleagues 

152
00:07:20,640 --> 00:07:22,480
said, OK, don't even think about
the folder structure. 

153
00:07:22,640 --> 00:07:25,480
Just make sure that you have 
references from one document to 

154
00:07:25,480 --> 00:07:26,960
another when they just make 
sense. 

155
00:07:27,600 --> 00:07:28,960
And then I have a graph 
visualization. 

156
00:07:28,960 --> 00:07:31,160
That's the only thing I can 
think of where I think, OK, this

157
00:07:31,160 --> 00:07:32,920
might be. 
I don't know if I would call it 

158
00:07:32,920 --> 00:07:34,600
a knowledge graph or if that's 
comparable. 

159
00:07:35,280 --> 00:07:36,320
Would you say that is? 
Yeah. 

160
00:07:36,320 --> 00:07:39,240
Yeah, what you're describing, I 
have a colleague, he calls it 

161
00:07:39,240 --> 00:07:42,280
the graph problem problem. 
OK, which means you don't know 

162
00:07:42,280 --> 00:07:44,000
that you have a graph problem. 
That's your. 

163
00:07:44,640 --> 00:07:46,400
Problem OK. 
And I love that. 

164
00:07:46,400 --> 00:07:48,640
And I think it's very fun 
because once you start thinking 

165
00:07:48,640 --> 00:07:50,640
about graphs, you start seeing 
everything as a graph. 

166
00:07:50,760 --> 00:07:53,600
OK? 
Like going to Amazon, buying a 

167
00:07:53,800 --> 00:07:57,040
laptop, It says other users who 
bought this laptop also 

168
00:07:57,040 --> 00:07:59,480
purchased a mouse. 
Yeah, that's also a graph. 

169
00:07:59,720 --> 00:08:03,120
OK, You're a node as a person. 
You bought a product that's a 

170
00:08:03,120 --> 00:08:05,480
node, and then there's another 
person that bought that same 

171
00:08:05,480 --> 00:08:07,400
product. 
So you're building this little 

172
00:08:07,400 --> 00:08:10,320
graph of connected entities and 
you're doing a recommendation. 

173
00:08:10,320 --> 00:08:13,080
These two nodes should be 
connected, you and this other 

174
00:08:13,080 --> 00:08:14,760
person, because you purchased 
the same product. 

175
00:08:15,160 --> 00:08:17,800
And then from that other person,
again, next step in the graph, 

176
00:08:18,000 --> 00:08:19,360
what other products did they 
buy? 

177
00:08:19,360 --> 00:08:21,720
A mouse? 
So then you're thinking in terms

178
00:08:21,720 --> 00:08:24,360
of connected entities. 
Do they use a graph under the 

179
00:08:24,360 --> 00:08:26,480
hood then also to recommend 
those things or? 

180
00:08:26,600 --> 00:08:28,680
I'm pretty sure. 
I'm not 100% sure, yeah, but it 

181
00:08:28,680 --> 00:08:31,560
feels like a very natural. 
Graph problem. 

182
00:08:31,560 --> 00:08:32,600
Interesting. 
Yeah, yeah, yeah. 

183
00:08:33,320 --> 00:08:34,960
So that apparently is a graph 
problem. 

184
00:08:35,120 --> 00:08:36,520
Yes. 
I never like, I never think 

185
00:08:36,520 --> 00:08:38,000
about that, what happens under 
the hood. 

186
00:08:38,480 --> 00:08:41,039
But it's very interesting that 
you go from one thing to an 

187
00:08:41,039 --> 00:08:43,120
action to then a person to their
actions. 

188
00:08:43,480 --> 00:08:45,360
And you can kind of follow steps
in that way. 

189
00:08:45,440 --> 00:08:48,760
Yeah, you can traverse those 
paths and you can use that with 

190
00:08:48,760 --> 00:08:50,520
regards to Gen. 
AI for reasoning then. 

191
00:08:51,440 --> 00:08:54,960
Yes, yeah. 
So my philosophy with Gen. 

192
00:08:54,960 --> 00:08:58,720
AI and Speaking of graph, right,
context is, is king. 

193
00:08:59,240 --> 00:09:01,560
The more context, the more 
relevant context you pass 

194
00:09:01,560 --> 00:09:04,320
through a large language model, 
the better it performs. 

195
00:09:04,600 --> 00:09:07,640
And graphs are just extremely 
good at context because you've 

196
00:09:07,640 --> 00:09:09,240
got a network of connected 
things. 

197
00:09:09,240 --> 00:09:12,240
It's a document living attached 
to entities that are connected 

198
00:09:12,240 --> 00:09:14,480
to other entities. 
And you can go very, very deep 

199
00:09:14,480 --> 00:09:16,480
in that graph. 
So you have a very rich context 

200
00:09:16,880 --> 00:09:19,560
of a ontology or a model that 
you've defined. 

201
00:09:20,160 --> 00:09:22,920
And that is just why I think 
graphs and graph rag is is very 

202
00:09:22,920 --> 00:09:24,960
cool for for this kind of thing.
Interesting. 

203
00:09:25,960 --> 00:09:28,560
I really wonder how it compares 
to like Rag. 

204
00:09:28,560 --> 00:09:32,080
In essence, how I've seen it is 
we use text and text is 

205
00:09:32,080 --> 00:09:34,400
interpretable for humans and 
then we make embeddings from 

206
00:09:34,400 --> 00:09:35,920
that and then we do reasoning 
for that. 

207
00:09:36,160 --> 00:09:38,240
It's a different way of kind of 
filtering this context. 

208
00:09:38,240 --> 00:09:41,840
But the way you explain a graph,
if I have an action and then I 

209
00:09:41,840 --> 00:09:44,120
purchase a product and we have 
other people that also purchase 

210
00:09:44,120 --> 00:09:47,120
that product, but also they 
bought different products, it 

211
00:09:47,120 --> 00:09:50,440
very much narrows the context. 
It makes it hyper specific. 

212
00:09:50,880 --> 00:09:53,160
And that might be a good thing. 
I don't know though. 

213
00:09:53,520 --> 00:09:55,280
It's it's, yeah, it's relevant, 
right? 

214
00:09:55,280 --> 00:09:57,240
You're defining. 
These are the relevant things. 

215
00:09:58,960 --> 00:10:01,760
Yeah, have any of you 
experimented with kind of 

216
00:10:01,760 --> 00:10:03,200
embeddings? 
You mentioned node 2 VAC is the 

217
00:10:03,200 --> 00:10:06,160
highest performing one. 
Well, it's just in our specific 

218
00:10:06,360 --> 00:10:10,360
situation, right? 
So we used graph stage before, 

219
00:10:10,480 --> 00:10:16,000
OK, which was a combination of 
we first took the nodes name and

220
00:10:16,000 --> 00:10:18,680
description and category. 
So it's like a certain drug, the

221
00:10:18,680 --> 00:10:25,400
name, maybe a certain 
description from drug DB or 

222
00:10:25,400 --> 00:10:30,000
something. 
And you take all of this and you

223
00:10:30,000 --> 00:10:33,000
throw it at an embedding model, 
the classic, well, not the the 

224
00:10:33,040 --> 00:10:35,840
opening icon, right, the the 
text embeddings. 

225
00:10:36,200 --> 00:10:38,400
And then you have a text 
embedding which describes the 

226
00:10:38,400 --> 00:10:40,120
node in isolation without its 
neighborhood. 

227
00:10:41,080 --> 00:10:45,040
And then graph Sage. 
I'm not going to be able to 

228
00:10:45,040 --> 00:10:46,960
perfectly recall it, but I 
believe it does a bunch of 

229
00:10:46,960 --> 00:10:55,480
random walks and then pertubes 
the the base embedding to also 

230
00:10:55,480 --> 00:11:00,240
encode the neighborhood. 
And so then this number contains

231
00:11:00,240 --> 00:11:03,200
both the information of the node
in isolation. 

232
00:11:04,160 --> 00:11:06,160
And the interesting theory is 
because you're throwing in, 

233
00:11:06,360 --> 00:11:07,960
you're throwing in a couple of 
things now, right? 

234
00:11:07,960 --> 00:11:10,600
You're having language models 
that read the entire Internet. 

235
00:11:10,880 --> 00:11:13,200
So they have very generalized 
knowledge about the world. 

236
00:11:13,760 --> 00:11:17,000
And you take that as a baseline.
And then you kind of manipulate 

237
00:11:17,000 --> 00:11:20,520
that that vector and make it 
point at something slightly 

238
00:11:20,520 --> 00:11:24,320
different given the environment.
And so then the resulting number

239
00:11:24,320 --> 00:11:27,840
is a mix of both the structure 
of your graph and it encodes the

240
00:11:28,520 --> 00:11:30,960
knowledge in general of the 
language model that the 

241
00:11:30,960 --> 00:11:36,520
embedding model was based on. 
That is nice because it's, it in

242
00:11:36,520 --> 00:11:39,000
theory should contain 
information both from your 

243
00:11:39,000 --> 00:11:42,440
knowledge graph, which is very 
codified as well as just general

244
00:11:42,440 --> 00:11:45,320
knowledge. 
But it's also problematic 

245
00:11:45,320 --> 00:11:47,640
because it might leak 
information because how do we 

246
00:11:47,640 --> 00:11:52,840
know that we're, we're, we're 
having a withholding set of, you

247
00:11:52,840 --> 00:11:54,760
know, you do a training and then
you withhold the data and you 

248
00:11:54,800 --> 00:11:57,960
try to predict whether your 
model is able to predict a 

249
00:11:57,960 --> 00:12:00,360
certain edge. 
And how do you know the language

250
00:12:00,360 --> 00:12:05,360
model that you use for the 
embedding doesn't have that 

251
00:12:05,360 --> 00:12:07,080
information stored in its 
weights? 

252
00:12:07,080 --> 00:12:09,600
And that kind of leaks through 
the language model. 

253
00:12:09,680 --> 00:12:14,680
So that's why no to vec is 
probably more pure because you 

254
00:12:14,680 --> 00:12:16,920
use a system that you have 
entire control of. 

255
00:12:17,360 --> 00:12:20,480
You start at a vector that is 
completely randomized and then 

256
00:12:20,480 --> 00:12:24,720
you purely from no to vec you 
get purely on the topology of 

257
00:12:24,720 --> 00:12:28,000
the graph, you get a vector. 
So the problem now is you're 

258
00:12:28,000 --> 00:12:30,760
throwing away this magic power 
from machine that we have 

259
00:12:30,760 --> 00:12:33,720
invented the last couple of 
years transformer models because

260
00:12:33,720 --> 00:12:37,200
we're just using Node 2 AC. 
Yeah. 

261
00:12:38,000 --> 00:12:40,880
I mean, for me like bringing 
Gen. 

262
00:12:40,880 --> 00:12:43,600
AI applications to production 
things that provide value, 

263
00:12:43,600 --> 00:12:46,320
that's already very challenging.
And I feel like what we're 

264
00:12:46,320 --> 00:12:49,200
discussing now with regards to 
the knowledge graphs, it feels 

265
00:12:49,200 --> 00:12:50,960
like it's 2 steps above that 
even. 

266
00:12:50,960 --> 00:12:53,840
When would I use that from a 
from a use case perspective? 

267
00:12:53,840 --> 00:12:56,600
I know you have a good one, but 
was it like the first thing you 

268
00:12:56,600 --> 00:12:59,360
gravitated towards or did you 
evolve into kind of this way of 

269
00:12:59,360 --> 00:13:01,520
thinking? 
Did it just make sense for the 

270
00:13:01,520 --> 00:13:05,720
use case? 
So in my case, my founders 

271
00:13:07,440 --> 00:13:11,520
essentially decided that there's
unique moment in time where we 

272
00:13:11,520 --> 00:13:14,600
have now machine learning models
at our disposal that can connect

273
00:13:14,600 --> 00:13:19,320
all the different information 
and they wanted to figure out, 

274
00:13:19,320 --> 00:13:21,800
OK, all the drugs, all the 
diseases, can we just search for

275
00:13:21,800 --> 00:13:27,320
any repurposing opportunity? 
That is why we use knowledge 

276
00:13:27,320 --> 00:13:31,120
graphs for this I think was just
because at the time that seemed 

277
00:13:31,120 --> 00:13:34,640
like the highest probability to 
be able to actually do a 

278
00:13:34,640 --> 00:13:38,520
prediction all versus all. 
But we're now mixing in a number

279
00:13:38,520 --> 00:13:41,760
of other because other methods 
because we realized actually 

280
00:13:41,760 --> 00:13:44,800
it's it's just a ranking problem
trying to rank everything. 

281
00:13:44,800 --> 00:13:50,000
You can rank things in many 
different ways, but I think when

282
00:13:50,200 --> 00:13:53,840
when you both were talking about
like it's so hard to get things 

283
00:13:53,840 --> 00:13:56,480
into production, I'm actually 
very happy that not everything 

284
00:13:56,480 --> 00:14:02,160
makes it to production because I
have real beef with the 

285
00:14:02,160 --> 00:14:04,240
complexity of large 
corporations. 

286
00:14:05,400 --> 00:14:11,640
I sat in a boardroom couple 
years ago and someone like 

287
00:14:12,120 --> 00:14:16,160
mentioned kind of an off way 
rise, a major insurance company 

288
00:14:17,080 --> 00:14:21,840
with 23,000 IT systems and 
everyone seemed to be completely

289
00:14:21,840 --> 00:14:23,280
fine with it. 
That's ridiculous. 

290
00:14:24,120 --> 00:14:27,040
I mean, how much can you, how 
many IT systems do you need to 

291
00:14:27,040 --> 00:14:29,400
manage? 
I don't know. 

292
00:14:29,400 --> 00:14:33,240
Even if you have 100 different 
forms of insurance, one Postgres

293
00:14:33,240 --> 00:14:36,600
database probably could handle 
all of your customers, all of 

294
00:14:36,600 --> 00:14:40,080
the insurance contracts that you
have, all of your employees. 

295
00:14:40,840 --> 00:14:43,360
Why do you need 20 + 1000 IT 
systems? 

296
00:14:44,120 --> 00:14:46,480
And I think the answer is they 
don't. 

297
00:14:46,480 --> 00:14:49,160
But now they have them and it's 
so hard to get rid of them. 

298
00:14:50,400 --> 00:14:54,040
So what I'm really waiting for 
is getting to the point where 

299
00:14:54,680 --> 00:14:56,840
maybe graphs of the answer 
there, I don't know, where 

300
00:14:56,840 --> 00:15:00,680
models were able to just stare 
at the system as a whole and go.

301
00:15:01,600 --> 00:15:06,560
All of these things we can 
consolidate so that people have 

302
00:15:06,560 --> 00:15:11,680
much less chaos to focus, to 
worry about, and then try to 

303
00:15:12,520 --> 00:15:15,040
instead do this one thing that 
is actually valued to the 

304
00:15:15,040 --> 00:15:17,520
business. 
Yeah, I think we're we're seeing

305
00:15:17,520 --> 00:15:20,720
early steps in that like we have
some some customers, you know 

306
00:15:20,720 --> 00:15:23,720
for J where they're trying to 
consolidate all these different 

307
00:15:23,720 --> 00:15:28,000
data sources into a massive 
enterprise knowledge graph or 

308
00:15:28,040 --> 00:15:30,880
digital twin or whatever you 
want to call it, which is 

309
00:15:30,880 --> 00:15:33,080
currently it's somehow 
automated. 

310
00:15:33,080 --> 00:15:36,240
It's a little bit manual, but 
23,000 systems that's obviously 

311
00:15:36,600 --> 00:15:39,440
far in the future. 
But I, I do think that's a big 

312
00:15:39,440 --> 00:15:41,920
area and I think we're going to 
see that because the, the 

313
00:15:41,920 --> 00:15:44,000
downside is there was 23,000 
systems. 

314
00:15:44,000 --> 00:15:45,760
There will be a lot of people's 
jobs depending on it. 

315
00:15:45,760 --> 00:15:49,960
So they're not going to be very 
eager to participate and get 

316
00:15:49,960 --> 00:15:51,720
their stuff out. 
There I follow the training. 

317
00:15:51,720 --> 00:15:54,320
It was about like architecture 
and architecture patterns from 

318
00:15:54,320 --> 00:15:56,400
from Gregor Hope. 
And he said for every 

319
00:15:56,560 --> 00:16:00,600
dysfunctional organizational 
behaviour, there's a vendor and 

320
00:16:00,600 --> 00:16:02,720
they'll be happy to take your 
money to kind of fix that. 

321
00:16:02,720 --> 00:16:04,520
But the behaviour is still 
dysfunctional. 

322
00:16:04,520 --> 00:16:06,920
So all of a sudden if you have a
process problem and you think, 

323
00:16:06,920 --> 00:16:09,120
OK, I'm going to have a SAS 
solution that's going to solve 

324
00:16:09,120 --> 00:16:12,240
that and it might not solve it 
or it just might alleviate it, 

325
00:16:12,240 --> 00:16:14,400
it's still contributing to this 
cost of ownership. 

326
00:16:15,080 --> 00:16:18,200
I feel like it's very 
challenging to keep things 

327
00:16:18,200 --> 00:16:20,560
simple. 
It's very easy to just buy, 

328
00:16:20,560 --> 00:16:23,640
acquire, to add things and then 
increase your complexity. 

329
00:16:25,160 --> 00:16:26,760
But yeah, it has long term 
consequences. 

330
00:16:26,760 --> 00:16:29,640
And I'm, I'm very not happy 
operating in a complex 

331
00:16:29,640 --> 00:16:31,960
landscape. 
Like I like to be productive. 

332
00:16:31,960 --> 00:16:34,160
I like to be effective. 
I think things should just make 

333
00:16:34,160 --> 00:16:35,840
sense. 
And when there's a lot of 

334
00:16:35,840 --> 00:16:38,200
dysfunctional behaviour and we 
have a lot of applications or 

335
00:16:38,200 --> 00:16:41,280
like tools for that to try and 
solve that, I'm just like, what 

336
00:16:41,280 --> 00:16:43,960
are we doing here? 
But still it grows like that. 

337
00:16:44,160 --> 00:16:46,680
It reminds me of that that 
thing, the programming thing, 

338
00:16:46,680 --> 00:16:49,520
where it's super easy to have 
200 lines of code, but to remove

339
00:16:49,520 --> 00:16:51,600
5 lines of code. 
No one wants to do. 

340
00:16:51,680 --> 00:16:54,560
That that feels like the the 
coding equivalent of what you're

341
00:16:54,560 --> 00:16:56,280
just saying. 
Yeah, yeah, yeah, you get that. 

342
00:16:56,560 --> 00:16:58,400
Yeah. 
I still don't know if I'm then 

343
00:16:58,400 --> 00:17:02,200
in an organization and I have an
actual use case where to go with

344
00:17:02,200 --> 00:17:03,720
that. 
What do I do with my data? 

345
00:17:03,720 --> 00:17:06,040
Do I actually need a craft 
database or knowledge? 

346
00:17:06,040 --> 00:17:09,200
Like you already said, I live in
this graph world and I can see 

347
00:17:09,200 --> 00:17:12,200
graph problems. 
What problems do you have, other

348
00:17:12,200 --> 00:17:14,200
than the Amazon example for 
example, that you've solved? 

349
00:17:14,480 --> 00:17:19,359
I think there's a lot out there 
and I'm being surprised still to

350
00:17:19,359 --> 00:17:22,000
this day. 
Unfortunately, I can't tell you 

351
00:17:22,000 --> 00:17:23,880
about all that. 
Of course. 

352
00:17:24,359 --> 00:17:28,960
I mean, there's a very obvious 
ones like real planning, right? 

353
00:17:29,080 --> 00:17:31,320
You're taking a route from a 
station to another station, the 

354
00:17:31,320 --> 00:17:33,200
extras algorithm, finding 
insurance path, 

355
00:17:33,960 --> 00:17:37,480
telecommunications industry, 
building a network of cables to 

356
00:17:37,480 --> 00:17:41,120
provide everyone fast Internet. 
That's a graph controlling the 

357
00:17:41,120 --> 00:17:44,400
load, understanding. 
If one link in my network goes 

358
00:17:44,400 --> 00:17:46,720
out, you know what, what will be
impacted? 

359
00:17:46,720 --> 00:17:49,120
Where do I have to reroute? 
You can translate. 

360
00:17:49,120 --> 00:17:52,920
You can take that idea and copy 
that over to logistics problems 

361
00:17:52,920 --> 00:17:55,960
as well, right? 
There's a boat stuck in the Suez

362
00:17:55,960 --> 00:17:58,080
Canal. 
How do I, how do I get around 

363
00:17:58,080 --> 00:18:03,040
and, and get my stuff out there?
But I think that the really 

364
00:18:03,040 --> 00:18:06,080
interesting ones are the ones 
that are not so obviously craft 

365
00:18:06,080 --> 00:18:08,520
related. 
I think what I was just saying, 

366
00:18:08,520 --> 00:18:10,080
the Amazon example is one of 
them, right? 

367
00:18:10,280 --> 00:18:12,680
You have people, you have 
products. 

368
00:18:13,200 --> 00:18:14,800
You can think of any 
organization really. 

369
00:18:14,800 --> 00:18:16,480
You have teams with people in 
them. 

370
00:18:16,960 --> 00:18:20,360
The people have are working on a
on a tool. 

371
00:18:20,480 --> 00:18:22,640
They have certain set of skills 
or expertises. 

372
00:18:22,640 --> 00:18:25,280
You can imagine a skill is a 
node in the graph, team is a 

373
00:18:25,280 --> 00:18:27,720
node in the graphs. 
Your team needs a Cotland 

374
00:18:27,720 --> 00:18:28,840
expert. 
We just discussed it. 

375
00:18:29,480 --> 00:18:32,000
You can look at your enterprise 
graph and see, OK, there's 

376
00:18:32,000 --> 00:18:34,680
someone working on the other 
side of the org that's a Cotland

377
00:18:34,680 --> 00:18:36,600
expert. 
Maybe I can bring that person 

378
00:18:36,600 --> 00:18:40,360
in, identify skill gaps in 
teams, and then you're starting 

379
00:18:40,360 --> 00:18:43,200
to think, OK, yeah, everything 
in my organization is a graph. 

380
00:18:43,440 --> 00:18:45,160
I can combine that in different 
ways. 

381
00:18:45,160 --> 00:18:48,400
I can and keep expanding that as
well. 

382
00:18:48,600 --> 00:18:49,200
Right. 
Interesting. 

383
00:18:49,280 --> 00:18:52,280
Yeah, Yeah, I mean, the 
reasoning part that for me is 

384
00:18:52,280 --> 00:18:54,560
most interesting, right? 
Because if I have something like

385
00:18:54,560 --> 00:18:58,280
that that depicts kind of in a 
graph way what my organization 

386
00:18:58,280 --> 00:19:00,240
structure looks like, and 
indeed, if I have a question or 

387
00:19:00,240 --> 00:19:03,400
a query and who do I need or do 
I know a person that is related 

388
00:19:03,400 --> 00:19:06,200
to any people in my network, 
that might be really helpful. 

389
00:19:06,520 --> 00:19:10,040
Especially nowadays where 
getting a job to stay, it's more

390
00:19:10,040 --> 00:19:12,040
and more difficult, like 
relationships are I think 

391
00:19:12,400 --> 00:19:15,040
something you build. 
Upon I think in consulting, 

392
00:19:15,520 --> 00:19:19,920
almost any consulting company 
should have at least a program 

393
00:19:19,920 --> 00:19:21,320
in place. 
Whether they try to solve that 

394
00:19:21,320 --> 00:19:26,000
with graphs is one question, but
how can we bring our best 

395
00:19:26,000 --> 00:19:29,400
possible person to this specific
line? 

396
00:19:31,880 --> 00:19:36,800
And I think, I mean, I've worked
in three consultancies now. 

397
00:19:37,240 --> 00:19:39,600
It's always very organic and 
human organized. 

398
00:19:39,600 --> 00:19:42,440
You've got staffers and staffers
have their network and then they

399
00:19:42,440 --> 00:19:45,640
try and think maybe they call 
some some partners up or some 

400
00:19:45,640 --> 00:19:49,840
colleagues in different offices.
And sure, you've got some 

401
00:19:49,840 --> 00:19:53,640
keyword based search systems 
where maybe you can you're 

402
00:19:53,640 --> 00:19:55,400
looking for, like you said, a 
Kotlin expert, right? 

403
00:19:55,400 --> 00:19:58,040
So then you search for Kotlin 
experts in in a consultancy in 

404
00:19:58,040 --> 00:20:02,560
your internal directory. 
What if the person didn't label 

405
00:20:02,560 --> 00:20:08,280
themselves with Kotlin because 
no one likes constantly updating

406
00:20:08,280 --> 00:20:09,760
their labels for themselves. 
It's kind of. 

407
00:20:10,000 --> 00:20:12,240
And the ones that are really 
good definitely don't need to 

408
00:20:12,240 --> 00:20:14,120
because they're constantly 
getting one project after 

409
00:20:14,120 --> 00:20:17,120
another. 
So having some form of a 

410
00:20:17,760 --> 00:20:21,960
representation of what's 
everyone's expertise and what is

411
00:20:21,960 --> 00:20:25,240
the most knowledgeable person. 
That could answer this question.

412
00:20:25,520 --> 00:20:27,880
And then probably that person's 
going to be completely swamped 

413
00:20:27,880 --> 00:20:29,600
because if you're the most 
knowledgeable person about a 

414
00:20:29,600 --> 00:20:31,200
popular topic, everyone's going 
to come to you. 

415
00:20:31,560 --> 00:20:35,160
What are the people around that 
person that still have capacity 

416
00:20:35,160 --> 00:20:38,560
and they're probably good enough
for the job to do some form of a

417
00:20:39,280 --> 00:20:42,640
load balancing across talents 
and also kind of nurture that. 

418
00:20:43,600 --> 00:20:46,200
That all happens in 
organizations quite organically,

419
00:20:46,800 --> 00:20:51,440
but it's certainly not optimized
to the sense of what is a near 

420
00:20:51,440 --> 00:20:53,760
optimal way of leveraging our 
resources. 

421
00:20:53,800 --> 00:20:56,080
Yeah, yeah. 
I like that use case a lot. 

422
00:20:56,280 --> 00:20:59,240
For me, like the tagging 
principle that you mentioned, it

423
00:20:59,240 --> 00:21:01,080
really hits on because we have 
exactly that. 

424
00:21:01,320 --> 00:21:04,680
We have a, a specific board on 
Monday and you're supposed to 

425
00:21:04,680 --> 00:21:06,560
tag like what technologies 
you're good at, what are you 

426
00:21:06,560 --> 00:21:08,040
comfortable with? 
It's going to help sales and 

427
00:21:08,040 --> 00:21:10,280
it's going to help your future 
assignment. 

428
00:21:10,760 --> 00:21:12,600
So if you want a great 
assignment, then you kind of 

429
00:21:12,600 --> 00:21:15,280
work to update that part of the 
tagging. 

430
00:21:15,760 --> 00:21:18,080
But wouldn't it be nice if do 
you guys use Slack? 

431
00:21:18,200 --> 00:21:21,600
Yeah, is something just 
constantly watches all the Slack

432
00:21:21,600 --> 00:21:25,200
channels and sees you talk about
certain topics and others 

433
00:21:25,200 --> 00:21:27,400
appreciating. 
And you can read from the way 

434
00:21:27,400 --> 00:21:30,920
that how many people respond, 
how many people react to it, And

435
00:21:30,920 --> 00:21:32,760
then pulling that all into a 
graph. 

436
00:21:32,880 --> 00:21:35,400
I guess you could do a lexical 
graph where you see every post 

437
00:21:35,400 --> 00:21:38,480
as a node, you're attached to 
the node as an author, so that's

438
00:21:38,480 --> 00:21:41,280
your relationship. 
And then how many people refer 

439
00:21:41,280 --> 00:21:43,280
to that. 
You can now pull out of the 

440
00:21:43,280 --> 00:21:47,440
graph structure the facts of 
you're going to be a central 

441
00:21:47,440 --> 00:21:50,720
node for a specific topic and so
you could find the experts 

442
00:21:50,720 --> 00:21:53,000
through that and you don't have 
to manually label. 

443
00:21:53,160 --> 00:21:55,160
Yeah. 
I like what you said about this,

444
00:21:55,280 --> 00:21:58,000
you you'd never want to get the 
ultimate expert because they're 

445
00:21:58,000 --> 00:22:00,200
going to be busy. 
But having a kind of community 

446
00:22:00,200 --> 00:22:02,680
again, I'm thinking in graph 
algorithms, bring a community 

447
00:22:02,680 --> 00:22:07,240
detection algorithm to find, you
know, Pascal's the expert on X, 

448
00:22:07,520 --> 00:22:09,160
but these are the people that 
work with him. 

449
00:22:09,160 --> 00:22:12,320
So by definition, they might 
also be the expert on Kotlin or 

450
00:22:12,360 --> 00:22:15,080
whatever. 
I think that's very interesting.

451
00:22:15,080 --> 00:22:18,160
And then then you start using 
data that's not super apparent. 

452
00:22:18,160 --> 00:22:20,800
So these tags, but you're 
starting to look at graph 

453
00:22:20,800 --> 00:22:23,080
patterns as part of your feature
set. 

454
00:22:23,160 --> 00:22:24,360
Yeah. 
Now you get that. 

455
00:22:24,960 --> 00:22:28,600
I mean, for me, like information
right now in the landscape that 

456
00:22:28,600 --> 00:22:31,320
I'm in is distributed because we
don't just use Slack, we have 

457
00:22:31,320 --> 00:22:33,400
Slack. 
Some organizations or some parts

458
00:22:33,400 --> 00:22:35,920
of the organizations use Teams. 
We have emails. 

459
00:22:35,920 --> 00:22:38,040
So I'll read there. 
The information is scattered, 

460
00:22:38,400 --> 00:22:40,760
but if you have indeed an 
automated way of already 

461
00:22:40,760 --> 00:22:42,760
plugging into Slack, that's 
incredibly valuable. 

462
00:22:43,400 --> 00:22:45,600
I know what's stopping you from 
also plugging in the other data 

463
00:22:45,600 --> 00:22:48,840
sources I feel like. 
So this was an interesting 1 

464
00:22:48,840 --> 00:22:53,920
during the classical RAG days 
where everyone wanted to stuff 

465
00:22:53,920 --> 00:22:57,000
things into a vector database. 
My hypothesis to this day still 

466
00:22:57,000 --> 00:23:00,000
is because most engineers who 
were put like made in charge of 

467
00:23:00,000 --> 00:23:02,800
a certain project and were 
vector database. 

468
00:23:02,840 --> 00:23:04,200
I didn't even know these things 
exist. 

469
00:23:04,200 --> 00:23:05,640
I want to build 1. 
I want to deploy one. 

470
00:23:06,880 --> 00:23:10,920
But so everything was trying to 
suck information into the vector

471
00:23:10,920 --> 00:23:12,760
representation and then copy 
stuff. 

472
00:23:12,760 --> 00:23:14,680
And of course, then you've got 
the outdated problem, you've got

473
00:23:14,680 --> 00:23:18,520
the access problem. 
Because if I take your direct 

474
00:23:18,520 --> 00:23:20,960
messages between the two of you 
and I chunk them, embed them, 

475
00:23:20,960 --> 00:23:24,360
put them in the knowledge graph,
how do I now make sure that that

476
00:23:24,360 --> 00:23:26,200
result isn't served to someone 
else, right? 

477
00:23:26,200 --> 00:23:28,160
So I need to attach metadata who
can read it. 

478
00:23:28,520 --> 00:23:31,680
It's a huge pain, I guess. 
Now I think the term that's 

479
00:23:31,840 --> 00:23:35,280
emerging is a gentic rags. 
So you have an agent search 

480
00:23:35,320 --> 00:23:38,520
through search AP is so it 
calls, it calls the Slack search

481
00:23:38,520 --> 00:23:41,280
API, it calls the team API, it 
calls SharePoint and Google 

482
00:23:41,280 --> 00:23:44,040
Drive and whatever. 
And then those systems return 

483
00:23:44,040 --> 00:23:48,080
them the most likely results. 
And then the agent kind of, 

484
00:23:48,280 --> 00:23:50,160
which is the same as what a 
normal human would do, right? 

485
00:23:50,160 --> 00:23:51,720
You kind of search for all the 
different systems. 

486
00:23:54,440 --> 00:23:58,640
But I was I was listening to a 
presentation from I think, I 

487
00:23:58,640 --> 00:24:01,080
don't know if it was the founder
of Exa, but there's a company 

488
00:24:01,080 --> 00:24:05,080
called Exa and they're calling 
their way of searching now 

489
00:24:05,080 --> 00:24:10,320
neural search, where they argue 
that can you reinvent search in 

490
00:24:10,320 --> 00:24:14,600
the age of neural networks where
yes, they do embed everything. 

491
00:24:14,600 --> 00:24:16,640
They do that for web-based 
search. 

492
00:24:18,240 --> 00:24:23,160
But their argument is that they 
that way actually out compete 

493
00:24:23,160 --> 00:24:25,640
companies like Google, which do 
keyword based search. 

494
00:24:26,360 --> 00:24:29,160
And so I think it's a really 
interesting feel to see what is 

495
00:24:29,160 --> 00:24:31,160
going to ultimately when is it 
going to be a hybrid? 

496
00:24:31,320 --> 00:24:33,800
Probably because everything's 
usually been mixed things. 

497
00:24:34,360 --> 00:24:38,240
Is it keyword search? 
Is it semantic vector embedding 

498
00:24:38,240 --> 00:24:41,680
based search? 
Is it an agent or some kind of 

499
00:24:41,680 --> 00:24:45,280
system that just calls a bunch 
of AP is and looks through the 

500
00:24:45,280 --> 00:24:48,080
different systems? 
Or can you really in the 

501
00:24:48,080 --> 00:24:53,080
Valhalla I guess is having this 
like 1 unified knowledge 

502
00:24:53,880 --> 00:24:58,400
structure of your company? 
And I know a graph companies say

503
00:24:58,400 --> 00:25:00,880
like this is what you, but I 
don't know if we can ever reach 

504
00:25:00,880 --> 00:25:04,600
that because it requires you to 
only pull in all of the 

505
00:25:04,600 --> 00:25:07,960
different IT systems and connect
them all to your graph system. 

506
00:25:07,960 --> 00:25:12,120
Have 20 in the insurance case, 
23,000 ETL pipelines that all 

507
00:25:12,120 --> 00:25:14,400
translate things into one 
unified ontology. 

508
00:25:15,760 --> 00:25:17,680
Sounds like sounds like a 
project that's never going to go

509
00:25:17,680 --> 00:25:20,200
live. 
Yeah, that would be painful. 

510
00:25:20,200 --> 00:25:22,000
Yeah. 
But again, like what what we 

511
00:25:22,000 --> 00:25:24,040
just said like not to get into 
production. 

512
00:25:24,240 --> 00:25:29,160
If I look and I think one thing 
that makes these projects fail, 

513
00:25:29,160 --> 00:25:32,200
it's the the ontology or the 
model, right. 

514
00:25:32,680 --> 00:25:37,360
So imagine you do have the ETL 
pipeline to suck in these 23,000

515
00:25:37,360 --> 00:25:39,840
systems. 
If your model is not right, if 

516
00:25:39,840 --> 00:25:42,000
you're not modeling the right 
entities, you're not modeling it

517
00:25:42,000 --> 00:25:44,200
in the right way, you're not 
gonna succeed. 

518
00:25:44,200 --> 00:25:47,400
But if you have the right model,
I think then you can do well. 

519
00:25:47,400 --> 00:25:49,200
And you can do this step by 
step, right? 

520
00:25:49,200 --> 00:25:52,600
You can start with 100 systems 
and 100 more, but the model is 

521
00:25:52,600 --> 00:25:54,240
really where you fail or. 
Succeed. 

522
00:25:55,160 --> 00:25:57,000
What ensures you have the right 
model? 

523
00:25:57,000 --> 00:25:58,720
Is it starting small and 
expanding? 

524
00:25:58,720 --> 00:26:01,560
Or how do you kind of test your 
assumptions and validate? 

525
00:26:02,000 --> 00:26:04,680
Yeah, I think so. 
The first thing when you're 

526
00:26:04,680 --> 00:26:06,800
making a graph model you're 
coming from relational is you 

527
00:26:06,800 --> 00:26:08,240
throw away your relational 
model. 

528
00:26:08,480 --> 00:26:09,440
That's step one. 
That's step. 

529
00:26:09,440 --> 00:26:11,360
One step one, throw it away. 
I have nothing now. 

530
00:26:12,520 --> 00:26:15,440
And then you. 
So The thing is, the way graphs 

531
00:26:15,440 --> 00:26:18,840
work, they don't do joints like 
there are no real time, there 

532
00:26:18,840 --> 00:26:20,320
are no joints done at query 
time. 

533
00:26:20,680 --> 00:26:25,400
We store the relationships at 
right time, So you have pointers

534
00:26:25,400 --> 00:26:27,720
between elements. 
So what you typically do is you 

535
00:26:27,720 --> 00:26:30,480
take a whiteboard, you start 
drawing, you say, this is my, 

536
00:26:31,200 --> 00:26:34,040
this is my employee, this is a 
skill, this is a team. 

537
00:26:34,360 --> 00:26:38,720
You draw that on a board. 
And almost always you can very 

538
00:26:39,000 --> 00:26:42,080
easily take that whiteboard 
model and convert it to a graph 

539
00:26:42,080 --> 00:26:44,000
model. 
So it's very much closer to 

540
00:26:44,000 --> 00:26:46,080
conceptual model. 
You don't need join tables and 

541
00:26:46,080 --> 00:26:47,760
all these awkward structures. 
Interesting. 

542
00:26:47,760 --> 00:26:50,240
So that's kind of a step one. 
And then I really like what you 

543
00:26:50,240 --> 00:26:51,760
were, what you were doing in 
your experiment. 

544
00:26:51,760 --> 00:26:54,720
Just getting back to that, like 
iterating on the model, seeing 

545
00:26:54,720 --> 00:26:56,640
what edges we can drop, what 
edges we can keep. 

546
00:26:56,640 --> 00:26:59,360
I think that's that's really 
interesting, especially in in 

547
00:26:59,360 --> 00:27:03,760
the age of LLM built ontologies 
and and you know. 

548
00:27:04,720 --> 00:27:08,000
I feel like right now, and it's 
more so now than ever, there's a

549
00:27:08,000 --> 00:27:10,280
lot of information that's like 
new and that's fresh. 

550
00:27:10,280 --> 00:27:13,520
And we were talking about graph 
databases and knowledge graphs 

551
00:27:13,520 --> 00:27:16,200
in the 1st place. 
For me, these are all like I 

552
00:27:16,200 --> 00:27:18,120
knew they existed. 
I've never experimented with 

553
00:27:18,120 --> 00:27:20,600
them because I also see there's 
two types of engineers. 

554
00:27:21,040 --> 00:27:23,880
One of the engineers that you 
said was vector database is 

555
00:27:23,880 --> 00:27:25,240
really cool is now up and 
coming. 

556
00:27:25,240 --> 00:27:26,840
Let me see if I can put that in 
production. 

557
00:27:27,080 --> 00:27:29,040
Regardless of the use case. 
Maybe they just want to play 

558
00:27:29,040 --> 00:27:32,360
around with technologies. 
I know I'm not that type of 

559
00:27:32,360 --> 00:27:33,880
engineer. 
I'm the engineer that's like, I 

560
00:27:33,880 --> 00:27:36,160
want to keep things simple and I
want to keep things maintainable

561
00:27:36,520 --> 00:27:38,480
and rather find like a good 
solution fit. 

562
00:27:38,480 --> 00:27:41,400
And even if I have many options,
I'll pick one, I'll pick, I'll 

563
00:27:41,400 --> 00:27:43,360
try and pick as fast as possible
and then validate my 

564
00:27:43,360 --> 00:27:44,600
assumptions. 
And if it doesn't work, I'll 

565
00:27:44,600 --> 00:27:46,640
pivot. 
And I know for like really good 

566
00:27:46,640 --> 00:27:49,440
solutions, you probably need 
both aspects because otherwise 

567
00:27:49,840 --> 00:27:52,120
one person that's going to look 
at established solutions will 

568
00:27:52,120 --> 00:27:54,960
always lag behind when it comes 
to trailblazing and kind of 

569
00:27:54,960 --> 00:27:56,840
newer technologies. 
And the people that experiment, 

570
00:27:57,120 --> 00:28:01,080
they will get a better feel for 
what works where and form an 

571
00:28:01,080 --> 00:28:03,160
opinion on that. 
And then they can actually apply

572
00:28:03,160 --> 00:28:05,880
that to production. 
So you need this balance of like

573
00:28:06,360 --> 00:28:08,560
I think newer tech and 
experimenting versus actually 

574
00:28:08,560 --> 00:28:11,040
looking at things and keeping 
things maintainable simple. 

575
00:28:11,320 --> 00:28:15,160
It's very often have these, 
maybe not student labs, but a 

576
00:28:15,160 --> 00:28:20,160
lot of young, young, career 
driven individuals joining these

577
00:28:20,240 --> 00:28:24,240
labs of larger companies, which 
is a bit of a ring fenced area 

578
00:28:24,240 --> 00:28:27,120
where you can experiment a lot. 
You can try stuff, you can throw

579
00:28:27,120 --> 00:28:32,720
things away because they can let
their creativity go wild in 

580
00:28:32,720 --> 00:28:36,640
there. 
While I guess the more seasoned 

581
00:28:36,920 --> 00:28:39,760
people that have burned 
themselves a couple times have 

582
00:28:40,040 --> 00:28:42,160
gone through the pain of now I 
have to deal with all this 

583
00:28:42,160 --> 00:28:45,800
legacy and all this complexity. 
They're the ones who then act as

584
00:28:45,800 --> 00:28:48,560
the stage gate to say, do we 
take this forward, yes or no. 

585
00:28:49,200 --> 00:28:51,600
Ideally they're not the grumpy 
kind that just doesn't want 

586
00:28:51,600 --> 00:28:53,680
anything. 
They don't want to be the You 

587
00:28:53,680 --> 00:28:55,800
don't want the person who just 
wants to sit through this for 

588
00:28:55,800 --> 00:28:57,520
another 10 years until they hit 
retirement. 

589
00:28:58,000 --> 00:29:02,040
But you want someone who has 
who's interested in new things, 

590
00:29:02,040 --> 00:29:05,960
but burn themselves enough times
that they don't want to deploy 

591
00:29:06,160 --> 00:29:12,880
the 74th Postgres database that 
does just stores 3 tables. 

592
00:29:15,080 --> 00:29:16,760
But then rather, let's 
consolidate and clean up. 

593
00:29:16,960 --> 00:29:19,520
Yeah, yeah, yeah. 
I'm interested then from your 

594
00:29:19,520 --> 00:29:22,560
perspective, Neil's looking at a
product and you have kind of a 

595
00:29:22,560 --> 00:29:26,240
complex product graph database. 
It has interesting users because

596
00:29:26,240 --> 00:29:27,720
those are then going to be 
developers. 

597
00:29:28,040 --> 00:29:31,280
Do you work on ease of use? 
Do you work on education? 

598
00:29:31,280 --> 00:29:34,480
Or how do you actually kind of 
inform users and make sure they 

599
00:29:34,480 --> 00:29:36,760
know this is the best solution 
for what problem they have? 

600
00:29:37,160 --> 00:29:39,160
Yeah, that's a good question. 
I don't know if graphs are 

601
00:29:39,160 --> 00:29:41,160
necessarily complex, but it's a 
different way of thinking. 

602
00:29:41,160 --> 00:29:42,480
Yeah, I think that's the biggest
thing. 

603
00:29:42,480 --> 00:29:45,000
The modeling is different, the 
querying is different, the 

604
00:29:45,000 --> 00:29:47,520
visualization is different. 
So there's just a a big step 

605
00:29:47,760 --> 00:29:51,040
that you need to take. 
So we do a lot of education. 

606
00:29:52,120 --> 00:29:54,720
One thing that is actually 
really cool is that and for J, 

607
00:29:54,720 --> 00:29:57,920
we had a query language cipher 
for a very long time because we 

608
00:29:57,960 --> 00:30:00,560
open sourced that. 
It wasn't de facto graph 

609
00:30:00,560 --> 00:30:02,600
language. 
And now there is now a standard 

610
00:30:02,760 --> 00:30:06,120
language GQL like Sequel SQL 
that's for graphs. 

611
00:30:06,400 --> 00:30:08,840
Which is super cool because this
is the first standard query 

612
00:30:08,840 --> 00:30:12,400
language in 40 years. 
Again, I'm geeking out this. 

613
00:30:12,920 --> 00:30:16,800
I know, I know, this really 
maybe sounds boring, but there 

614
00:30:16,800 --> 00:30:19,480
has never been a standardized 
query language for no SQL 

615
00:30:19,480 --> 00:30:21,240
databases. 
They just couldn't get to the 

616
00:30:21,240 --> 00:30:22,880
consensus. 
But the fact that there's not 

617
00:30:22,880 --> 00:30:25,520
one for graphs is awesome 
because you can then pivot 

618
00:30:25,520 --> 00:30:27,040
between different graph 
databases. 

619
00:30:27,360 --> 00:30:30,760
It's standard and it's just, I 
don't know, that's just feels 

620
00:30:30,760 --> 00:30:32,800
really cool to me. 
And we actually the people that 

621
00:30:32,800 --> 00:30:34,920
were involved in creating the 
GQL standard where some of the 

622
00:30:34,920 --> 00:30:37,320
people working on my original 
SQL standard 40 years ago. 

623
00:30:37,320 --> 00:30:39,960
So it's amazing that these 
people just came back. 

624
00:30:39,960 --> 00:30:42,000
Some came back from, I don't 
know what they were doing, 

625
00:30:42,000 --> 00:30:45,120
retired or retired, but it's 
just amazing and I think it's. 

626
00:30:45,240 --> 00:30:47,640
Really nice to have them on. 
Yeah, I think that in terms of 

627
00:30:47,680 --> 00:30:50,520
education, you know, first of 
all, if you have a standard 

628
00:30:50,520 --> 00:30:53,280
language, people will Start 
learning that and then this 

629
00:30:53,280 --> 00:30:56,800
other thinking, modelling, 
visualization, that comes later.

630
00:30:56,800 --> 00:30:59,520
But having a good language is is
is really important. 

631
00:30:59,520 --> 00:31:01,560
Yeah, Yeah. 
So that is, but that's not 

632
00:31:01,560 --> 00:31:03,440
really what you focus on for 
product sense, right? 

633
00:31:03,440 --> 00:31:05,440
Is that more the education side 
that you focus on? 

634
00:31:05,680 --> 00:31:07,560
Yeah. 
So I'm, so my, my job right now 

635
00:31:07,560 --> 00:31:09,640
is I'm working on visualization 
tools, right? 

636
00:31:09,640 --> 00:31:12,400
So I'm, I'm working on 
dashboarding tool for graphs. 

637
00:31:12,680 --> 00:31:16,040
So you can imagine your standard
BI tool works really well with 

638
00:31:16,040 --> 00:31:17,800
tables. 
Yeah, graphs, not so much. 

639
00:31:18,600 --> 00:31:21,560
So we're making a tool that has 
a graph visualization component 

640
00:31:21,560 --> 00:31:25,440
in it, ease of use using then 
cipher or G call. 

641
00:31:26,680 --> 00:31:30,640
And from that perspective, what 
I'm focusing on is user 

642
00:31:30,640 --> 00:31:34,000
experience. 
So my, my thinking is graphs are

643
00:31:34,000 --> 00:31:36,960
for many people difficult, new, 
I don't know, difficult is the 

644
00:31:36,960 --> 00:31:38,200
right word, but it's new is 
different. 

645
00:31:38,560 --> 00:31:41,120
So I feel like we should have a 
higher standard of user 

646
00:31:41,120 --> 00:31:43,760
experience compared to 
established tools because 

647
00:31:43,760 --> 00:31:48,000
there's already the cognitive 
jump to move to graphs. 

648
00:31:48,200 --> 00:31:50,480
If there's a cognitive jump to 
move to a different user 

649
00:31:50,480 --> 00:31:52,360
interface as well or, or the 
experience. 

650
00:31:52,400 --> 00:31:53,840
Yeah, I think that's difficult 
for people. 

651
00:31:53,840 --> 00:31:57,080
So I, I really care about user 
experience and I've had hour 

652
00:31:57,080 --> 00:32:00,200
long discussions with my team 
about very minor things like 

653
00:32:00,440 --> 00:32:02,520
button placement. 
But I but I love that. 

654
00:32:02,520 --> 00:32:06,920
I think it's, it's really nice 
to have amazing user experience.

655
00:32:07,120 --> 00:32:09,160
Yeah, yeah. 
I've said this many times on the

656
00:32:09,160 --> 00:32:12,160
podcast, like if I have to go 
through a hurdle to try 

657
00:32:12,160 --> 00:32:14,320
something out, I'm gone. 
I'm just, I'll search something 

658
00:32:14,320 --> 00:32:15,920
else. 
Yeah, if there's a paywall, I'm 

659
00:32:15,920 --> 00:32:17,080
out. 
But that's me. 

660
00:32:17,600 --> 00:32:20,360
I think people are impatient. 
I think it's also with the age 

661
00:32:20,360 --> 00:32:23,320
of LLMS getting even worse. 
Like people don't have patience 

662
00:32:23,320 --> 00:32:25,400
anymore. 
And if you build for the most 

663
00:32:25,400 --> 00:32:29,360
impatient person, then the more 
patient people will just also be

664
00:32:29,360 --> 00:32:31,120
happy because for them it feels 
like a breeze. 

665
00:32:31,240 --> 00:32:34,440
You know they won't struggle. 
The struggles that they would 

666
00:32:34,440 --> 00:32:36,320
normally have patience with, 
They're not there. 

667
00:32:36,320 --> 00:32:38,960
So it feels amazing to them. 
And the inpatient ones will 

668
00:32:38,960 --> 00:32:41,200
still be grumpy, but at least 
they'll they'll stick. 

669
00:32:41,200 --> 00:32:42,400
They won't give up. 
Yeah. 

670
00:32:42,920 --> 00:32:45,440
Yeah. 
Have you seen more adoption with

671
00:32:45,440 --> 00:32:47,600
regards to Gen. 
AI, because we mentioned RAG, we

672
00:32:47,600 --> 00:32:49,920
mentioned graph RAG is this way 
of kind of distilling your 

673
00:32:49,920 --> 00:32:53,320
context, giving it in a 
different way to reason about 

674
00:32:53,320 --> 00:32:56,760
and it might even outperform 
some other embeddings or might 

675
00:32:56,760 --> 00:32:59,400
have higher accuracy. 
Have you seen like improved 

676
00:32:59,400 --> 00:33:01,680
usage? 
I've got it in graph databases 

677
00:33:01,680 --> 00:33:04,120
in general because of that. 
Oh yeah, 100%, yeah, yeah. 

678
00:33:04,120 --> 00:33:06,480
Everyone is moving towards graph
Rag, especially right now. 

679
00:33:06,480 --> 00:33:09,520
It's we've had a lot of people 
come to us and say, hey guys, we

680
00:33:09,520 --> 00:33:11,120
tried Rag, but it's just not 
working out. 

681
00:33:11,160 --> 00:33:13,320
OK. 
That's the pattern that we see, 

682
00:33:13,520 --> 00:33:14,520
OK. 
And then they say. 

683
00:33:14,520 --> 00:33:17,200
To cater to that then, because 
they're already frustrated, 

684
00:33:17,200 --> 00:33:18,840
right? 
So you now have to be the 

685
00:33:18,840 --> 00:33:21,120
saviour for them. 
That's a tough spot to be in. 

686
00:33:21,480 --> 00:33:23,840
We tried vector databases, it 
didn't cut it for us. 

687
00:33:23,840 --> 00:33:27,840
Now this next, like we've got 
one more shot at getting this 

688
00:33:27,840 --> 00:33:29,680
right before we're going to get 
our funding cut. 

689
00:33:31,520 --> 00:33:34,080
Now fix it. 
Yeah, No, your database now has 

690
00:33:34,080 --> 00:33:35,640
to be the same here. 
It's a tough one. 

691
00:33:35,920 --> 00:33:37,360
Yeah. 
I mean, on the upside though, 

692
00:33:37,360 --> 00:33:40,040
people realize that simple rack 
doesn't cut it. 

693
00:33:40,040 --> 00:33:43,240
So they say, OK, we're OK with 
investing this time into making 

694
00:33:43,240 --> 00:33:46,680
a model and getting things set 
up right to get that boost in 

695
00:33:46,680 --> 00:33:48,040
accuracy, that graph rack. 
Yeah, first. 

696
00:33:48,640 --> 00:33:52,040
But things are moving so fast. 
Like I have had people on half a

697
00:33:52,040 --> 00:33:54,000
year ago, a few months ago, 
maybe a year ago. 

698
00:33:54,000 --> 00:33:56,880
They were like vector DBS of the
thing and now you're already 

699
00:33:56,880 --> 00:33:59,040
saying it's not cutting it and 
we need to go to this graph 

700
00:33:59,040 --> 00:34:00,800
database to like fix. 
It, I don't know, I think vector

701
00:34:00,800 --> 00:34:04,040
DBS have have a good place, but 
I think if you really care about

702
00:34:04,160 --> 00:34:06,680
accuracy, traceability is 
something that comes up all the 

703
00:34:06,680 --> 00:34:09,000
time. 
People say, I want to know that 

704
00:34:09,199 --> 00:34:11,800
where this answer comes from, 
how was it derived? 

705
00:34:11,800 --> 00:34:15,840
And again, back to the graph, 
what are the entities in my 

706
00:34:15,840 --> 00:34:18,199
graph that were used to compose 
this answer from this large 

707
00:34:18,199 --> 00:34:20,040
language model? 
So that's another thing. 

708
00:34:20,040 --> 00:34:23,560
And and for big companies, the 
difference between 80% accurate 

709
00:34:23,560 --> 00:34:27,840
and 90% accurate or 90% to 95, 
that's, that's huge. 

710
00:34:27,880 --> 00:34:28,600
That's huge. 
Yeah. 

711
00:34:28,679 --> 00:34:30,320
Gotcha. 
What I find a bit curious is 

712
00:34:30,320 --> 00:34:33,120
that all these companies are 
trying to build this themselves.

713
00:34:33,719 --> 00:34:38,199
There's RAG came about and then 
every company decided to build 

714
00:34:38,199 --> 00:34:41,600
their own knowledge assistant. 
And I get the desire. 

715
00:34:41,800 --> 00:34:47,199
Having a Volkswagen GPT or a 
Continental GPT or, you know, 

716
00:34:47,360 --> 00:34:51,960
HSPCGPT, well, it's too many 
acronyms, but that's very 

717
00:34:51,960 --> 00:34:53,000
powerful. 
You want that. 

718
00:34:53,760 --> 00:34:57,640
But then at the same time, 
Google dominates search because 

719
00:34:57,640 --> 00:35:01,880
all of this is a search problem.
Microsoft tried to come in and 

720
00:35:01,880 --> 00:35:04,520
it took them ages to get a 
decent search engine going. 

721
00:35:04,920 --> 00:35:07,320
It's very, very hard to build a 
good search engine. 

722
00:35:07,720 --> 00:35:11,320
So it's curious that now every 
company under the sun has a team

723
00:35:11,560 --> 00:35:15,240
somewhere that is supposed to 
build their internal knowledge 

724
00:35:15,240 --> 00:35:19,320
agent. 
And as one problem out of many, 

725
00:35:19,320 --> 00:35:21,840
I mean, they have to get user 
usability, right? 

726
00:35:21,840 --> 00:35:25,360
Access management, right? 
What is the information that you

727
00:35:25,360 --> 00:35:27,160
can actually have access to all 
the Etls? 

728
00:35:27,720 --> 00:35:30,040
And then they also just have to 
like on the side, solve the 

729
00:35:30,040 --> 00:35:31,920
retrieval problem, which is a 
search problem. 

730
00:35:31,920 --> 00:35:36,760
It just seems like I would, I 
would rather say, OK, can we sit

731
00:35:36,760 --> 00:35:40,200
this one out and wait for a year
instead, get all the stars 

732
00:35:40,200 --> 00:35:42,360
align. 
So what, what will we need to 

733
00:35:42,360 --> 00:35:44,480
do? 
We will need to make sure that 

734
00:35:44,480 --> 00:35:48,960
our data is accessible, well 
structured, that we have some 

735
00:35:48,960 --> 00:35:52,200
form of a grip of what are the 
main queries that we need 

736
00:35:52,200 --> 00:35:55,800
answering. 
And then if and when a start up 

737
00:35:55,800 --> 00:35:59,240
from Y Combinator or maybe one 
of the big tech companies bring 

738
00:35:59,240 --> 00:36:00,960
something that solves this 
problem. 

739
00:36:00,960 --> 00:36:04,360
So you can really feel like you 
can find the information in your

740
00:36:04,360 --> 00:36:07,480
entire internal knowledge world 
the same way as you can just ask

741
00:36:07,480 --> 00:36:10,560
strategy BT for something. 
When that technology comes about

742
00:36:10,560 --> 00:36:12,720
that has get that gets 
everything right, the right mix 

743
00:36:12,720 --> 00:36:16,480
of graph database, vector 
database, keyword search, index,

744
00:36:16,480 --> 00:36:20,280
look UPS, ranking, re ranking, 
synthesizing. 

745
00:36:20,920 --> 00:36:25,480
Then you plug that one in. 
Yeah, I still, I would have 

746
00:36:25,480 --> 00:36:28,200
probably, I gave this advice two
years ago and I would still give

747
00:36:28,200 --> 00:36:30,720
it to companies like sit this 
one out, wait until someone 

748
00:36:30,720 --> 00:36:34,880
solves and productizes it, and 
focus on the things you need to 

749
00:36:34,880 --> 00:36:38,360
actually get right data use 
cases. 

750
00:36:38,480 --> 00:36:40,080
What are the main queries and 
what are the. 

751
00:36:40,600 --> 00:36:42,440
Yeah. 
Where are the biggest levers of 

752
00:36:43,120 --> 00:36:44,360
optimization for you as a 
company? 

753
00:36:44,520 --> 00:36:45,880
Yeah. 
I mean, I feel like there's 

754
00:36:45,880 --> 00:36:49,680
never been as much FOMO as there
is right now. 

755
00:36:50,000 --> 00:36:53,120
Yeah, right. 
And I think I can kind of think 

756
00:36:53,120 --> 00:36:55,440
of how to explain it. 
Proof of concepts are quite 

757
00:36:55,440 --> 00:36:58,800
easy, right? 
Models are decently accessible. 

758
00:36:58,800 --> 00:37:01,960
They're getting more and more 
accessible, giving access to 

759
00:37:01,960 --> 00:37:04,720
proof of concept data set to 
prove value of a certain 

760
00:37:04,720 --> 00:37:07,280
business case without even 
looking at cost factors. 

761
00:37:07,280 --> 00:37:10,200
Because especially in larger 
organizations, cost doesn't come

762
00:37:10,200 --> 00:37:11,640
to play in proof of concepts 
yet. 

763
00:37:12,000 --> 00:37:15,400
Especially if you look at like 
say your IT landscape versus 

764
00:37:15,400 --> 00:37:16,760
using everything that's out 
there. 

765
00:37:16,960 --> 00:37:19,360
If you can do everything that's 
out there in a proof of concept,

766
00:37:19,360 --> 00:37:21,560
you can get something decently 
up and running. 

767
00:37:21,880 --> 00:37:25,440
And then that's the next step is
like, OK, user adoption is like 

768
00:37:25,440 --> 00:37:27,800
a huge thing. 
If people have this ingrained 

769
00:37:27,800 --> 00:37:30,680
way of working and they do that 
over and over again, it's like 

770
00:37:30,680 --> 00:37:32,640
their pride. 
That's what they're there for, 

771
00:37:32,640 --> 00:37:34,440
That's their job. 
And you come in with a tool and 

772
00:37:34,440 --> 00:37:37,640
that's like 75% accurate. 
Now let's see if you're more 

773
00:37:37,640 --> 00:37:38,800
productive. 
People will be like, what is 

774
00:37:38,800 --> 00:37:41,000
this shit? 
And even though it might be 

775
00:37:41,000 --> 00:37:44,040
accurate, let's say from a 
theoretical sense, if the 

776
00:37:44,040 --> 00:37:46,600
linguistics are not the same, if
the tone of voice is not the 

777
00:37:46,600 --> 00:37:49,120
same, people that be like, it 
doesn't work for me. 

778
00:37:49,440 --> 00:37:51,720
They'll throw it in the trash 
faster than anything else 

779
00:37:52,080 --> 00:37:54,200
because they also see that it's 
just another tool. 

780
00:37:54,200 --> 00:37:57,440
And then instead of going from 
14 screens, they now have a 15th

781
00:37:57,440 --> 00:38:00,040
screen that they sometimes have 
to do something with and it's 

782
00:38:00,040 --> 00:38:01,520
just another tool in their tool 
belt. 

783
00:38:02,240 --> 00:38:05,280
I feel like that is the most 
underestimated part is the 

784
00:38:05,280 --> 00:38:07,920
adoption sense. 
There might definitely be really

785
00:38:07,920 --> 00:38:11,040
good use cases in organizations.
And I do agree that taking 

786
00:38:11,040 --> 00:38:13,440
something off the shelf versus 
like really looking at the 

787
00:38:13,440 --> 00:38:16,320
context that you have, 
leveraging that to make a more 

788
00:38:16,320 --> 00:38:17,840
optimal solution. 
I think the second one is 

789
00:38:17,840 --> 00:38:21,400
better, but I also recognise 
that not a lot of people, not a 

790
00:38:21,400 --> 00:38:24,400
lot of organizations are 
equipped to be able to execute 

791
00:38:25,400 --> 00:38:28,000
to the high standards that it 
needs to to succeed in that way.

792
00:38:28,640 --> 00:38:30,560
For me, I feel like that's the 
bigger challenge. 

793
00:38:31,280 --> 00:38:33,560
We saw that, at least in 
consultancy, a lot of 

794
00:38:33,560 --> 00:38:37,320
organizations have this FOMO 
fear, like fear missing out, 

795
00:38:37,320 --> 00:38:38,840
basically. 
I don't know why I said FOMO 

796
00:38:38,840 --> 00:38:41,800
fear, but in any case, they're 
like, we need to do something 

797
00:38:41,800 --> 00:38:44,240
with AI because our competitors 
are doing something with AI. 

798
00:38:44,560 --> 00:38:47,240
And if everyone has that 
mindset, then indeed it UPS this

799
00:38:47,240 --> 00:38:50,760
fear of missing out completely. 
And then people were looking at 

800
00:38:50,760 --> 00:38:52,360
their own organizations and 
seeing what's out there. 

801
00:38:52,360 --> 00:38:55,240
And then chat bots was like the 
fast food Jenny I solution that 

802
00:38:55,240 --> 00:38:57,080
a lot of companies were building
themselves. 

803
00:38:57,520 --> 00:39:00,120
And then I asked the same exact 
question that you did. 

804
00:39:00,680 --> 00:39:01,920
Why are you building this 
yourself? 

805
00:39:01,960 --> 00:39:04,400
Like there are companies and 
there are start-ups that this is

806
00:39:04,400 --> 00:39:06,840
the problem. 
How do we distill context from 

807
00:39:06,840 --> 00:39:09,400
an organization and create 
tooling for that to have an 

808
00:39:09,400 --> 00:39:13,600
optimal user customer journey 
with regards to chat And 

809
00:39:13,600 --> 00:39:15,520
companies think they can do that
better themselves. 

810
00:39:15,520 --> 00:39:18,960
I was like, I don't, I don't see
this yet, but waiting it out. 

811
00:39:19,160 --> 00:39:22,240
I feel like there's also not 
really a convincing alternative,

812
00:39:22,680 --> 00:39:24,880
which means you need to start 
experimenting and start small 

813
00:39:24,880 --> 00:39:27,400
and kind of incrementally grow, 
which I think is healthy. 

814
00:39:27,400 --> 00:39:29,760
But then, yeah, you see a lot of
failure and hopefully that's 

815
00:39:29,760 --> 00:39:31,120
good. 
If there's a culture of failing 

816
00:39:31,120 --> 00:39:33,280
is OK and we learn, then that's 
the best. 

817
00:39:34,080 --> 00:39:37,480
I think there's also, of course,
it's not black or white, right? 

818
00:39:37,480 --> 00:39:39,760
But let me paint 2 extreme 
cases. 

819
00:39:39,760 --> 00:39:41,160
And then there's companies all 
in between. 

820
00:39:41,560 --> 00:39:45,040
You've got companies where, and 
I think there's probably a vast 

821
00:39:45,040 --> 00:39:49,360
majority of the companies out 
there, there's going to be 

822
00:39:49,360 --> 00:39:51,680
dozens of initiatives going on 
at the same time. 

823
00:39:52,040 --> 00:39:56,960
People are running 55, you know,
hundreds of projects depending 

824
00:39:56,960 --> 00:39:59,600
on the size of the company, of 
course, that that number keeps 

825
00:39:59,600 --> 00:40:03,920
growing. 
But I think that the hallmark or

826
00:40:03,920 --> 00:40:07,000
the the marker that I would look
for is if you're in a group of 

827
00:40:07,000 --> 00:40:11,200
managers that are responsible 
for a department or for a 

828
00:40:11,200 --> 00:40:14,720
certain region or whatever and 
you ask all of them, what are 

829
00:40:14,720 --> 00:40:18,680
the three overarching? 
Priorities for us for this year 

830
00:40:18,880 --> 00:40:24,320
or for this quarter in a company
that is very chaolic these and 

831
00:40:24,320 --> 00:40:26,640
everyone writes them down 
individually and then you put 

832
00:40:26,640 --> 00:40:28,480
them side by side, they will be 
very misaligned. 

833
00:40:28,640 --> 00:40:30,920
Everyone will kind of be in 
their own mind, in their own 

834
00:40:30,920 --> 00:40:32,920
headspace. 
And then you've got a laser 

835
00:40:32,920 --> 00:40:38,080
focus companies on the other end
of the extreme where you could 

836
00:40:38,280 --> 00:40:40,760
probably go at through all of 
the different levels of the 

837
00:40:40,760 --> 00:40:44,120
company and you could say again,
write down like what are the 

838
00:40:44,120 --> 00:40:48,000
three top priorities for you? 
They will be very unified 

839
00:40:48,120 --> 00:40:50,840
because everyone's aligned, 
everyone's kind of on the same 

840
00:40:50,840 --> 00:40:52,920
radar, on the same, on the same 
track. 

841
00:40:53,560 --> 00:40:55,040
And of course it starts at the 
top. 

842
00:40:55,040 --> 00:40:57,120
You need someone to just give a 
laser sharp vision. 

843
00:40:57,320 --> 00:40:58,880
This is our objective, this is 
our vision. 

844
00:40:58,880 --> 00:41:01,200
This is what we what this 
company was made for. 

845
00:41:02,440 --> 00:41:06,800
If you're over here, then you 
have such clarity. 

846
00:41:07,440 --> 00:41:10,440
People are able to now deploy a 
new project. 

847
00:41:10,440 --> 00:41:14,240
They can say, OK, we are now 
going to go after customer 

848
00:41:14,240 --> 00:41:17,520
support agents or customer 
support AI bots because it'll 

849
00:41:17,520 --> 00:41:20,120
drastically improve our customer
attention, Whatever. 

850
00:41:21,360 --> 00:41:23,800
The companies over here I think 
are the ones that you and I are 

851
00:41:23,800 --> 00:41:27,120
familiar with where these 
projects actually get kicked 

852
00:41:27,120 --> 00:41:28,960
off. 
And then it's one project that 

853
00:41:28,960 --> 00:41:32,160
gets drowned in all the other 
things that are going on. 

854
00:41:32,160 --> 00:41:34,880
All the other initiatives, 
probably they have three half 

855
00:41:34,880 --> 00:41:39,360
finished transformation projects
still going on Guilty having 

856
00:41:39,360 --> 00:41:43,240
been in a consultancy where you 
then also have the access to the

857
00:41:43,240 --> 00:41:46,240
executive level, right. 
So then there's probably stuff 

858
00:41:46,240 --> 00:41:50,560
that gets pushed through with 
like you have the the rank, 

859
00:41:50,560 --> 00:41:52,520
right. 
So this is an executive sponsor 

860
00:41:52,520 --> 00:41:54,680
project, so it gets priority 
treatment. 

861
00:41:54,680 --> 00:41:57,920
So then others have to wait. 
That's frustrating in that chaos

862
00:41:57,920 --> 00:41:59,880
projects. 
It's very hard to get a project 

863
00:41:59,880 --> 00:42:04,480
to to succeed unless you make it
absolutely minimal, absolutely 

864
00:42:04,480 --> 00:42:07,120
simple. 
Yeah. 

865
00:42:07,200 --> 00:42:11,480
But just whipping up a company 
wide knowledge agent there, 

866
00:42:11,880 --> 00:42:12,880
tough one. 
Yeah. 

867
00:42:13,280 --> 00:42:15,520
It's not good enough. 
I feel like like you need to 

868
00:42:15,520 --> 00:42:19,480
figure out indeed, what does it 
contribute to in in business 

869
00:42:19,480 --> 00:42:22,080
outcomes and if there's no laser
focus from an organizational 

870
00:42:22,080 --> 00:42:26,240
sense, then for some how shape 
or form you need to figure that 

871
00:42:26,240 --> 00:42:30,240
out bottom up right. 
This use case that I have, does 

872
00:42:30,240 --> 00:42:32,600
it actually make sense with 
regards to cost and benefit and 

873
00:42:32,600 --> 00:42:35,320
is going to be enough time saved
for people to actually work to 

874
00:42:35,320 --> 00:42:37,600
adopt it? 
Like is it going to stand true 

875
00:42:37,920 --> 00:42:40,760
or do we need to indeed focus on
kind of the exact decision and 

876
00:42:40,760 --> 00:42:43,040
like the highest priority cases 
and do we focus on that? 

877
00:42:43,720 --> 00:42:46,240
I also wonder what would happen 
if you take a lot of the 

878
00:42:46,240 --> 00:42:48,760
companies and you just said, OK,
we're going to have to get rid 

879
00:42:48,760 --> 00:42:50,520
of 50% of all of our 
initiatives. 

880
00:42:50,680 --> 00:42:53,320
That's just a extrinsic 
requirement. 

881
00:42:53,320 --> 00:42:57,640
We're going to crank up the like
the evolutionary pressure a 

882
00:42:57,640 --> 00:42:59,880
little bit. 
You do not have capacity to do 

883
00:42:59,880 --> 00:43:01,280
all these things. 
So 50% has to go. 

884
00:43:02,120 --> 00:43:03,480
That doesn't mean that people 
have to go. 

885
00:43:03,480 --> 00:43:06,680
It just means that the 
multitasking of the Organism as 

886
00:43:06,680 --> 00:43:12,960
a whole has to be toned down. 
I think that would actually be a

887
00:43:12,960 --> 00:43:14,680
very healthy exercise for a lot 
of companies. 

888
00:43:14,680 --> 00:43:17,560
Yeah, I'm there with you. 
But that's the only reason I'm 

889
00:43:17,560 --> 00:43:20,440
there with you is because I had 
an inkling that was the case. 

890
00:43:20,440 --> 00:43:23,720
And then I was responsible for 
product and I already inherited 

891
00:43:23,720 --> 00:43:25,480
a road map that had like many 
tracks in parallel. 

892
00:43:25,480 --> 00:43:27,000
And I was like, this is this is 
impossible. 

893
00:43:27,160 --> 00:43:28,800
We cannot do all of this in 
parallel. 

894
00:43:28,880 --> 00:43:31,840
Nothing's going to work. 
Everything will go slower as a 

895
00:43:31,840 --> 00:43:35,320
response to that. 
So then one of my, and I also 

896
00:43:35,320 --> 00:43:38,000
thought that was my role was to 
say we do one thing at a time, 

897
00:43:38,040 --> 00:43:40,200
we execute, we deliver and we 
move on to the next. 

898
00:43:40,520 --> 00:43:42,120
And there will not be many 
things in parallel. 

899
00:43:42,120 --> 00:43:45,200
There will be nothing because I 
have this ordered list and we go

900
00:43:45,200 --> 00:43:46,720
top to bottom and then that's 
it. 

901
00:43:47,000 --> 00:43:48,840
I was very, I, I did not have a 
road map. 

902
00:43:48,840 --> 00:43:50,400
I don't, I didn't have any Gantt
charts. 

903
00:43:50,480 --> 00:43:53,920
Just made one XL sheet, it has 
already rose and it's very easy 

904
00:43:53,920 --> 00:43:57,600
to follow how we do things stop 
to bottom that's it in it's 

905
00:43:57,600 --> 00:44:01,440
simplest and kind of in essence 
priority and that's how we 

906
00:44:01,440 --> 00:44:03,520
execute it. 
And in the end it was really 

907
00:44:03,520 --> 00:44:05,360
hard to get through, but it was 
quite effective. 

908
00:44:05,360 --> 00:44:08,160
I. 
Have this hate love hate 

909
00:44:08,160 --> 00:44:12,040
relationship with gang charts. 
My first full time job, I worked

910
00:44:12,040 --> 00:44:15,680
for a subcontractor for the 
European Space Agency and we 

911
00:44:15,680 --> 00:44:18,760
built the gang chart that plans 
out everything that you need to 

912
00:44:18,760 --> 00:44:22,720
bring to the space station. 
Because it was the, I forgot 

913
00:44:22,720 --> 00:44:25,000
what the name was, but it's like
the management platform for the 

914
00:44:25,000 --> 00:44:28,880
ISS. 
The ISS has some pretty hard 

915
00:44:28,880 --> 00:44:32,760
deadlines because the rocket is 
going up if you forgot your 

916
00:44:32,760 --> 00:44:34,720
screwdriver. 
You're screwed. 

917
00:44:34,800 --> 00:44:37,400
You can't go back. 
That was a good pun. 

918
00:44:37,400 --> 00:44:41,760
I'm kind of proud of that one. 
And so that one, there was no 

919
00:44:41,760 --> 00:44:44,880
gun software for this kind of 
use case because gun softwares 

920
00:44:44,880 --> 00:44:48,120
are made with this idea that 
you've got dependencies, you can

921
00:44:48,120 --> 00:44:50,360
move them forward and backward, 
you can make them the blocks 

922
00:44:50,360 --> 00:44:52,800
large and smaller. 
But it didn't have this concept 

923
00:44:52,800 --> 00:44:56,320
of you have launch windows and 
you're going to launch something

924
00:44:56,520 --> 00:44:59,360
and you have certain capacity. 
And so all of these, this linear

925
00:44:59,360 --> 00:45:01,280
optimization had to be custom 
built. 

926
00:45:01,680 --> 00:45:05,920
So I built gun software and I 
was kind of proud of it. 

927
00:45:05,920 --> 00:45:07,600
It's super hard to build on the 
front end. 

928
00:45:07,600 --> 00:45:10,120
You've got all this state 
management with the reactive 

929
00:45:10,120 --> 00:45:15,040
redox and whatever, but then you
see a lot of project plans and 

930
00:45:15,040 --> 00:45:18,080
you just see all of these blocks
on top of each other. 

931
00:45:19,080 --> 00:45:21,520
And I often look at these and 
then I just go, yeah, this 

932
00:45:21,520 --> 00:45:23,160
project's going to be massively 
delayed. 

933
00:45:24,160 --> 00:45:29,480
And then I mean, how, why? 
If you then get asked, like, how

934
00:45:29,480 --> 00:45:32,200
do you know? 
I don't know, just intuition. 

935
00:45:32,200 --> 00:45:34,200
This is not going to work 
because we can't possibly do all

936
00:45:34,200 --> 00:45:35,720
these things at the same time. 
People are going to be 

937
00:45:35,720 --> 00:45:38,080
distracted. 
This is just modelling what you 

938
00:45:38,080 --> 00:45:39,280
know. 
There's just all the stuff that 

939
00:45:39,280 --> 00:45:42,480
you don't know and your team of 
five people already has 25 

940
00:45:42,800 --> 00:45:45,480
active things going on just in 
your project management tool, 

941
00:45:46,080 --> 00:45:47,960
plus all the stuff that you 
don't even model. 

942
00:45:48,400 --> 00:45:51,280
No way. 
But then if you distill it down 

943
00:45:51,280 --> 00:45:54,560
to a gun chart where you would 
really say, OK, we have 3 

944
00:45:54,560 --> 00:45:58,480
objectives and that's the most 
we can paralyze. 

945
00:45:59,560 --> 00:46:01,800
So when we finish those, then we
start with the next thing. 

946
00:46:02,200 --> 00:46:04,800
Then you don't need a gun chart 
anymore because you always have 

947
00:46:04,800 --> 00:46:06,400
three things. 
You can just write them on a 

948
00:46:06,400 --> 00:46:10,680
list and you finish like you 
cross one off and then a new one

949
00:46:10,680 --> 00:46:13,360
makes it. 
So then gun charts, they come 

950
00:46:13,360 --> 00:46:15,160
from engineering. 
That makes sense if you build a 

951
00:46:15,160 --> 00:46:18,120
skyscraper, but they don't make 
sense in project management in 

952
00:46:18,120 --> 00:46:20,920
software in my mind. 
It sounds like a graph problem. 

953
00:46:22,920 --> 00:46:24,320
Dependencies. 
Yeah, exactly. 

954
00:46:24,440 --> 00:46:26,560
Dependencies between sorry, 
broke a record. 

955
00:46:27,280 --> 00:46:29,880
Like, as a final thought, I was 
wondering what you both think of

956
00:46:29,880 --> 00:46:33,200
this because for me, newer 
technology pops up, right? 

957
00:46:33,200 --> 00:46:35,720
We have, first of all, newer 
models that just keep popping 

958
00:46:35,720 --> 00:46:37,000
up. 
There's tools that are trying to

959
00:46:37,000 --> 00:46:40,200
solve organizational 
dysfunctional behaviour, but 

960
00:46:40,200 --> 00:46:42,760
also behaviour that like 
alleviate certain problems that 

961
00:46:42,760 --> 00:46:44,920
you didn't even know you had. 
And all of a sudden now with 

962
00:46:44,920 --> 00:46:48,440
tooling and technology, it might
be a problem or it might be a 

963
00:46:48,440 --> 00:46:52,160
solution that's offered to you. 
Which means me as an individual 

964
00:46:52,160 --> 00:46:54,040
or let's take me as a software 
engineer. 

965
00:46:55,120 --> 00:46:57,600
I have a challenge with regards 
to my learning journey like you 

966
00:46:57,600 --> 00:46:59,440
with regards to Gantt charts. 
I'm not going to learn 

967
00:46:59,440 --> 00:47:00,960
everything at the same time 
incrementally. 

968
00:47:00,960 --> 00:47:04,320
I like to focus and prioritize. 
But then also from a software 

969
00:47:04,320 --> 00:47:07,800
engineering hat, I need not just
the theory, I need to be able to

970
00:47:07,800 --> 00:47:09,680
walk the walk. 
So I also need to build stuff. 

971
00:47:09,680 --> 00:47:13,000
And I feel like, especially 
nowadays, I feel like I need to 

972
00:47:13,000 --> 00:47:14,760
build more. 
I don't feel like I'm building 

973
00:47:14,760 --> 00:47:17,640
enough. 
How do you guys learn or 

974
00:47:17,640 --> 00:47:20,800
familiarize, familiarize 
yourself with new topics with 

975
00:47:20,800 --> 00:47:23,400
regards to executing and then 
trying to be kind of an 

976
00:47:23,400 --> 00:47:26,600
authority on that topic? 
How do you handle that? 

977
00:47:28,360 --> 00:47:30,880
Good question. 
Yeah, so I, I started being a 

978
00:47:30,880 --> 00:47:34,680
product manager about a year ago
and I was an engineer before 

979
00:47:34,720 --> 00:47:36,600
solution architect doing a lot 
of hands on work. 

980
00:47:36,800 --> 00:47:39,360
And what I found myself doing 
recently is instead of writing 

981
00:47:39,360 --> 00:47:43,000
an RFPI just vibe called 
something which is really fun 

982
00:47:43,240 --> 00:47:46,520
because it kind of helps you 
think about a problem one step 

983
00:47:46,520 --> 00:47:49,040
ahead. 
And it also is an opportunity to

984
00:47:49,040 --> 00:47:50,440
get in touch with new 
technologies, right. 

985
00:47:50,440 --> 00:47:51,880
If you're building something 
front end, you can try a 

986
00:47:51,880 --> 00:47:55,880
different stack every time. 
So I think coding something, 

987
00:47:55,880 --> 00:47:58,560
vibe coding, whether it's vibe 
coding or actually learning like

988
00:47:58,560 --> 00:48:01,480
a language, whatever, just 
having these tiny problems in 

989
00:48:01,480 --> 00:48:05,000
your life and thinking about 
those in a software way, could 

990
00:48:05,040 --> 00:48:07,560
we could really help, yeah. 
What have you Vibe called it 

991
00:48:07,680 --> 00:48:09,920
lately? 
What have I vibe coded recently?

992
00:48:11,800 --> 00:48:14,840
Actually something really fun. 
So we're building a cipher 

993
00:48:14,840 --> 00:48:17,760
editor. 
So for querying the graph, and 

994
00:48:17,760 --> 00:48:20,760
what I wanted to do is I wanted 
to have intelligence suggestions

995
00:48:20,760 --> 00:48:22,920
based on the questions to the 
user. 

996
00:48:22,920 --> 00:48:26,560
So, OK, a little bit context. 
OK, so I'll keep it short. 

997
00:48:26,920 --> 00:48:29,760
We have for the graph, we have a
cyber interface which is 

998
00:48:29,760 --> 00:48:30,800
queries. 
And then we have a natural 

999
00:48:30,800 --> 00:48:33,320
language interface where you can
type normal text and it 

1000
00:48:33,320 --> 00:48:35,320
translated to a graph query 
using an LLM. 

1001
00:48:35,480 --> 00:48:36,640
Love that. 
That's really cool. 

1002
00:48:36,840 --> 00:48:38,960
So you can use and you don't 
need to learn, don't need to 

1003
00:48:38,960 --> 00:48:41,600
learn the graph language because
you can just use natural 

1004
00:48:41,600 --> 00:48:44,040
language. 
But there's still the problem of

1005
00:48:44,040 --> 00:48:47,080
looking at a blank screen. 
It says natural language input 

1006
00:48:47,080 --> 00:48:49,000
right here. 
What am I going to write? 

1007
00:48:49,200 --> 00:48:51,800
You know, that's I need to think
about what is the question? 

1008
00:48:51,800 --> 00:48:56,000
So what if I could suggest 
questions automatically by 

1009
00:48:56,000 --> 00:48:59,360
looking at the user's schema, by
looking at the type of chart 

1010
00:48:59,360 --> 00:49:01,720
they want to create? 
So I've coded this super 

1011
00:49:01,720 --> 00:49:04,960
quickly, this little demo that 
generates that natural language 

1012
00:49:04,960 --> 00:49:08,040
queries that then could also be 
translated to me for jqueries. 

1013
00:49:08,280 --> 00:49:11,400
So you kind of have this 
automated question generator for

1014
00:49:11,800 --> 00:49:13,320
for graphs. 
Yeah, yeah. 

1015
00:49:13,480 --> 00:49:15,760
Is that going to make it in? 
Because that's very much like 

1016
00:49:15,760 --> 00:49:18,840
it's very tangible to go then 
from this idea that I have, 

1017
00:49:18,840 --> 00:49:22,080
look, here's how it works to 
making it even in the product at

1018
00:49:22,080 --> 00:49:23,040
the end. 
Let's see. 

1019
00:49:23,040 --> 00:49:24,880
Yeah, let's see. 
I hope it makes it in. 

1020
00:49:24,880 --> 00:49:28,920
I think it's, I don't know it it
as an engineer, it's always fun 

1021
00:49:28,920 --> 00:49:31,560
to do these things. 
But yeah, it does help you think

1022
00:49:31,560 --> 00:49:32,840
and it helps you reach your 
limitations. 

1023
00:49:32,840 --> 00:49:34,880
Like OK, maybe I should have 
shorter questions. 

1024
00:49:34,880 --> 00:49:37,960
Maybe I should look forward 
questions, check in at two 

1025
00:49:37,960 --> 00:49:40,760
months and then see if it's in 
the product and try it out. 

1026
00:49:41,800 --> 00:49:45,200
Yeah, what about you post call? 
What have I coded or how do I 

1027
00:49:45,200 --> 00:49:46,160
learn? 
How do you learn? 

1028
00:49:49,280 --> 00:49:54,960
I think I actually have the the 
inverse problem, so I'm very 

1029
00:49:54,960 --> 00:49:58,720
curious about just trying all 
sorts of things, but I realize I

1030
00:49:58,720 --> 00:50:02,440
can't spend my whole day trying 
things because I'm being paid to

1031
00:50:02,440 --> 00:50:05,600
run a team and, you know, lead 
engineer. 

1032
00:50:05,600 --> 00:50:08,080
Do the job. 
So what I want to make sure as I

1033
00:50:08,080 --> 00:50:12,960
time box my kind of curiosity 
journeys to a certain amount of 

1034
00:50:12,960 --> 00:50:16,360
time where I say, OK, I allow 
myself 10 percent, 15% of my 

1035
00:50:16,360 --> 00:50:19,640
time budget or say, and then 
whatever I do after work hours 

1036
00:50:20,360 --> 00:50:23,240
that sometimes that goes out of 
hand, but you know, whatever. 

1037
00:50:25,120 --> 00:50:27,760
And then within those boxes I 
just follow whatever I feel 

1038
00:50:27,760 --> 00:50:31,680
like. 
I think you can say I'm going to

1039
00:50:31,680 --> 00:50:34,800
be really structured about it 
and I'm going to do this. 

1040
00:50:35,600 --> 00:50:38,760
I'm going to build an agenda of 
my learning journey and I'm 

1041
00:50:38,760 --> 00:50:40,560
going to do it like in a 
university, but I hate it. 

1042
00:50:40,600 --> 00:50:42,200
I hate it. 
Structured learning. 

1043
00:50:43,600 --> 00:50:47,960
I much rather go off into 
letting my curiosity be my 

1044
00:50:47,960 --> 00:50:51,960
guide, and then I just hope that
I'm curious about things that on

1045
00:50:51,960 --> 00:50:54,520
average, other people are also 
curious about. 

1046
00:50:54,680 --> 00:50:56,440
Statistically speaking, it's 
likely right. 

1047
00:50:56,760 --> 00:50:58,880
I'm most likely going to be 
interested in something that 

1048
00:50:58,880 --> 00:51:02,400
other people are also interested
in, because it's very unlikely 

1049
00:51:02,760 --> 00:51:06,760
that I'm in the 0.001% of people
that only like this one weird 

1050
00:51:06,760 --> 00:51:09,320
little niche corner. 
So I'm going to try and learn, 

1051
00:51:09,360 --> 00:51:11,480
or I'm going to be interested in
a topic that other shows find 

1052
00:51:11,480 --> 00:51:12,760
interesting. 
I'm going to learn about it. 

1053
00:51:12,760 --> 00:51:15,720
I'm driven by my intrinsic 
curiosity, so I'm going to go 

1054
00:51:15,720 --> 00:51:17,280
deeper and I'm going to have 
more fun on it. 

1055
00:51:17,280 --> 00:51:20,800
Time's going to go fast, 
quickly, and I'll learn more 

1056
00:51:20,840 --> 00:51:24,560
quickly. 
Yeah, I think learning you 

1057
00:51:24,560 --> 00:51:30,200
shouldn't overthink in a way of 
this is a job, do your job. 

1058
00:51:30,200 --> 00:51:33,680
And then just learn whatever you
find cool. 

1059
00:51:33,680 --> 00:51:37,920
Like right now I find multi 
criteria decision analysis and 

1060
00:51:37,920 --> 00:51:39,640
stage gate modelling really 
interesting. 

1061
00:51:40,080 --> 00:51:42,600
So I'm trying to think of can 
you model a lot of business 

1062
00:51:42,600 --> 00:51:48,920
problem as like a multi stage 
gate system with MCDA ranking at

1063
00:51:48,920 --> 00:51:52,200
each level? 
I don't know why I find that 

1064
00:51:52,200 --> 00:51:53,600
interesting. 
Just an article somewhere 

1065
00:51:53,600 --> 00:51:56,040
random. 
Yeah, I think it reminds me of 

1066
00:51:56,040 --> 00:52:00,320
decision theory in university. 
And I think it's what we do is 

1067
00:52:00,320 --> 00:52:02,320
we rank drugs, right? 
So we have to figure out which 

1068
00:52:02,320 --> 00:52:03,880
of the 60 million do we want to 
do. 

1069
00:52:04,440 --> 00:52:07,160
But I saw a podcast you had 
where you were talking about 

1070
00:52:08,400 --> 00:52:12,040
people asking how do I get like 
a high paying remote job from 

1071
00:52:12,040 --> 00:52:13,720
the US and how do I get in 
there? 

1072
00:52:14,600 --> 00:52:17,040
I think another big one in 
Europe is where do I live? 

1073
00:52:17,080 --> 00:52:19,400
People constantly I'm in this 
bubble where everyone talks 

1074
00:52:19,400 --> 00:52:21,080
about which is the best city to 
live in. 

1075
00:52:21,600 --> 00:52:24,240
London is Amsterdam is a Paris. 
Should I move to New York? 

1076
00:52:24,240 --> 00:52:27,920
Should I try the Asia? 
So this is a ranking problem. 

1077
00:52:27,920 --> 00:52:31,200
Which is the best city for my 
unique situation? 

1078
00:52:31,200 --> 00:52:35,080
What is my criteria grid and 
what is the weight of each 

1079
00:52:35,080 --> 00:52:37,880
criterion? 
And So what is my personalized 

1080
00:52:37,880 --> 00:52:40,520
ranking for the all options of 
all cities? 

1081
00:52:41,040 --> 00:52:44,800
So that's a multi criteria 
decision problem and there's 

1082
00:52:44,800 --> 00:52:47,080
many, many like it. 
What is the stock that a company

1083
00:52:47,080 --> 00:52:49,680
should invest in? 
What is a compound that a 

1084
00:52:50,120 --> 00:52:52,960
pharmaceutical company should 
put more money into and which 

1085
00:52:52,960 --> 00:52:56,280
should they kill? 
We have the same problem, so I 

1086
00:52:56,280 --> 00:52:58,520
find the class of problems 
interesting, and so I'm reading 

1087
00:52:58,520 --> 00:53:01,960
up in the literature. 
If I have to force myself to 

1088
00:53:01,960 --> 00:53:04,440
read that literature because 
it's on my schedule now, I would

1089
00:53:04,440 --> 00:53:06,720
not read it, no. 
I feel like if I ever need like 

1090
00:53:06,720 --> 00:53:10,520
a, a project or like I'm not 
aware of any problems that I can

1091
00:53:10,520 --> 00:53:11,440
solve. 
I just need to have a 

1092
00:53:11,440 --> 00:53:13,960
conversation with you because I 
feel like you just, you just 

1093
00:53:13,960 --> 00:53:18,120
have many already, the one where
you know what's for my use case,

1094
00:53:18,120 --> 00:53:20,040
the best city to live in. 
I think that's fascinating. 

1095
00:53:20,040 --> 00:53:22,640
I don't know if there's a tool 
out there somewhere, but I feel 

1096
00:53:22,640 --> 00:53:24,240
like that solves the problem 
that many people have. 

1097
00:53:25,000 --> 00:53:26,600
That's a funny one. 
I think once you're out of high 

1098
00:53:26,600 --> 00:53:31,880
school or university, you have 
this, this possibility now to 

1099
00:53:31,880 --> 00:53:34,640
guide your own learning journey.
We were all forced to follow an 

1100
00:53:34,640 --> 00:53:38,400
agenda where 9th grade, we have 
to learn about Napoleon. 

1101
00:53:38,960 --> 00:53:42,400
You don't. 
Who says every 9th grader in 

1102
00:53:42,400 --> 00:53:45,640
September is interested in 
Napoleon at that point in time 

1103
00:53:45,640 --> 00:53:47,320
in their life, in their unique 
journey? 

1104
00:53:47,440 --> 00:53:48,720
Doesn't matter. 
It's on the agenda. 

1105
00:53:49,440 --> 00:53:50,400
General knowledge. 
Yeah. 

1106
00:53:50,520 --> 00:53:53,800
As an adult now, you get to just
do whatever you want. 

1107
00:53:54,360 --> 00:53:57,840
Just make sure you time box it 
so that you also get to still do

1108
00:53:57,840 --> 00:53:59,400
the things that other people 
expect from you. 

1109
00:53:59,560 --> 00:54:01,840
Yeah, yeah, I love that. 
Cool man. 

1110
00:54:01,960 --> 00:54:03,640
Thanks, guys. 
This was a lot of fun. 

1111
00:54:04,080 --> 00:54:06,040
I think it kind of kind of went 
everywhere. 

1112
00:54:06,040 --> 00:54:08,280
I do have a better picture of 
graph databases, knowledge 

1113
00:54:08,280 --> 00:54:09,640
graphs. 
I'm definitely going to have 

1114
00:54:09,640 --> 00:54:12,720
many conversations, I think. 
I hope also after the show see 

1115
00:54:12,720 --> 00:54:14,520
if I can build some stuff with 
the rest of that because I think

1116
00:54:14,520 --> 00:54:17,360
it's fascinating. 
It wasn't necessarily the most 

1117
00:54:17,360 --> 00:54:19,360
structured. 
These are all three topics 

1118
00:54:19,360 --> 00:54:21,720
conversation, but that's what 
graphs are all about. 

1119
00:54:21,760 --> 00:54:24,600
They're they're network, they're
messy, they're kind of chaotic. 

1120
00:54:24,920 --> 00:54:27,520
Good stuff, good mess. 
Then we're going to round it off

1121
00:54:27,520 --> 00:54:29,120
here. 
If you're still listening, leave

1122
00:54:29,120 --> 00:54:30,440
a like. 
If you like the episode, they're

1123
00:54:30,440 --> 00:54:32,080
free. 
They only take a second and 

1124
00:54:32,080 --> 00:54:33,480
otherwise we'll see you in the 
next one.