Meeting Title: Hannah <> Casie: n8n Case Study Date: 2025-09-04 Meeting participants: Casie Aviles, Hannah Wang


WEBVTT

1 00:00:28.960 00:00:29.770 Casie Aviles: Hey.

2 00:00:30.790 00:00:32.190 Hannah Wang: Hi, how’s it going?

3 00:00:33.760 00:00:34.679 Casie Aviles: It’s doing good.

4 00:00:34.980 00:00:35.810 Casie Aviles: I love this.

5 00:00:37.400 00:00:45.120 Hannah Wang: Good. I have a quick question. Is my mic okay? Like, my audio? Is it fuzzy, or is it clear?

6 00:00:45.120 00:00:46.150 Casie Aviles: Yeah, it’s clear.

7 00:00:46.500 00:00:53.360 Hannah Wang: Oh, okay. I was having issues yesterday, so… that was weird, but…

8 00:00:54.030 00:00:57.449 Hannah Wang: Okay, so… I feel like we’ve done…

9 00:00:57.600 00:01:07.250 Hannah Wang: a lot of these, which is good. So, it’ll be the same, structure and format. So, let’s get started with…

10 00:01:07.800 00:01:13.160 Hannah Wang: The NAN case study,

11 00:01:13.940 00:01:18.810 Hannah Wang: Yeah, UTAM… well, usually for case studies.

12 00:01:19.140 00:01:26.419 Hannah Wang: I get, like, a clear, oh, this is the project that we want to demo and make a case study for, but I think.

13 00:01:26.910 00:01:29.610 Hannah Wang: We just need something N8N.

14 00:01:30.000 00:01:32.560 Hannah Wang: Related and focused, so…

15 00:01:32.810 00:01:41.509 Hannah Wang: Yeah, were you able to think about a case study or, like, a project that you did that was, like, primarily NAN?

16 00:01:41.700 00:01:46.070 Hannah Wang: like, heavy on NAN, and… Stuff like that.

17 00:01:47.560 00:01:51.470 Casie Aviles: Yeah, I think, I think we can do…

18 00:01:52.270 00:01:55.600 Casie Aviles: ABC there, because we use NHN.

19 00:01:56.210 00:02:04.709 Casie Aviles: For ABC, so… I can… Walk through the… The workflow that we have.

20 00:02:06.200 00:02:13.049 Hannah Wang: Yeah, so before we start that, I… we already have an ABC.

21 00:02:13.050 00:02:14.229 Casie Aviles: Oh, okay.

22 00:02:14.230 00:02:18.009 Hannah Wang: study, but it’s for Andy, so I don’t know if this is different.

23 00:02:18.910 00:02:21.479 Casie Aviles: Yeah, it’s going to be the same.

24 00:02:22.620 00:02:23.600 Hannah Wang: Okay.

25 00:02:23.800 00:02:24.530 Hannah Wang: Okay.

26 00:02:25.880 00:02:26.990 Hannah Wang: Let me think.

27 00:02:39.140 00:02:40.950 Hannah Wang: So this is for Andy, right?

28 00:02:42.130 00:02:45.950 Casie Aviles: Yes, what else can I…

29 00:02:46.080 00:02:53.850 Casie Aviles: So, sorry, I thought it was going to be, like… Talking about the tool, so…

30 00:02:55.090 00:02:57.949 Hannah Wang: I see. I… yeah, I think… I do think…

31 00:02:58.230 00:03:06.719 Hannah Wang: talking about the tool is helpful, but I think probably within the context of client work is probably the best.

32 00:03:07.130 00:03:12.649 Hannah Wang: Use of it.

33 00:03:13.530 00:03:18.559 Casie Aviles: Okay, I think I can do one of the clients that I worked on

34 00:03:19.210 00:03:21.560 Casie Aviles: Which is… off the record.

35 00:03:22.980 00:03:31.630 Hannah Wang: Okay, let me see… We already have… let’s see, is it… so, I already have case studies for off the record.

36 00:03:31.630 00:03:32.040 Casie Aviles: Wonderful.

37 00:03:32.040 00:03:38.590 Hannah Wang: Or a query agent, and then… One for CRM automation.

38 00:03:38.730 00:03:40.990 Hannah Wang: Were you gonna talk about those already?

39 00:03:43.690 00:03:50.559 Casie Aviles: Yeah, if we have a case study there, then yeah, that’s pretty much what I was going to talk about for off the record.

40 00:03:51.490 00:03:54.539 Hannah Wang: Let me try to think.

41 00:03:54.760 00:04:01.490 Hannah Wang: I mean, like, for both of those case studies, for both ABC and Off the Record, like, it’s not…

42 00:04:02.190 00:04:07.150 Hannah Wang: like… I just mentioned…

43 00:04:07.970 00:04:16.480 Hannah Wang: N-A-N, like, and there’s, like, a section for tools used, and I just, like, slapped the logo on there, but, like.

44 00:04:16.829 00:04:21.330 Hannah Wang: Within the actual… case study, like, I didn’t…

45 00:04:22.450 00:04:30.309 Hannah Wang: Explicitly go into detail, like, oh, this is how we used N-A-N.

46 00:04:31.280 00:04:32.630 Casie Aviles: Hmm, okay.

47 00:04:33.290 00:04:41.059 Hannah Wang: Yeah, so maybe… It might be worth it to… it’s okay if, like, we redo…

48 00:04:41.330 00:04:45.169 Hannah Wang: The same case study, but maybe this time, like, you can just…

49 00:04:45.620 00:04:52.950 Hannah Wang: heavily emphasize NAN, because, for example, for Andy, like, it’s very high level.

50 00:04:55.530 00:04:56.250 Casie Aviles: Okay.

51 00:04:57.840 00:05:04.329 Casie Aviles: Yeah. Did we also… have we also done, like, the client hub agents that we have?

52 00:05:06.020 00:05:12.969 Hannah Wang: Client hub agent… could you give me an example of what those are?

53 00:05:13.800 00:05:20.390 Casie Aviles: So these are agents that… That, like, have… access to Slack.

54 00:05:21.240 00:05:25.220 Casie Aviles: Slack messages and Zoom meeting recordings.

55 00:05:27.050 00:05:29.530 Hannah Wang: I see, like the Zoom meeting platform.

56 00:05:30.160 00:05:31.260 Hannah Wang: And…

57 00:05:32.370 00:05:42.820 Casie Aviles: Yeah, it’s kind of like that, but this one is more like, consolidating… plant-specific, info.

58 00:05:44.740 00:05:49.859 Hannah Wang: Could you, like, just demo it for me, real quick? Or, like, show me what it looks like?

59 00:05:50.970 00:05:59.830 Casie Aviles: Yes, so… we have some… I believe we have some… Tests…

60 00:06:01.060 00:06:03.669 Casie Aviles: Yeah, here, like, going stuff I use.

61 00:06:05.300 00:06:06.480 Hannah Wang: Oh…

62 00:06:07.470 00:06:14.960 Casie Aviles: So, like, for example, I would ask… I think default.

63 00:06:15.870 00:06:19.130 Casie Aviles: What were the… Passed.

64 00:06:19.860 00:06:25.070 Casie Aviles: Or, like… Can you give… Can you assign me off of it.

65 00:06:28.320 00:06:29.950 Hannah Wang: I see. Something like that.

66 00:06:30.290 00:06:31.110 Hannah Wang: Okay.

67 00:06:31.330 00:06:34.320 Hannah Wang: I don’t think we did that yet.

68 00:06:34.820 00:06:45.050 Hannah Wang: Yeah, I only have, like… for internal projects, I only have… like…

69 00:06:45.230 00:06:52.600 Hannah Wang: lead… enriching lead automation, Zoom meeting platform… MQS… .

70 00:06:53.230 00:06:53.950 Casie Aviles: Okay.

71 00:06:54.280 00:06:57.040 Casie Aviles: Yeah, we also did something like this.

72 00:06:58.810 00:06:59.590 Hannah Wang: Okay.

73 00:07:01.070 00:07:03.879 Hannah Wang: And is it, like, helpful and being used?

74 00:07:04.110 00:07:08.130 Hannah Wang: within… The teams or, like, clients?

75 00:07:08.330 00:07:10.529 Hannah Wang: Or is this just internal for us?

76 00:07:11.260 00:07:13.380 Casie Aviles: Yeah, this is only internally,

77 00:07:14.840 00:07:19.369 Casie Aviles: Adoption is not yet, like, it’s not being widely used.

78 00:07:19.980 00:07:23.130 Casie Aviles: But we do have it on the platform as well.

79 00:07:23.710 00:07:24.690 Casie Aviles: And…

80 00:07:26.100 00:07:34.409 Casie Aviles: Yeah, like, basically it’s for, like… the idea is to help the team, like, if they have questions about

81 00:07:35.390 00:07:39.610 Casie Aviles: You know, if we have to onboard someone to a new client, and then…

82 00:07:39.740 00:07:42.289 Casie Aviles: Ideally, the client hub agent would help.

83 00:07:43.110 00:07:45.490 Casie Aviles: Onboard them much faster.

84 00:07:47.260 00:07:48.030 Hannah Wang: Okay.

85 00:07:48.420 00:07:51.499 Hannah Wang: Sure, let’s… Let’s do that.

86 00:07:52.290 00:08:00.189 Hannah Wang: Sorry, let me… give me one second. I’m forgetting what…

87 00:08:01.040 00:08:07.100 Hannah Wang: Our next case study is gonna be about… oh, it’s measuring AI agents, right? That’s our…

88 00:08:07.100 00:08:07.790 Casie Aviles: Yes.

89 00:08:07.960 00:08:10.180 Casie Aviles: Or the dash, I believe the dashboard.

90 00:08:10.690 00:08:26.120 Hannah Wang: Yeah, the ABC dashboard. Okay, yeah. Alright, so for this one, we can focus on NAN. Okay, so I think we can try the client hub, and then we’ll just see, like, what Utam says. So, let me pull up the questions.

91 00:08:28.980 00:08:39.350 Hannah Wang: Okay, so I guess first question is, how long did this project take? I know it’s still, like, in progress, but how long have you guys been working on it for?

92 00:08:41.350 00:08:44.260 Casie Aviles: I believe it’s been… Oh, well, and I think…

93 00:08:45.670 00:08:50.739 Casie Aviles: I don’t really remember, like, the first time we did this anymore, but I think it was,

94 00:08:51.110 00:08:52.410 Casie Aviles: Around March?

95 00:08:53.160 00:08:53.900 Hannah Wang: Okay

96 00:08:56.260 00:09:06.680 Hannah Wang: Okay, and then who primarily worked on this? I… I know Miguel probably worked on it, but did you also work on it, too?

97 00:09:07.530 00:09:12.330 Casie Aviles: Yeah, so it’s, yeah, it was mostly Miguel at first.

98 00:09:12.680 00:09:18.529 Casie Aviles: And then I also started to come in and work on…

99 00:09:18.820 00:09:23.469 Casie Aviles: integrating the agents here on Slack, like, adding them to Slack.

100 00:09:25.290 00:09:27.759 Casie Aviles: And then Miguel worked primarily on…

101 00:09:29.370 00:09:38.670 Casie Aviles: I’m, like, getting the knowledge base for… yeah, for this… yeah, getting the knowledge base working for these agents, so…

102 00:09:40.400 00:09:45.720 Casie Aviles: That’s where he did some RAG for these agents.

103 00:09:46.630 00:09:47.400 Hannah Wang: Okay.

104 00:09:47.820 00:09:56.560 Hannah Wang: Okay, cool. So, let’s go over the context next. So, I guess before…

105 00:09:57.340 00:10:06.679 Hannah Wang: creating these client agents. What… was the environment like? Like…

106 00:10:06.980 00:10:10.700 Hannah Wang: Or what was, like, the primary…

107 00:10:11.520 00:10:18.499 Hannah Wang: Like, issue that made you guys want to start developing this tool, or the… the agent?

108 00:10:19.190 00:10:21.990 Casie Aviles: Yeah, so I think kind of similar to…

109 00:10:22.220 00:10:25.419 Casie Aviles: what I mentioned before for the Zoom platform.

110 00:10:27.100 00:10:34.080 Casie Aviles: Sorry, so the idea is that, there are some questions, like, Client-specific questions that…

111 00:10:36.100 00:10:38.650 Casie Aviles: That are some… are being asked, and…

112 00:10:39.220 00:10:41.940 Casie Aviles: You know, sometimes they’re not getting answered.

113 00:10:43.910 00:10:50.159 Casie Aviles: Or, like, it’s kind of hard to… so it’s… yeah, it’s kind of hard to look for the information, so it’s similar…

114 00:10:50.440 00:10:57.340 Casie Aviles: idea behind it, behind the platform, where we want to consolidate the the information, so…

115 00:10:58.200 00:11:05.580 Casie Aviles: Like, before our… the plan was to consolidate the information from

116 00:11:06.060 00:11:11.040 Casie Aviles: not just Slack and Zoom, but also GitHub, and then…

117 00:11:13.090 00:11:18.109 Casie Aviles: What else? Notion, yeah, and… Even emails, so…

118 00:11:19.230 00:11:22.889 Casie Aviles: That that was the idea, like, we wanted to get everything there.

119 00:11:23.540 00:11:27.240 Casie Aviles: So the agent would be able to answer questions.

120 00:11:27.570 00:11:28.910 Casie Aviles: Much easily.

121 00:11:30.180 00:11:33.120 Casie Aviles: So yeah, that was the idea behind it.

122 00:11:34.140 00:11:39.190 Hannah Wang: But currently, it only pulls info… info from Slack.

123 00:11:41.120 00:11:45.550 Casie Aviles: Yeah, right now it’s just those.

124 00:11:46.690 00:11:53.670 Casie Aviles: We’ve… yeah, we have. In the past, we’ve tried to work on the other data sources, but it was kind of…

125 00:11:53.910 00:11:55.540 Casie Aviles: It… it got… it got…

126 00:11:55.950 00:12:00.879 Casie Aviles: Difficult quickly, because, you know, it was… there are a lot of sources, it’s kind of…

127 00:12:01.200 00:12:03.490 Casie Aviles: Hard to make sure that the bot is…

128 00:12:03.900 00:12:06.840 Casie Aviles: Pulling from the right sources, so that’s where we.

129 00:12:06.840 00:12:07.300 Hannah Wang: Yeah.

130 00:12:07.450 00:12:09.649 Casie Aviles: That’s a challenge that we face then.

131 00:12:11.070 00:12:21.149 Hannah Wang: Okay, and it’s still… like, are there plans to try to do that again? Like, integrate with Notion and email, or is it still gonna be Slack and Zoom?

132 00:12:21.150 00:12:24.050 Casie Aviles: Yeah, I think there’s still,

133 00:12:24.590 00:12:29.110 Casie Aviles: I… we still want to do that, because, there’s…

134 00:12:29.500 00:12:34.740 Casie Aviles: like, Slap and Zoom is just… Like, it’s not…

135 00:12:35.630 00:12:41.070 Casie Aviles: what do you call this? The context is not just there, we also have…

136 00:12:41.690 00:12:47.120 Casie Aviles: Ideally, we also want to have Notion, because we have a lot of documentation there, like.

137 00:12:48.050 00:12:52.050 Casie Aviles: So we want to be able to get those as well.

138 00:12:53.330 00:12:55.020 Hannah Wang: Okay, yes.

139 00:12:55.020 00:12:55.620 Casie Aviles: Yeah.

140 00:12:55.730 00:12:56.719 Casie Aviles: Thank you very much.

141 00:12:57.650 00:13:03.310 Hannah Wang: Okay, and I guess before these agents were made,

142 00:13:04.560 00:13:15.410 Hannah Wang: like, basically, one person who had the context… like, if someone was asking questions, like, if there was a new person onboarding to a client, I’m assuming before…

143 00:13:15.690 00:13:31.499 Hannah Wang: this agent, like, either Utam or whoever was, like, PMing, the client would have to, like, type out or, like, send a bunch of documentation to the new person onboarding, and that was just, like, really time-consuming, right?

144 00:13:31.500 00:13:31.829 Casie Aviles: Thank you.

145 00:13:31.830 00:13:34.349 Hannah Wang: Kind of makes it easier for…

146 00:13:34.710 00:13:40.410 Hannah Wang: it to be, like, a bot to answer those questions instead of, like, UTAM, for example.

147 00:13:40.590 00:13:41.990 Casie Aviles: Yeah, exactly, Lynn.

148 00:13:42.180 00:13:51.420 Casie Aviles: That’s the main pain point that we were trying to address, where, you know, someone does not become a bottleneck for…

149 00:13:52.670 00:13:53.690 Casie Aviles: For this…

150 00:13:55.350 00:13:56.910 Hannah Wang: Yeah, that makes sense.

151 00:14:02.050 00:14:03.610 Casie Aviles: Yeah, I’m just… Jeez.

152 00:14:10.190 00:14:14.739 Casie Aviles: Yeah, okay, yeah, hmm, anything else?

153 00:14:16.970 00:14:22.959 Hannah Wang: Okay, so that’s the context, and I guess you kind of answered the challenge already.

154 00:14:23.240 00:14:29.470 Hannah Wang: It’s just that people were being bottlenecks, basically, and it was hard to…

155 00:14:29.940 00:14:38.299 Hannah Wang: answer client-specific questions, to new people, to the… to people who are new to the client, and that just, like, slowed the team down.

156 00:14:38.690 00:14:39.780 Hannah Wang: Mmm.

157 00:14:40.660 00:14:44.700 Hannah Wang: Okay, so getting into the solution…

158 00:14:44.840 00:14:52.790 Hannah Wang: here’s where we can make N8N shine. So, I guess, what was the solution and the setup and infra?

159 00:14:53.900 00:14:55.210 Casie Aviles: Okay.

160 00:14:55.940 00:14:59.710 Casie Aviles: So, I can show one of the workflows.

161 00:15:03.240 00:15:05.380 Casie Aviles: Let me just look for one of them.

162 00:15:06.040 00:15:06.820 Casie Aviles: Bye.

163 00:15:09.660 00:15:14.110 Casie Aviles: So, we have this, workflow,

164 00:15:14.560 00:15:20.820 Casie Aviles: So I’m not gonna go through each node, but… The idea here is,

165 00:15:22.180 00:15:29.309 Casie Aviles: this is, like, the core of it, like, this is the AI agent that’s… receiving the input.

166 00:15:30.410 00:15:33.620 Casie Aviles: Performing actions and generating output.

167 00:15:36.070 00:15:41.880 Casie Aviles: So… The good thing about N810 is it already comes built

168 00:15:42.040 00:15:44.859 Casie Aviles: In with a lot of integrations, so…

169 00:15:45.500 00:15:49.080 Casie Aviles: As opposed to, like, if we were to do this…

170 00:15:49.500 00:15:53.320 Casie Aviles: like, I guess from scratch, or, like, completely from code.

171 00:15:54.230 00:15:59.489 Casie Aviles: We’d have to implement our own connectors or integrations to Slack.

172 00:15:59.880 00:16:02.960 Casie Aviles: So that’s definitely gonna be more time-consuming.

173 00:16:03.250 00:16:09.829 Casie Aviles: And… also, like, There’s going to be a lot of work to be done for that.

174 00:16:09.960 00:16:10.700 Casie Aviles: And…

175 00:16:11.300 00:16:19.370 Casie Aviles: What’s good about NA10 is these integrations are already available for us, so we just… they come out of the box, so…

176 00:16:19.550 00:16:26.410 Casie Aviles: We just have to add a Slack node, and just set the… I guess the credentials to…

177 00:16:27.330 00:16:31.450 Casie Aviles: To our… to one of our accounts, and then, yeah, that should be good to go, and…

178 00:16:31.880 00:16:35.670 Casie Aviles: So that’s one good thing, and

179 00:16:35.910 00:16:37.519 Casie Aviles: So the idea is we want

180 00:16:37.830 00:16:41.100 Casie Aviles: the AI agents to live where we do the work.

181 00:16:41.840 00:16:45.660 Casie Aviles: So, wherever the team is, most…

182 00:16:45.920 00:16:50.000 Casie Aviles: Or, like, where we spend time the most, so, which is probably on Slack.

183 00:16:50.460 00:16:54.530 Casie Aviles: So that’s the idea why we wanted to have it on Slack, and then…

184 00:16:55.920 00:17:01.549 Casie Aviles: Additionally, since we developed our, the internal platform that we have now.

185 00:17:03.600 00:17:10.640 Casie Aviles: And A10 also, makes it easier for us to integrate it to our own front end because of.

186 00:17:11.720 00:17:15.330 Casie Aviles: The triggers that… that come with N8N.

187 00:17:15.480 00:17:26.019 Casie Aviles: So we can… activate this agent, via webhooks, so… Essentially, we’re just,

188 00:17:26.579 00:17:29.339 Casie Aviles: Yeah, we’re just sending out, requests.

189 00:17:29.480 00:17:35.840 Casie Aviles: And… which is… which are received by this, workflow, and then the agent will just…

190 00:17:35.970 00:17:40.150 Casie Aviles: Generate an output, so that’s what makes it… convenient.

191 00:17:40.270 00:17:46.979 Casie Aviles: So we could do something, yeah, so the agent setup is not the most…

192 00:17:47.680 00:17:53.870 Casie Aviles: challenging part, I guess, it would be the… the vectorization.

193 00:17:54.560 00:18:01.079 Casie Aviles: But yeah, the good thing about… another good thing about NA10 is… It has these…

194 00:18:02.070 00:18:06.709 Casie Aviles: vector database tools, or RAG tools.

195 00:18:06.950 00:18:11.640 Casie Aviles: So… We also don’t have to implement our own I mean…

196 00:18:12.780 00:18:15.480 Casie Aviles: At least for, like, the retrieval.

197 00:18:15.630 00:18:18.260 Casie Aviles: part. We don’t have to implement that.

198 00:18:18.770 00:18:23.950 Casie Aviles: We just use Supabase, which is another integration that comes with N80.

199 00:18:24.560 00:18:32.729 Casie Aviles: And then we just… Plug in these models, so… Yeah, I think that’s… I think that’s what…

200 00:18:34.170 00:18:39.989 Casie Aviles: That’s what… those are the good things about N80 that made this possible for us.

201 00:18:42.510 00:18:48.919 Hannah Wang: Maybe I should have asked this before we started talking, but what exactly is N8N?

202 00:18:49.080 00:18:56.900 Casie Aviles: Oh, okay. Yeah, so NA10 is, like, an agent builder, but… or a workflow builder as well, like, you know.

203 00:18:57.240 00:19:07.030 Casie Aviles: So… It’s… it’s kind of… it lets us do, like, automations in a node-based manner, so…

204 00:19:07.730 00:19:13.740 Casie Aviles: Basically, we have these… Little nodes here that we can string together.

205 00:19:14.070 00:19:19.509 Casie Aviles: And… It lets us build automation, so…

206 00:19:19.640 00:19:27.070 Casie Aviles: And additionally, we can also build, like, workflows where AI could be added.

207 00:19:27.560 00:19:28.710 Casie Aviles: So…

208 00:19:29.500 00:19:36.310 Casie Aviles: Yeah, if we want… if we have steps that need to be AI-generated, like, if we want to automate

209 00:19:36.650 00:19:39.969 Casie Aviles: email generation.

210 00:19:40.230 00:19:48.040 Casie Aviles: or, like, email draft generation, then we could have, like, a step, an LLM step, that We’ll generate the…

211 00:19:49.150 00:19:53.329 Casie Aviles: the draft for us. So, yeah, that’s what pretty much…

212 00:19:53.730 00:19:58.180 Casie Aviles: N8N is. It’s like a workflow builder, and also a…

213 00:19:59.170 00:20:03.490 Casie Aviles: With AI features and with a lot… with a lot of integrations.

214 00:20:04.460 00:20:06.209 Hannah Wang: Okay. Kind of like Zapier.

215 00:20:07.060 00:20:07.790 Hannah Wang: Got it.

216 00:20:08.100 00:20:12.290 Hannah Wang: Zapier and, I think, Make is another workflow builder, right?

217 00:20:12.420 00:20:13.220 Hannah Wang: those two.

218 00:20:14.820 00:20:29.130 Hannah Wang: Okay, that… I understand that a little more now, because I’ve… I’ve used Make before, so… N-A-N, yeah, it looks similar, it’s just workflow building, basically. Yeah. Okay. And…

219 00:20:29.590 00:20:36.989 Hannah Wang: Perhaps adoption hasn’t been that great, but…

220 00:20:37.260 00:20:37.770 Casie Aviles: Yeah.

221 00:20:37.770 00:20:39.179 Hannah Wang: even, like.

222 00:20:39.670 00:20:50.839 Hannah Wang: with the small amount of usage that we got, I guess, was there any feedback that you guys got for the agent?

223 00:20:50.980 00:20:55.739 Hannah Wang: either from Utam or from Amber, I feel like those are the names.

224 00:20:55.740 00:20:56.430 Casie Aviles: two people.

225 00:20:56.430 00:20:57.499 Hannah Wang: that use it.

226 00:20:58.170 00:21:06.170 Casie Aviles: Yeah, that… yeah, those are, like, the main two people that… that use the client hubs, so…

227 00:21:06.320 00:21:10.180 Casie Aviles: like, the positive feedback, I guess, would be…

228 00:21:10.930 00:21:16.840 Casie Aviles: It’s helped Amber before, like, for ABC, when she was starting out.

229 00:21:17.030 00:21:26.219 Casie Aviles: It’s helped her generate, like, email summaries for the work that we’ve done, and… also asked.

230 00:21:26.560 00:21:33.150 Casie Aviles: Questions, or, you know, things that… She doesn’t have context on…

231 00:21:33.640 00:21:38.200 Casie Aviles: Because she, she was, I think…

232 00:21:38.470 00:21:42.000 Casie Aviles: She came later, to the project.

233 00:21:43.680 00:21:48.049 Casie Aviles: So, me and Miguel were there first, so we built this for her.

234 00:21:48.270 00:21:50.120 Casie Aviles: When she was new, so…

235 00:21:51.520 00:21:54.619 Casie Aviles: Yeah, I think that’s the positive side to it.

236 00:21:56.120 00:22:04.310 Casie Aviles: But then, it’s also… when we built this, it was also very challenging, because we had a lot of,

237 00:22:04.470 00:22:06.020 Casie Aviles: Quality issues.

238 00:22:06.660 00:22:09.169 Casie Aviles: Then, because we haven’t figured out, like.

239 00:22:09.630 00:22:16.450 Casie Aviles: the best way to feed the data to this AI agent

240 00:22:16.850 00:22:20.860 Casie Aviles: we were kind of also learning how to do RAG as we did

241 00:22:21.570 00:22:25.700 Casie Aviles: These client hub agents, so… I guess that’s what…

242 00:22:25.880 00:22:29.620 Casie Aviles: That’s, like, the negative feedback that we got, which is…

243 00:22:30.870 00:22:34.480 Casie Aviles: It was kind of hard to, like, trust this agent.

244 00:22:34.960 00:22:35.800 Casie Aviles: Yeah.

245 00:22:36.370 00:22:44.709 Casie Aviles: Especially since… It doesn’t give the data, sometimes it doesn’t give the data, the info that is…

246 00:22:45.220 00:22:50.520 Casie Aviles: That we’re looking for, and… You’ve got incomplete data, and sometimes it’s just…

247 00:22:50.760 00:22:55.260 Casie Aviles: hallucinating. So we had a lot of technical challenges with

248 00:22:55.620 00:22:59.259 Casie Aviles: Building this, primarily because of, like, the…

249 00:23:00.410 00:23:08.350 Casie Aviles: Yeah, the data part, like, the knowledge base, so… fabulous, yeah.

250 00:23:08.350 00:23:09.899 Hannah Wang: I guess…

251 00:23:10.850 00:23:14.179 Casie Aviles: Yeah, I was just gonna say that was a lot of,

252 00:23:14.940 00:23:19.750 Casie Aviles: iterations, and on… so we were working on this on and off as well, so…

253 00:23:19.750 00:23:20.410 Hannah Wang: Yeah.

254 00:23:21.690 00:23:22.070 Casie Aviles: Yeah.

255 00:23:22.070 00:23:22.690 Hannah Wang: Got it.

256 00:23:23.390 00:23:35.200 Hannah Wang: And, I guess after building, this workflow, did it help you like, build other…

257 00:23:35.840 00:23:43.979 Hannah Wang: tools? Like, did… was it… like, for example, like, I know you integrated this into the Zoom platform, right? So I’m just…

258 00:23:43.980 00:23:44.430 Casie Aviles: Yes.

259 00:23:44.430 00:23:46.950 Hannah Wang: Yeah, asking, like, oh, was this…

260 00:23:47.240 00:23:51.960 Hannah Wang: Building this, like, a helpful foundation for other projects that you guys worked on.

261 00:23:52.440 00:23:54.899 Casie Aviles: Yeah, actually, we did learn a lot from

262 00:23:55.120 00:23:59.310 Casie Aviles: Building the client hub agents, so…

263 00:23:59.580 00:24:03.149 Casie Aviles: There, like, for example, the email…

264 00:24:03.740 00:24:07.930 Casie Aviles: like, summary generation, we decided to just…

265 00:24:08.120 00:24:11.230 Casie Aviles: Create, like, an entirely separate workflow for that.

266 00:24:13.050 00:24:14.060 Casie Aviles: And…

267 00:24:14.290 00:24:24.680 Casie Aviles: Yeah, so that’s what… something that, Mustafa also worked on. But yeah, we kind of specialized it from there.

268 00:24:25.870 00:24:30.959 Casie Aviles: And so, yeah, like, It’s… Ideally, this agent would have done…

269 00:24:31.140 00:24:35.600 Casie Aviles: a lot of the stuff, but we decided to just compartmentalize, I guess.

270 00:24:35.670 00:24:37.140 Hannah Wang: For now.

271 00:24:37.220 00:24:44.219 Casie Aviles: But yeah, also a lot of the challenges we faced building this Has been good.

272 00:24:44.370 00:24:51.259 Casie Aviles: at least a good learning experience. We were able to build, like, other more Complex workflows, for instance,

273 00:24:51.600 00:24:58.269 Casie Aviles: like, with… even with ABC, we… we would, from… based on what we learned here, we were able to, like.

274 00:24:58.680 00:25:01.600 Casie Aviles: Make some improvements to other clients.

275 00:25:02.200 00:25:04.590 Casie Aviles: Existing client workflows that we have.

276 00:25:04.860 00:25:08.269 Casie Aviles: So… Yeah, definitely, I think.

277 00:25:08.400 00:25:15.020 Casie Aviles: Even though the… there were a lot of issues that we faced, we learned a lot here.

278 00:25:15.560 00:25:16.160 Hannah Wang: Hmm.

279 00:25:17.420 00:25:18.330 Hannah Wang: Cool.

280 00:25:18.510 00:25:21.760 Hannah Wang: Okay, I think this is…

281 00:25:21.950 00:25:36.430 Hannah Wang: This is good. We’ll see how the case study comes out. But yeah, I think we can hop on to the next meeting to talk about, measuring agents with the ABC dashboard. So, I’ll see you in that meeting.

282 00:25:37.280 00:25:37.980 Casie Aviles: Okay.

283 00:25:38.460 00:25:39.020 Casie Aviles: Bye-bye.