Meeting Title: ABC Planning Sync Date: 2025-12-09 Meeting participants: Casie Aviles, Mustafa Raja, Samuel Roberts
WEBVTT
1 00:00:26.120 ⇒ 00:00:27.260 Casie Aviles: Hey, Masafa.
2 00:00:27.970 ⇒ 00:00:31.330 Mustafa Raja: Yeah, how are you? Long time no see.
3 00:00:32.320 ⇒ 00:00:38.179 Casie Aviles: Yeah, I think you had… Doing better, I mean…
4 00:00:38.360 ⇒ 00:00:40.660 Casie Aviles: There’s a lot of stuff going on.
5 00:00:41.580 ⇒ 00:00:43.070 Mustafa Raja: Just doing…
6 00:00:43.070 ⇒ 00:00:44.350 Casie Aviles: Doing our best.
7 00:00:45.120 ⇒ 00:00:50.249 Mustafa Raja: Yeah, yeah That’s all we can do, right? Trying out.
8 00:00:50.250 ⇒ 00:00:52.210 Casie Aviles: Yeah. Yep, yep.
9 00:00:52.890 ⇒ 00:00:56.179 Casie Aviles: I’m just curious, by the way,
10 00:00:56.380 ⇒ 00:00:58.879 Casie Aviles: Will you… are you still, like…
11 00:00:59.090 ⇒ 00:01:03.370 Casie Aviles: assigned to work on ABC, or… because I wasn’t sure if, you know, if…
12 00:01:03.370 ⇒ 00:01:04.180 Mustafa Raja: Yes.
13 00:01:04.610 ⇒ 00:01:06.650 Mustafa Raja: I’m assigned 5Rs.
14 00:01:07.310 ⇒ 00:01:08.810 Mustafa Raja: Oh, okay, okay.
15 00:01:09.190 ⇒ 00:01:10.480 Casie Aviles: I see, I see.
16 00:01:10.480 ⇒ 00:01:13.909 Mustafa Raja: Yeah, so feel free to, you know, throw some stuff at me.
17 00:01:14.310 ⇒ 00:01:18.210 Mustafa Raja: Because I… because, I think last month I only…
18 00:01:18.340 ⇒ 00:01:21.380 Mustafa Raja: Who worked 5 hours for the whole month, you know?
19 00:01:21.380 ⇒ 00:01:22.600 Casie Aviles: Oh, really? Oh.
20 00:01:22.600 ⇒ 00:01:24.759 Mustafa Raja: of ours for a week, so… Yeah.
21 00:01:25.960 ⇒ 00:01:28.750 Casie Aviles: Yeah, that’s… that’s… That’s fine.
22 00:01:29.960 ⇒ 00:01:30.930 Mustafa Raja: Yay.
23 00:01:34.630 ⇒ 00:01:35.610 Casie Aviles: Hey, Sean.
24 00:01:40.970 ⇒ 00:01:46.340 Casie Aviles: Sorry, I think… I think you’re… you’re… if you’re talking…
25 00:01:46.340 ⇒ 00:01:46.780 Mustafa Raja: Yeah.
26 00:01:46.780 ⇒ 00:01:48.150 Casie Aviles: We’re not hearing anything.
27 00:01:48.630 ⇒ 00:01:49.130 Mustafa Raja: Yeah.
28 00:01:49.130 ⇒ 00:01:50.420 Samuel Roberts: You guys hear me at all?
29 00:01:50.700 ⇒ 00:01:51.650 Mustafa Raja: Yes.
30 00:01:52.220 ⇒ 00:01:54.779 Samuel Roberts: Now you can? Alright. Okay. Everything.
31 00:01:59.480 ⇒ 00:02:04.159 Casie Aviles: Yeah, okay. So I guess I’ll just start off.
32 00:02:04.760 ⇒ 00:02:05.090 Samuel Roberts: Okay.
33 00:02:05.090 ⇒ 00:02:05.739 Casie Aviles: I think.
34 00:02:06.610 ⇒ 00:02:09.799 Casie Aviles: Yeah, thank you, thank you guys for making the time.
35 00:02:09.800 ⇒ 00:02:10.420 Samuel Roberts: Of course.
36 00:02:10.789 ⇒ 00:02:14.459 Casie Aviles: I wasn’t sure, like, with Mustafa, but yeah, thank you as well for joining.
37 00:02:14.759 ⇒ 00:02:17.189 Casie Aviles: But, but yeah, I think mainly…
38 00:02:18.299 ⇒ 00:02:23.989 Casie Aviles: I think we all know the context behind, like, Andy, from last.
39 00:02:23.990 ⇒ 00:02:24.640 Mustafa Raja: Yeah.
40 00:02:25.210 ⇒ 00:02:34.170 Casie Aviles: So… I guess, like, I just have some more on, like, questions on, like, what’s the best
41 00:02:35.960 ⇒ 00:02:42.130 Casie Aviles: path forward, like, I think that’s where I’m having a little trouble getting some clarity.
42 00:02:42.730 ⇒ 00:02:46.030 Casie Aviles: And kind of… Yeah, I mean…
43 00:02:46.510 ⇒ 00:02:54.030 Casie Aviles: You know, like, for example, I think some… some of the issues, or, like, some of the things that I want to…
44 00:02:54.260 ⇒ 00:03:00.770 Casie Aviles: Clarify and, you know, form a decision with you guys is, like, What, what are, like…
45 00:03:01.350 ⇒ 00:03:07.860 Casie Aviles: You know, long-term and short-term, like, what can… what should we be doing?
46 00:03:08.630 ⇒ 00:03:16.400 Casie Aviles: Should we… should we have, like, a… should we establish, like, a clear… a clear QA process?
47 00:03:16.780 ⇒ 00:03:18.039 Casie Aviles: Stuff like that.
48 00:03:19.130 ⇒ 00:03:27.690 Casie Aviles: And also, I guess… so that’s more of, like, a high-level… Question… That I, I’m…
49 00:03:28.320 ⇒ 00:03:36.930 Casie Aviles: I’m hoping we could, we could also kind of… talk about… What else?
50 00:03:37.270 ⇒ 00:03:44.340 Casie Aviles: Like, yeah, as well, like, I know we have the migration plan, which is probably our answer to the long-term solutions.
51 00:03:44.660 ⇒ 00:03:51.299 Casie Aviles: So… I’m not sure if it’s a question for this call, but there are also stuff like
52 00:03:51.720 ⇒ 00:03:56.350 Casie Aviles: for example, I was just in a call earlier with Janiece and Amber, and
53 00:03:56.500 ⇒ 00:04:00.979 Casie Aviles: It turns out, like, we designed… the database…
54 00:04:01.400 ⇒ 00:04:04.459 Casie Aviles: There’s, like, another… another missing thing.
55 00:04:04.600 ⇒ 00:04:06.139 Casie Aviles: That we just learned of.
56 00:04:06.840 ⇒ 00:04:09.060 Casie Aviles: That we have to do again.
57 00:04:09.610 ⇒ 00:04:13.750 Casie Aviles: So we have to kind of think about like…
58 00:04:14.140 ⇒ 00:04:19.450 Casie Aviles: Updating the database again with new stuff, so… like…
59 00:04:19.450 ⇒ 00:04:24.280 Samuel Roberts: Okay, what is that? Is it more, a different department or something, or…
60 00:04:24.450 ⇒ 00:04:27.100 Casie Aviles: Yeah, yeah, it’s more like that, where…
61 00:04:27.520 ⇒ 00:04:28.780 Samuel Roberts: Okay.
62 00:04:29.630 ⇒ 00:04:35.479 Casie Aviles: You know, like, for example, with… Inspectors, or, like.
63 00:04:35.610 ⇒ 00:04:44.150 Casie Aviles: Technicians, there’s… there’s not, like… For inspectors, we designed it so that they are distinguished between commercial or residential.
64 00:04:44.610 ⇒ 00:04:54.470 Casie Aviles: And then… There’s… but for technicians, we didn’t, and now we’re learning that Oh, they should be distinguished.
65 00:04:54.790 ⇒ 00:05:01.020 Casie Aviles: And that’s kind of… Okay, so we need to work on that, but… like…
66 00:05:01.450 ⇒ 00:05:04.469 Casie Aviles: We’re also working on the migration, so…
67 00:05:05.000 ⇒ 00:05:06.969 Casie Aviles: Should we, like, do that?
68 00:05:07.600 ⇒ 00:05:17.479 Casie Aviles: that’s kind of, like, what I’m trying to think of, like, what should I do first? Should I work on that first, or… but what about the migration plan, you know, like…
69 00:05:18.620 ⇒ 00:05:19.080 Samuel Roberts: Right.
70 00:05:19.080 ⇒ 00:05:20.340 Casie Aviles: Those are kinds of… Okay.
71 00:05:20.460 ⇒ 00:05:21.230 Casie Aviles: Yeah.
72 00:05:24.360 ⇒ 00:05:26.820 Casie Aviles: Yeah, I think… I think that’s… that’s all…
73 00:05:26.920 ⇒ 00:05:29.690 Casie Aviles: I’m not sure how to best run…
74 00:05:29.830 ⇒ 00:05:32.270 Casie Aviles: The call, but those are kind of, like…
75 00:05:33.170 ⇒ 00:05:36.680 Casie Aviles: My questions, or, like, hoping to discuss with you guys.
76 00:05:36.680 ⇒ 00:05:38.249 Samuel Roberts: Yeah. Okay, cool.
77 00:05:39.370 ⇒ 00:05:41.039 Samuel Roberts: Yeah, so let’s just, I mean…
78 00:05:41.380 ⇒ 00:05:44.530 Samuel Roberts: So I got a little bit of the context from last week.
79 00:05:44.620 ⇒ 00:05:45.480 Casie Aviles: But…
80 00:05:45.600 ⇒ 00:05:52.050 Samuel Roberts: Even going back further to, like, the migration plan, I think…
81 00:05:52.190 ⇒ 00:05:58.770 Samuel Roberts: Well, the database is not necessarily something that needs to get touched as quickly, but… I think…
82 00:05:59.350 ⇒ 00:06:07.729 Samuel Roberts: I mean, some of the… well, I guess. You made a… hold on, sorry, let me back up even further. Sure. You posted something on Notion, let me get that open here.
83 00:06:08.790 ⇒ 00:06:12.629 Samuel Roberts: You put, like, a little summary… Issue summary report.
84 00:06:13.220 ⇒ 00:06:17.849 Samuel Roberts: I just wanted to… I took a quick look at it this morning, but I wanted to open it up again.
85 00:06:18.070 ⇒ 00:06:23.230 Casie Aviles: Yeah, this is just a summary, it’s not super comprehensive yet, okay. Yeah.
86 00:06:25.000 ⇒ 00:06:31.050 Samuel Roberts: I guess my… My question here would be… what…
87 00:06:31.790 ⇒ 00:06:37.250 Samuel Roberts: Out of these, like, two, the infrastructure issues and the content things, what do we think are the bigger…
88 00:06:38.030 ⇒ 00:06:43.240 Samuel Roberts: percentage of the problems that we’re seeing. That might help us dictate what we focus on.
89 00:06:45.390 ⇒ 00:06:46.600 Casie Aviles: Hmm, yeah.
90 00:06:47.370 ⇒ 00:06:50.670 Casie Aviles: I don’t have, like, numbers here yet, but…
91 00:06:50.990 ⇒ 00:06:51.780 Samuel Roberts: Sure.
92 00:06:51.780 ⇒ 00:06:57.290 Casie Aviles: I believe, like… There… there’s… there’s definitely a lot more…
93 00:06:58.100 ⇒ 00:07:00.940 Casie Aviles: Issues coming from, like, the content.
94 00:07:00.940 ⇒ 00:07:02.999 Samuel Roberts: Accuracy, the second part here.
95 00:07:03.300 ⇒ 00:07:08.060 Casie Aviles: And… I think that’s also mainly because…
96 00:07:08.590 ⇒ 00:07:13.919 Casie Aviles: Well, their data is… or, like, the data source that they gave us is not very clean.
97 00:07:14.480 ⇒ 00:07:15.740 Samuel Roberts: Yeah. And…
98 00:07:15.740 ⇒ 00:07:19.750 Casie Aviles: That’s also something we have to wrangle or, like, clean up.
99 00:07:20.190 ⇒ 00:07:26.719 Casie Aviles: And… That can either be solved by them going into, you know, updating
100 00:07:26.830 ⇒ 00:07:29.179 Casie Aviles: That’s why we did, like, the update forms.
101 00:07:29.840 ⇒ 00:07:30.290 Samuel Roberts: Right.
102 00:07:30.380 ⇒ 00:07:33.860 Casie Aviles: But then there’s also, like, probably a more…
103 00:07:34.720 ⇒ 00:07:39.650 Casie Aviles: implementation side, where, like, the RAG itself may not be the best
104 00:07:40.060 ⇒ 00:07:43.140 Casie Aviles: And getting the content that it needs to get.
105 00:07:43.560 ⇒ 00:07:49.439 Casie Aviles: So those are, like… but yeah, definitely we have a lot of feedback coming from the second part here.
106 00:07:49.440 ⇒ 00:07:49.840 Samuel Roberts: Okay.
107 00:07:49.840 ⇒ 00:07:50.590 Casie Aviles: DRC.
108 00:07:50.910 ⇒ 00:07:52.720 Casie Aviles: And infrastructure… Okay.
109 00:07:53.050 ⇒ 00:07:53.850 Casie Aviles: Yeah.
110 00:07:54.120 ⇒ 00:08:03.220 Casie Aviles: for infrastructure, I think that’s… Kind of more frustrating, though, because it’s just gonna return Errors and undesirable.
111 00:08:03.220 ⇒ 00:08:03.600 Samuel Roberts: Yeah, right?
112 00:08:03.600 ⇒ 00:08:08.710 Casie Aviles: And that’s actually where we get a lot… we also get, like, more urgent
113 00:08:09.840 ⇒ 00:08:16.880 Casie Aviles: questions. The second one, there’s more volume, but not as urgent, probably. But number… the first part…
114 00:08:17.800 ⇒ 00:08:24.839 Casie Aviles: like, it’s just a bad… I guess it looks bad because it’s going to, like, just error out, spit something that’s…
115 00:08:24.980 ⇒ 00:08:25.819 Casie Aviles: you know.
116 00:08:26.030 ⇒ 00:08:27.250 Samuel Roberts: Right, right.
117 00:08:29.660 ⇒ 00:08:41.919 Samuel Roberts: Okay, yeah, I think… Based on… That, then focusing on the… Migration is probably a good… kind of…
118 00:08:43.260 ⇒ 00:08:48.519 Samuel Roberts: First pass, because it will help us alleviate some of those initial, like.
119 00:08:49.310 ⇒ 00:08:52.160 Samuel Roberts: you know, running out of memory errors, or…
120 00:08:52.160 ⇒ 00:08:52.630 Casie Aviles: YF.
121 00:08:52.630 ⇒ 00:08:53.340 Samuel Roberts: just, like.
122 00:08:53.450 ⇒ 00:09:00.720 Samuel Roberts: things that we don’t get notified for, all kinds of stuff, and then I think we can probably have a little more flexibility to work with the RAG system.
123 00:09:02.960 ⇒ 00:09:07.659 Samuel Roberts: And tie in other tools or whatever else we might need that might make it more robust.
124 00:09:08.460 ⇒ 00:09:15.239 Samuel Roberts: Which can help with the content accuracy. So I think that’s still probably the path forward.
125 00:09:15.410 ⇒ 00:09:18.549 Samuel Roberts: For, like, short and long-term kind of thing.
126 00:09:20.450 ⇒ 00:09:21.210 Casie Aviles: Okay.
127 00:09:22.030 ⇒ 00:09:39.200 Samuel Roberts: I just think, like, getting off of N8… I think we’ve just kind of hit the limit, literally and, kind of figuratively, with what N8N is good for here, because I think you’re right down here, like, we’re just gonna… more correction workflows, more memory pressure, more crashes, more update… yeah. It just… it becomes a, yeah, bad…
128 00:09:39.630 ⇒ 00:09:41.579 Samuel Roberts: Bad path in general, so…
129 00:09:43.140 ⇒ 00:09:48.460 Samuel Roberts: So I think at least the first thing is to get the logic off of there.
130 00:09:51.080 ⇒ 00:09:51.800 Casie Aviles: Okay.
131 00:09:52.680 ⇒ 00:09:54.799 Samuel Roberts: I think the database stuff…
132 00:09:55.240 ⇒ 00:10:00.149 Samuel Roberts: Can be happening along with that a little bit, because we can still make use of the…
133 00:10:01.180 ⇒ 00:10:04.619 Samuel Roberts: like, data that’s there. We don’t necessarily have to get off of Superbase
134 00:10:04.740 ⇒ 00:10:07.829 Samuel Roberts: As part of the initial step of the migration, right?
135 00:10:08.470 ⇒ 00:10:10.170 Samuel Roberts: Yeah, yeah, we don’t have suite yet.
136 00:10:11.040 ⇒ 00:10:20.779 Samuel Roberts: So, I think, you know, maybe eventually we need to move to something, you know, more GCP for them, but for now, that’s a part that we can still just access and update.
137 00:10:21.290 ⇒ 00:10:27.400 Samuel Roberts: And maybe that means having to change the data model a little bit, but…
138 00:10:28.490 ⇒ 00:10:31.559 Samuel Roberts: Making it a little more flexible in general is probably smart.
139 00:10:32.840 ⇒ 00:10:35.110 Samuel Roberts: Because if there’s other things that come down the…
140 00:10:35.890 ⇒ 00:10:39.779 Samuel Roberts: Down the line that we didn’t realize it, like this, kind of thing.
141 00:10:42.020 ⇒ 00:10:42.720 Casie Aviles: Okay.
142 00:10:43.070 ⇒ 00:10:46.369 Casie Aviles: So… Yeah, so I think to…
143 00:10:46.850 ⇒ 00:10:53.059 Casie Aviles: It’s still a good idea, like, to continue with the migration plan, of course, so that’s still going to be our priority.
144 00:10:54.160 ⇒ 00:10:58.800 Casie Aviles: So, yeah.
145 00:10:59.450 ⇒ 00:11:00.200 Casie Aviles: Or…
146 00:11:00.200 ⇒ 00:11:01.570 Samuel Roberts: Yeah, I would say… oh, go ahead.
147 00:11:03.410 ⇒ 00:11:09.309 Casie Aviles: Yeah, like, additionally, there was, like, from the email that I got from Enid.
148 00:11:11.450 ⇒ 00:11:17.580 Casie Aviles: They were suggesting, like… Optimizations.
149 00:11:21.450 ⇒ 00:11:28.340 Casie Aviles: Yeah, I’m not sure if… We should, like, continue… still work on those, or just focus on…
150 00:11:29.270 ⇒ 00:11:31.159 Casie Aviles: On the migration itself.
151 00:11:32.680 ⇒ 00:11:38.050 Samuel Roberts: Yeah, that’s a good… respond to that email.
152 00:11:39.300 ⇒ 00:11:41.339 Casie Aviles: Yeah, let me look for it as well.
153 00:11:42.920 ⇒ 00:11:44.469 Samuel Roberts: Let’s go back to this problem.
154 00:11:49.960 ⇒ 00:11:50.599 Samuel Roberts: Thank you.
155 00:11:50.600 ⇒ 00:11:51.170 Casie Aviles: And…
156 00:11:51.670 ⇒ 00:11:52.859 Samuel Roberts: It’s November 26th.
157 00:11:55.130 ⇒ 00:11:56.820 Samuel Roberts: Okay, I think I got it here.
158 00:12:00.260 ⇒ 00:12:03.800 Casie Aviles: Yeah, so we have this email…
159 00:12:05.820 ⇒ 00:12:09.230 Casie Aviles: These are kind of also some of the stuff they pointed out.
160 00:12:09.660 ⇒ 00:12:10.530 Samuel Roberts: Yeah.
161 00:12:11.830 ⇒ 00:12:16.740 Casie Aviles: I haven’t really gotten around to actually optimizing a lot, like.
162 00:12:19.370 ⇒ 00:12:26.650 Casie Aviles: Yeah, it’s more… most… just mostly just band-aid solutions, honestly, that I’ve been… band-aid fixes, that I’ve been doing lately.
163 00:12:28.100 ⇒ 00:12:34.759 Samuel Roberts: No, I… I think the better strategy is to focus on the migration kind of as…
164 00:12:35.120 ⇒ 00:12:39.819 Samuel Roberts: Quickly as we can get it up and running, like, on par with what it’s doing now.
165 00:12:39.820 ⇒ 00:12:40.560 Casie Aviles: Okay.
166 00:12:41.110 ⇒ 00:12:42.400 Samuel Roberts: Just because I feel…
167 00:12:42.750 ⇒ 00:12:48.680 Samuel Roberts: that if we focus time and energy on these band-aids, we’re just… there’s gonna be other things that creep up.
168 00:12:49.030 ⇒ 00:12:53.789 Samuel Roberts: And if we can get off of there and have more observability.
169 00:12:53.920 ⇒ 00:12:58.590 Samuel Roberts: And notifications, then we’ll be able to more cleanly fix these things.
170 00:12:59.990 ⇒ 00:13:00.420 Casie Aviles: Okay.
171 00:13:03.040 ⇒ 00:13:08.390 Samuel Roberts: That’s my thought. I don’t know exactly how quickly we can make that happen, but I think the quicker the better.
172 00:13:08.530 ⇒ 00:13:10.060 Samuel Roberts: So.
173 00:13:12.130 ⇒ 00:13:12.970 Casie Aviles: Okay, okay.
174 00:13:13.860 ⇒ 00:13:17.819 Samuel Roberts: That’s… yeah, I… you know, I feel a little,
175 00:13:18.700 ⇒ 00:13:27.559 Samuel Roberts: unsure saying that, because obviously, like, the N8N stuff is what they’re still going to be using, but if we can get something stood up quickly that is, you know, we can test out
176 00:13:28.540 ⇒ 00:13:34.169 Samuel Roberts: then I think that’s probably a better, strategy than trying to do both in parallel.
177 00:13:34.410 ⇒ 00:13:36.099 Samuel Roberts: Cause it’ll just slow down the…
178 00:13:37.060 ⇒ 00:13:37.540 Casie Aviles: Yeah.
179 00:13:37.540 ⇒ 00:13:44.530 Samuel Roberts: the migration, and probably, to be honest, not improve the N8N as much as they think it would, in terms of N8N.
180 00:13:44.810 ⇒ 00:13:52.849 Samuel Roberts: Like, they’re probably like, oh yeah, try all these other things, but, like, it might just be not too much work right now for the gains we’re gonna get.
181 00:13:53.950 ⇒ 00:13:55.410 Casie Aviles: Yeah, yeah, that makes sense.
182 00:13:55.900 ⇒ 00:14:01.190 Casie Aviles: Yeah, I guess, yeah, my main concern there was just… Like…
183 00:14:02.260 ⇒ 00:14:07.760 Casie Aviles: They’re still actively using the N8N1, so, if it’s still…
184 00:14:07.880 ⇒ 00:14:15.700 Casie Aviles: if there are some issues there, like, I was just worried that that… that might be…
185 00:14:16.160 ⇒ 00:14:20.160 Casie Aviles: That might be, like, you know, a huge problem if it keeps…
186 00:14:20.160 ⇒ 00:14:21.050 Samuel Roberts: That’s true.
187 00:14:21.180 ⇒ 00:14:25.979 Samuel Roberts: I guess, okay, so then that leads to, like, how quickly can we get a migration done?
188 00:14:25.980 ⇒ 00:14:26.860 Casie Aviles: Yep, yep.
189 00:14:27.450 ⇒ 00:14:35.000 Samuel Roberts: So, I guess that kind of becomes, like, what is… what does the time look like on this, and other clients, and other work?
190 00:14:35.430 ⇒ 00:14:40.560 Samuel Roberts: You know, how much can we… Pour into this.
191 00:14:40.960 ⇒ 00:14:45.940 Casie Aviles: Yeah, I believe, we’ve been… there have been reallocations.
192 00:14:46.340 ⇒ 00:14:46.730 Samuel Roberts: Yeah.
193 00:14:46.730 ⇒ 00:14:48.469 Casie Aviles: Right? So… Yep.
194 00:14:48.970 ⇒ 00:14:54.810 Casie Aviles: I think I should be able to just give more of my focus and time on ABC.
195 00:14:55.660 ⇒ 00:14:56.230 Samuel Roberts: Okay.
196 00:14:56.690 ⇒ 00:15:05.349 Samuel Roberts: I think, yeah, I think that’s a good… if that’s the case, then I would say, like, you know, we could probably get, like, a V1 of a migrated thing out
197 00:15:05.780 ⇒ 00:15:09.189 Samuel Roberts: What week? Tuesday.
198 00:15:09.720 ⇒ 00:15:13.690 Samuel Roberts: Then maybe by… I mean…
199 00:15:15.340 ⇒ 00:15:22.150 Samuel Roberts: Mustafa, you did some of the work, or I… actually, I don’t know who did the work, but Mustafa, you were telling me about the work for the generic client hub, right?
200 00:15:22.630 ⇒ 00:15:24.369 Casie Aviles: Yeah, Mustafo worked on that.
201 00:15:24.370 ⇒ 00:15:24.850 Mustafa Raja: Yeah.
202 00:15:24.850 ⇒ 00:15:29.380 Samuel Roberts: Okay, so that was… that was kind of a similar thing, where we’re moving over N8NLogic.
203 00:15:29.790 ⇒ 00:15:31.830 Samuel Roberts: I guess…
204 00:15:31.830 ⇒ 00:15:32.720 Mustafa Raja: Yes.
205 00:15:32.940 ⇒ 00:15:33.430 Mustafa Raja: So…
206 00:15:33.430 ⇒ 00:15:34.699 Samuel Roberts: Based on that experience.
207 00:15:34.700 ⇒ 00:15:36.110 Mustafa Raja: There’s two parts.
208 00:15:37.540 ⇒ 00:15:40.560 Samuel Roberts: There’s two parts to this, right?
209 00:15:40.580 ⇒ 00:15:44.610 Mustafa Raja: One is, you know, just Andy, and then one is.
210 00:15:44.610 ⇒ 00:15:44.960 Samuel Roberts: cooking.
211 00:15:44.960 ⇒ 00:15:46.320 Mustafa Raja: bidding pipelines.
212 00:15:46.640 ⇒ 00:15:53.289 Mustafa Raja: And we haven’t, in our plan, talked about how we are going to, you know, handle the embeddings.
213 00:15:53.480 ⇒ 00:16:11.930 Mustafa Raja: But just the agent, I think the only thing that is going to take time here would be, you know, getting the integrations in code, you know? Because in NHS, it’s just a quick, you know, plug-in, whereas in code, it might be a little different.
214 00:16:13.170 ⇒ 00:16:22.189 Mustafa Raja: Other than that, I feel, if we are confident in getting this by… by end of week.
215 00:16:23.570 ⇒ 00:16:25.799 Mustafa Raja: I think it could be a win.
216 00:16:26.260 ⇒ 00:16:26.890 Mustafa Raja: What do you think?
217 00:16:26.890 ⇒ 00:16:27.650 Samuel Roberts: Okay.
218 00:16:28.620 ⇒ 00:16:34.340 Samuel Roberts: Yeah, I guess I had kind of… that slipped my mind about the embedding pipelines, but do we think…
219 00:16:35.060 ⇒ 00:16:40.660 Samuel Roberts: That some of the content accuracy problems are related to the embedding.
220 00:16:40.820 ⇒ 00:16:41.870 Samuel Roberts: Right now?
221 00:16:42.890 ⇒ 00:16:44.619 Samuel Roberts: Or is it more how we’re using that?
222 00:16:44.980 ⇒ 00:16:47.760 Samuel Roberts: in the… in Andy, because I think…
223 00:16:48.380 ⇒ 00:16:49.090 Mustafa Raja: Okay.
224 00:16:49.340 ⇒ 00:16:54.270 Mustafa Raja: Okay, so let me, let me, let me answer this,
225 00:16:55.610 ⇒ 00:17:05.259 Mustafa Raja: So, when we started, the content was a lot smaller, right? You know, we started with just one document, one department that was pissed department.
226 00:17:05.450 ⇒ 00:17:12.609 Mustafa Raja: And, if I remember correctly, there weren’t many accuracy issues.
227 00:17:12.730 ⇒ 00:17:18.350 Mustafa Raja: At that time. But we grew into maybe 5 to 6… 6 departments now.
228 00:17:18.550 ⇒ 00:17:32.850 Mustafa Raja: So… so you… you see how much context does the Andy has now, and there are, similar words among the… among the departments, and that is what.
229 00:17:32.850 ⇒ 00:17:33.440 Samuel Roberts: Missy.
230 00:17:33.440 ⇒ 00:17:34.390 Mustafa Raja: confused.
231 00:17:35.950 ⇒ 00:17:40.800 Samuel Roberts: That makes sense. And the routing logic was introduced because of this problem.
232 00:17:42.530 ⇒ 00:17:43.010 Mustafa Raja: You know?
233 00:17:43.010 ⇒ 00:17:43.550 Samuel Roberts: Okay.
234 00:17:43.550 ⇒ 00:17:55.210 Mustafa Raja: So, to mitigate this, people should be looking into whatever department they are referring to, but that didn’t work out quite
235 00:17:55.730 ⇒ 00:17:58.639 Mustafa Raja: Quite like we thought it would work.
236 00:17:58.820 ⇒ 00:18:13.750 Mustafa Raja: So my, my, my, goal, was to, you know, because I, we, we, we had planned to move from, move away from Google Chat.
237 00:18:13.840 ⇒ 00:18:20.290 Mustafa Raja: And if we did that, and went to a custom UI, we could have a dropdown to select the department.
238 00:18:20.520 ⇒ 00:18:24.159 Mustafa Raja: And that… that would have been a crazy situation for us.
239 00:18:24.310 ⇒ 00:18:30.669 Mustafa Raja: I don’t know if we are going to do that or not. So maybe a better routing logic?
240 00:18:31.120 ⇒ 00:18:34.540 Mustafa Raja: Or something like that, would be.
241 00:18:34.540 ⇒ 00:18:38.969 Samuel Roberts: Yeah, I mean, even… even if we move off of N8 N into something
242 00:18:39.830 ⇒ 00:18:45.560 Samuel Roberts: you know, codebase… like, there’s… there’s no reason Andy can’t ask for clarification, potentially.
243 00:18:46.820 ⇒ 00:18:47.720 Samuel Roberts: If there’s…
244 00:18:47.720 ⇒ 00:18:48.690 Mustafa Raja: Yes.
245 00:18:49.560 ⇒ 00:18:50.180 Samuel Roberts: So…
246 00:18:50.180 ⇒ 00:18:55.619 Mustafa Raja: the Andy, Andy’s, what’s it called? The system prompt is a lot, is like…
247 00:18:55.720 ⇒ 00:18:58.190 Mustafa Raja: The system prompt is too big, to be honest.
248 00:18:58.190 ⇒ 00:18:59.480 Samuel Roberts: Yeah, well, I think that’s what we…
249 00:18:59.480 ⇒ 00:19:00.219 Mustafa Raja: Let me open that up.
250 00:19:00.220 ⇒ 00:19:01.250 Samuel Roberts: migration.
251 00:19:01.720 ⇒ 00:19:02.640 Samuel Roberts: I’m sorry?
252 00:19:04.380 ⇒ 00:19:08.790 Mustafa Raja: I was asking if Casey can open it up in LineFuse, and we can take a look at it together.
253 00:19:08.790 ⇒ 00:19:10.749 Samuel Roberts: Oh yeah, that’d actually be really nice, yeah.
254 00:19:21.940 ⇒ 00:19:23.000 Casie Aviles: It’s loading.
255 00:19:25.300 ⇒ 00:19:26.980 Mustafa Raja: Yeah, it takes some time.
256 00:19:32.220 ⇒ 00:19:35.210 Casie Aviles: G… Yeah, it’s this one.
257 00:19:36.290 ⇒ 00:19:38.570 Casie Aviles: Yeah, it’s quite big right now.
258 00:19:39.150 ⇒ 00:19:39.570 Samuel Roberts: Okay.
259 00:19:39.570 ⇒ 00:19:40.770 Casie Aviles: and kind of…
260 00:19:41.550 ⇒ 00:19:53.800 Casie Aviles: Another thing here is, in order to fix some of the issues, we kind of hard-coded some of, like, the context here that Andy was not consistently getting right.
261 00:19:54.410 ⇒ 00:19:59.849 Samuel Roberts: Right. Okay, well that’s… That’s something we can start to work as we migrate.
262 00:19:59.970 ⇒ 00:20:06.720 Samuel Roberts: We can probably strip out things from the context and make those tool calls or lookups or something.
263 00:20:06.910 ⇒ 00:20:15.050 Mustafa Raja: So, I believe, 75% of the system prompt is just content, you know?
264 00:20:15.520 ⇒ 00:20:17.829 Samuel Roberts: Yeah, that’s definitely something we can fix.
265 00:20:18.050 ⇒ 00:20:18.640 Samuel Roberts: Okay.
266 00:20:18.640 ⇒ 00:20:27.019 Mustafa Raja: So if we fix that, maybe we would have more space, or more context to fit in Andy, so it could, you know.
267 00:20:27.020 ⇒ 00:20:32.629 Samuel Roberts: Perform better in terms of asking clarifications and other stuff.
268 00:20:34.400 ⇒ 00:20:35.070 Samuel Roberts: Okay.
269 00:20:38.000 ⇒ 00:20:38.790 Samuel Roberts: Yeah.
270 00:20:38.940 ⇒ 00:20:42.169 Samuel Roberts: I think that’s… I think that’s definitely right. I didn’t…
271 00:20:42.600 ⇒ 00:20:48.940 Samuel Roberts: remember how big this was. Okay.
272 00:20:50.410 ⇒ 00:20:54.659 Mustafa Raja: Also, what GPT model are we using with this?
273 00:20:54.770 ⇒ 00:21:00.370 Mustafa Raja: It might be a good time to, you know, Change that to… Yeah.
274 00:21:01.500 ⇒ 00:21:02.469 Casie Aviles: Yes, that’s possible.
275 00:21:03.240 ⇒ 00:21:06.220 Mustafa Raja: Do you have… do you know, Casey, which one are we using?
276 00:21:06.560 ⇒ 00:21:08.800 Casie Aviles: I think it’s still using 4.0.
277 00:21:10.130 ⇒ 00:21:12.069 Mustafa Raja: Hmm. Okay. Do… hmm…
278 00:21:12.570 ⇒ 00:21:13.540 Casie Aviles: Yeah, yeah.
279 00:21:14.530 ⇒ 00:21:16.110 Samuel Roberts: Yeah, I get what you’re saying.
280 00:21:16.110 ⇒ 00:21:18.650 Mustafa Raja: We do 5.1 or something.
281 00:21:18.790 ⇒ 00:21:22.610 Mustafa Raja: It’s just, you know, a different… Definitely worth testing out.
282 00:21:23.120 ⇒ 00:21:23.900 Samuel Roberts: Yeah.
283 00:21:24.930 ⇒ 00:21:31.139 Mustafa Raja: I mean, it’s been, like, 2 years since 4 came out, right?
284 00:21:31.400 ⇒ 00:21:32.310 Mustafa Raja: So… Hmm.
285 00:21:32.430 ⇒ 00:21:33.169 Mustafa Raja: Might be a good.
286 00:21:33.170 ⇒ 00:21:41.210 Samuel Roberts: Yeah, I think that’s… that’s definitely something we can test as we get into code, because that’s, you know, I don’t want to… I worry about just changing that, because it could change…
287 00:21:42.050 ⇒ 00:21:43.370 Samuel Roberts: A lot, potentially.
288 00:21:43.370 ⇒ 00:21:43.990 Mustafa Raja: Yeah.
289 00:21:43.990 ⇒ 00:21:47.030 Samuel Roberts: But if we, if we test it out, we could probably…
290 00:21:47.870 ⇒ 00:21:48.670 Mustafa Raja: Yeah, yeah.
291 00:21:49.870 ⇒ 00:21:50.470 Samuel Roberts: Okay.
292 00:21:50.660 ⇒ 00:21:53.310 Mustafa Raja: So, what do we think about the routing one?
293 00:21:53.580 ⇒ 00:21:58.080 Mustafa Raja: I mean, for now, it’s just turned off because it didn’t quite work the…
294 00:21:58.190 ⇒ 00:22:01.259 Mustafa Raja: Worked the best, but rather created more problems.
295 00:22:03.740 ⇒ 00:22:06.499 Samuel Roberts: Yeah, I think if we can avoid…
296 00:22:06.650 ⇒ 00:22:10.069 Samuel Roberts: That, like, using that, and maybe have it…
297 00:22:10.730 ⇒ 00:22:14.979 Samuel Roberts: Well, I don’t know, this opens up the door, if we’re gonna use Monstra as, like, a, you know.
298 00:22:15.340 ⇒ 00:22:20.100 Samuel Roberts: Agent Builder. This opens up a little more flexibility
299 00:22:21.100 ⇒ 00:22:24.230 Samuel Roberts: In how we structure these workflows.
300 00:22:28.520 ⇒ 00:22:33.790 Samuel Roberts: So, you know, like, we were, like, we were kind of stuck with a few things because of N8N’s…
301 00:22:34.030 ⇒ 00:22:42.530 Samuel Roberts: abilities… But… We can be a little more… creative, maybe?
302 00:22:42.880 ⇒ 00:22:56.259 Samuel Roberts: Where there is one agent that does some kind of routing and makes sure we know what department we’re talking about. If we don’t know what department we’re talking about, we make sure to ask that. If we clarify that, we can then look up the right things,
303 00:22:57.850 ⇒ 00:23:00.060 Casie Aviles: Rather than, like, one…
304 00:23:00.260 ⇒ 00:23:02.660 Samuel Roberts: some… mostly linear flow. Yeah.
305 00:23:04.060 ⇒ 00:23:04.700 Mustafa Raja: Right.
306 00:23:05.310 ⇒ 00:23:06.249 Samuel Roberts: So I think…
307 00:23:07.110 ⇒ 00:23:12.040 Samuel Roberts: What… like, what we’ve thought about for the migration is good, but we probably need a little more…
308 00:23:12.150 ⇒ 00:23:16.600 Samuel Roberts: And maybe that’s something I just need to sit down and, like, diagram out a little bit.
309 00:23:16.880 ⇒ 00:23:20.460 Samuel Roberts: like, what could we do with Mastra?
310 00:23:21.040 ⇒ 00:23:25.630 Samuel Roberts: That’s, you know, making sure we have all the information that’s necessary to do some kind of lookup.
311 00:23:25.930 ⇒ 00:23:31.190 Samuel Roberts: Making sure we know if we’re looking up you know, like…
312 00:23:32.410 ⇒ 00:23:35.860 Samuel Roberts: this might be a little bit where I need to get into some of these,
313 00:23:36.350 ⇒ 00:23:42.840 Samuel Roberts: the details of what gets asked and stuff. Like, what kind of issues are we having with the content? Accuracy, is it?
314 00:23:44.120 ⇒ 00:23:45.430 Samuel Roberts: Is it based on…
315 00:23:45.900 ⇒ 00:23:48.440 Samuel Roberts: Parts of the… I just haven’t dug.
316 00:23:48.440 ⇒ 00:24:01.680 Mustafa Raja: Yeah, it’s just, you know, some of the… a lot of the times, it’s the same words, you know, that get… gets asked, that are mentioned multiple times in multiple different documents, and
317 00:24:01.830 ⇒ 00:24:07.249 Mustafa Raja: You know, the… the vector store would pull all of them, and…
318 00:24:07.380 ⇒ 00:24:11.569 Mustafa Raja: That’s all the context that Andy would have.
319 00:24:12.100 ⇒ 00:24:12.450 Samuel Roberts: Yeah.
320 00:24:12.450 ⇒ 00:24:20.260 Mustafa Raja: And… the information it would tell us wouldn’t be incorrect, but it would be, you know, Not really.
321 00:24:20.260 ⇒ 00:24:23.219 Samuel Roberts: The wrong… yeah, yeah, the wrong part, okay.
322 00:24:23.400 ⇒ 00:24:25.719 Samuel Roberts: Okay, while we’re…
323 00:24:27.570 ⇒ 00:24:35.750 Samuel Roberts: while we’re all on, can we talk a little bit about the embedding pipeline? Because I think, like I said, or like you said, we talked about migrating Andy’s
324 00:24:36.000 ⇒ 00:24:42.279 Samuel Roberts: like, the chat functionality, but I know less about the actual embedding. Is it possible to…
325 00:24:43.670 ⇒ 00:24:46.259 Mustafa Raja: Yeah, we… let’s go to the diagram.
326 00:24:46.420 ⇒ 00:24:48.119 Mustafa Raja: That we have in FigDem.
327 00:24:49.710 ⇒ 00:24:50.489 Casie Aviles: For, for…
328 00:24:50.490 ⇒ 00:24:51.340 Samuel Roberts: Yes, that’s right.
329 00:24:51.820 ⇒ 00:24:52.670 Samuel Roberts: For what?
330 00:24:52.850 ⇒ 00:24:55.620 Casie Aviles: For, is it for ABC, or…
331 00:24:55.620 ⇒ 00:24:58.569 Samuel Roberts: Yeah, yeah, still ABC stuff. Yeah, I know, ABC, I’m curious.
332 00:25:00.010 ⇒ 00:25:02.469 Mustafa Raja: Yeah, let’s go to FigJam, we have it there.
333 00:25:02.470 ⇒ 00:25:04.019 Samuel Roberts: That’s right, I forgot we had.
334 00:25:07.450 ⇒ 00:25:10.130 Casie Aviles: Let me see if this is the right…
335 00:25:13.960 ⇒ 00:25:18.180 Casie Aviles: But yeah, right now, it looks… we have these.
336 00:25:18.370 ⇒ 00:25:20.890 Casie Aviles: similar to.
337 00:25:20.890 ⇒ 00:25:21.709 Mustafa Raja: Yeah, these are the…
338 00:25:21.710 ⇒ 00:25:23.150 Casie Aviles: hubs, you know?
339 00:25:23.150 ⇒ 00:25:25.139 Samuel Roberts: Okay, so what do we have, a coverage…
340 00:25:25.140 ⇒ 00:25:27.910 Mustafa Raja: store, and the Central Dock store? Okay.
341 00:25:31.870 ⇒ 00:25:33.030 Casie Aviles: Hmm, this is the wrong one.
342 00:25:33.400 ⇒ 00:25:34.020 Casie Aviles: That means…
343 00:25:34.020 ⇒ 00:25:35.550 Samuel Roberts: Is it the page? Oh, okay.
344 00:25:41.660 ⇒ 00:25:42.830 Samuel Roberts: Right, okay.
345 00:25:43.720 ⇒ 00:25:44.400 Casie Aviles: I think this one…
346 00:25:44.400 ⇒ 00:25:44.920 Mustafa Raja: inboxes.
347 00:25:44.920 ⇒ 00:25:45.659 Casie Aviles: We were talking about.
348 00:25:45.660 ⇒ 00:25:47.100 Mustafa Raja: congestion pipeline.
349 00:25:47.910 ⇒ 00:25:48.580 Samuel Roberts: Yeah.
350 00:25:50.000 ⇒ 00:25:50.800 Casie Aviles: This one?
351 00:25:51.390 ⇒ 00:26:00.220 Mustafa Raja: So if you would scroll, scroll, sorry, zoom out, there would be, yeah, on top right, there’s a central… yeah, this one.
352 00:26:00.540 ⇒ 00:26:02.250 Mustafa Raja: ingestion pipeline.
353 00:26:02.640 ⇒ 00:26:05.699 Mustafa Raja: So this is the flow that we are following right now.
354 00:26:06.240 ⇒ 00:26:06.940 Samuel Roberts: Okay.
355 00:26:07.360 ⇒ 00:26:10.949 Samuel Roberts: Essential Doc Triggers… One for each central dock.
356 00:26:11.820 ⇒ 00:26:16.789 Samuel Roberts: That’s the raw content for the doc and JSON, convert it to sections, headings, content…
357 00:26:21.780 ⇒ 00:26:22.490 Samuel Roberts: Okay.
358 00:26:24.130 ⇒ 00:26:31.200 Samuel Roberts: Yeah, so maybe… while we’re…
359 00:26:31.310 ⇒ 00:26:36.290 Samuel Roberts: migrating the rest to Mastra, we want to think about this more than I was anticipating, because I think
360 00:26:36.730 ⇒ 00:26:41.240 Samuel Roberts: What we’re doing here… is probably…
361 00:26:41.550 ⇒ 00:26:44.859 Samuel Roberts: influenced by the fact that it was NAN as well, right?
362 00:26:45.170 ⇒ 00:26:45.830 Casie Aviles: Yeah, yeah.
363 00:26:45.830 ⇒ 00:26:50.730 Samuel Roberts: So… So, let’s… let’s think about this a little bit. Let’s… I got… okay.
364 00:26:50.940 ⇒ 00:26:53.639 Samuel Roberts: So, one for each central dock. How many central docks are there?
365 00:26:55.160 ⇒ 00:27:02.960 Mustafa Raja: I think it’s 6. I’m not sure. Let me… let me just go ahead and look, to be honest.
366 00:27:13.070 ⇒ 00:27:16.639 Samuel Roberts: Now, did we look at contextual a bit?
367 00:27:18.890 ⇒ 00:27:20.619 Mustafa Raja: What Contextune does is…
368 00:27:21.560 ⇒ 00:27:27.919 Mustafa Raja: Go ahead. Yeah, the only thing that contextual… contextual, does not do is the… I think it does not…
369 00:27:28.240 ⇒ 00:27:31.449 Mustafa Raja: 8 or something, if the document is updated.
370 00:27:31.680 ⇒ 00:27:32.319 Mustafa Raja: So we would have.
371 00:27:32.320 ⇒ 00:27:32.650 Samuel Roberts: Okay.
372 00:27:32.650 ⇒ 00:27:33.450 Mustafa Raja: into that.
373 00:27:35.870 ⇒ 00:27:37.290 Casie Aviles: Oh, you mean syncing?
374 00:27:38.180 ⇒ 00:27:38.870 Mustafa Raja: Yeah.
375 00:27:39.200 ⇒ 00:27:39.950 Casie Aviles: Okay.
376 00:27:43.080 ⇒ 00:27:50.080 Casie Aviles: Yeah, I was also kind of thinking about whether we should just use, like, an off-the-shelf tool, or…
377 00:27:51.010 ⇒ 00:27:54.769 Samuel Roberts: Yeah, because I… I… especially for something that’s, like.
378 00:27:55.110 ⇒ 00:28:02.560 Samuel Roberts: documents like this. I’m wondering how much we, you know… and things that are gonna have, you know, words that are similar.
379 00:28:03.170 ⇒ 00:28:05.080 Samuel Roberts: M.
380 00:28:05.950 ⇒ 00:28:11.470 Samuel Roberts: Could be worth that, and then what we have to just figure out a way to update it ourselves.
381 00:28:11.790 ⇒ 00:28:13.170 Samuel Roberts: Instead of automatically.
382 00:28:13.360 ⇒ 00:28:17.110 Casie Aviles: I think that they should have an API contextual,
383 00:28:17.210 ⇒ 00:28:24.789 Casie Aviles: like, I guess my worry is, like, if we were to build it again, like our, like, the embedding pipeline.
384 00:28:25.510 ⇒ 00:28:28.200 Casie Aviles: I mean, there’s only a few of us…
385 00:28:28.840 ⇒ 00:28:38.219 Casie Aviles: I don’t know, maybe it might… might be difficult, like, there’s a lot of things that can go wrong if we build it ourselves.
386 00:28:38.840 ⇒ 00:28:46.010 Samuel Roberts: Exactly, exactly. And especially for something that, like, these are 6 documents, there’s, like, a lot in them that is similar.
387 00:28:46.400 ⇒ 00:28:48.030 Mustafa Raja: Yeah, this is the document.
388 00:28:48.380 ⇒ 00:28:50.690 Mustafa Raja: Yeah, we have 6 documents,
389 00:28:51.300 ⇒ 00:28:54.130 Mustafa Raja: 5 central docks, and then 1 coverage dock.
390 00:28:55.250 ⇒ 00:28:55.850 Samuel Roberts: Okay.
391 00:28:57.900 ⇒ 00:29:01.169 Samuel Roberts: The central… Or 5 Central 1 coverage.
392 00:29:01.340 ⇒ 00:29:03.090 Samuel Roberts: But that follows the same pipeline.
393 00:29:03.790 ⇒ 00:29:04.739 Mustafa Raja: Yeah.
394 00:29:04.890 ⇒ 00:29:05.790 Mustafa Raja: All of them.
395 00:29:07.620 ⇒ 00:29:11.750 Samuel Roberts: Over at… Okay,
396 00:29:13.230 ⇒ 00:29:19.239 Samuel Roberts: Yeah, so I think it’s probably worth doing, have you guys worked with Contextual much?
397 00:29:20.460 ⇒ 00:29:23.769 Mustafa Raja: I have not, but I would love to.
398 00:29:24.330 ⇒ 00:29:29.790 Samuel Roberts: I haven’t… we haven’t really applied it to any existing clients, I think. Yeah.
399 00:29:29.790 ⇒ 00:29:33.580 Casie Aviles: The main concern there was, like, the cost.
400 00:29:33.980 ⇒ 00:29:37.250 Samuel Roberts: Okay, that’s what I was gonna add, I didn’t know how that scales and stuff.
401 00:29:37.250 ⇒ 00:29:43.319 Casie Aviles: I forgot how much it exactly was, but I remember it being expensive.
402 00:29:43.780 ⇒ 00:29:54.180 Casie Aviles: But I think for a tool like Contextual, that would be… that would help us a lot, because we don’t have to do all of this.
403 00:29:57.150 ⇒ 00:30:05.280 Casie Aviles: You know? I mean, we could build it, but yeah, it’s going to be… A bigger… like, task.
404 00:30:05.780 ⇒ 00:30:06.440 Samuel Roberts: Yeah.
405 00:30:07.020 ⇒ 00:30:08.710 Samuel Roberts: I think so, okay.
406 00:30:12.980 ⇒ 00:30:22.380 Samuel Roberts: So I think in the… Meantime… Let’s… Let’s focus on Andy.
407 00:30:22.380 ⇒ 00:30:24.300 Casie Aviles: The conversational one, okay.
408 00:30:24.300 ⇒ 00:30:25.050 Samuel Roberts: Yeah.
409 00:30:25.290 ⇒ 00:30:31.400 Samuel Roberts: But let’s… let’s… I might need to do a little bit more of the… or calculate those numbers, or see, just so I have a sense of it.
410 00:30:31.620 ⇒ 00:30:32.510 Samuel Roberts: Because, like.
411 00:30:32.680 ⇒ 00:30:36.799 Samuel Roberts: You know, we have a relationship with Contextual, but there are other tools as well that we could use.
412 00:30:39.100 ⇒ 00:30:42.770 Samuel Roberts: So it might be worth doing a little bit of a deep dive into that.
413 00:30:43.250 ⇒ 00:30:47.899 Casie Aviles: Yeah, if we could do contextual, that would be okay. Yeah, that would be fine.
414 00:30:48.830 ⇒ 00:30:49.450 Samuel Roberts: Okay.
415 00:30:49.640 ⇒ 00:30:55.170 Casie Aviles: Just wondering, like, I don’t know, like, the specifics about, like, The cost, or, like…
416 00:30:55.500 ⇒ 00:30:57.490 Casie Aviles: How we will work that out.
417 00:30:58.310 ⇒ 00:30:58.850 Casie Aviles: Like I say?
418 00:30:58.850 ⇒ 00:30:59.540 Samuel Roberts: Yeah.
419 00:30:59.800 ⇒ 00:31:05.230 Casie Aviles: the embedding that they do, I think that was pretty good. It showed good qualities.
420 00:31:05.430 ⇒ 00:31:11.749 Casie Aviles: I just haven’t tried it… tried connecting it as, like, a tool to an agent.
421 00:31:12.560 ⇒ 00:31:13.230 Samuel Roberts: Yeah.
422 00:31:13.890 ⇒ 00:31:20.729 Casie Aviles: But we can create, like, our own agents there as well, using contextual, and I think the results are not bad.
423 00:31:20.930 ⇒ 00:31:21.660 Casie Aviles: And…
424 00:31:21.660 ⇒ 00:31:22.290 Samuel Roberts: Okay.
425 00:31:23.020 ⇒ 00:31:26.199 Casie Aviles: Certainly beats having to worry about
426 00:31:26.510 ⇒ 00:31:29.970 Casie Aviles: You know, what chunking strategy we have to do, like.
427 00:31:29.970 ⇒ 00:31:30.990 Samuel Roberts: Yeah…
428 00:31:31.150 ⇒ 00:31:42.999 Samuel Roberts: Exactly, there’s a lot of, like, tuning that I just think we don’t need to worry about if we do something else, but if the cost is an issue, then… a cost or security or something goes.
429 00:31:43.670 ⇒ 00:31:51.480 Samuel Roberts: Okay, but I think in the meantime, If we could,
430 00:31:52.570 ⇒ 00:31:55.819 Samuel Roberts: start moving off N8N for the conversation, at least?
431 00:31:56.720 ⇒ 00:31:57.420 Casie Aviles: Okay.
432 00:31:59.640 ⇒ 00:32:01.479 Samuel Roberts: I think that’s probably, like, just…
433 00:32:01.480 ⇒ 00:32:02.440 Mustafa Raja: Nope.
434 00:32:04.420 ⇒ 00:32:08.109 Mustafa Raja: I was saying, does end of week look like a reasonable time for that?
435 00:32:09.080 ⇒ 00:32:09.750 Casie Aviles: Sorry?
436 00:32:10.030 ⇒ 00:32:11.140 Casie Aviles: End of week.
437 00:32:11.570 ⇒ 00:32:12.380 Mustafa Raja: Yeah.
438 00:32:14.130 ⇒ 00:32:18.539 Casie Aviles: Yeah, I think… I should be able to…
439 00:32:18.650 ⇒ 00:32:21.920 Casie Aviles: work on it, given that I have more time, but…
440 00:32:21.920 ⇒ 00:32:22.350 Samuel Roberts: Okay.
441 00:32:22.350 ⇒ 00:32:30.049 Casie Aviles: I think another thing that I also wanted to kind of think about was, like, how will the QA process be?
442 00:32:30.310 ⇒ 00:32:36.460 Casie Aviles: Yeah. But I also don’t kind of want to bog down, like, the process, like, with too much…
443 00:32:37.010 ⇒ 00:32:43.530 Casie Aviles: You know… like, I still want to ship at a reasonable rate, but I also don’t want to.
444 00:32:43.530 ⇒ 00:32:43.960 Samuel Roberts: Of course.
445 00:32:43.960 ⇒ 00:32:46.570 Casie Aviles: Ship it with lots of holes, you know.
446 00:32:46.940 ⇒ 00:32:49.500 Samuel Roberts: No, no, I understand. I mean, the nice thing…
447 00:32:49.890 ⇒ 00:32:56.950 Samuel Roberts: Is that if, once it’s in code, and we can add some testing on top of it, and probably a more…
448 00:32:57.570 ⇒ 00:32:59.930 Samuel Roberts: Ergonomic way than N8N lets us do.
449 00:33:01.500 ⇒ 00:33:02.210 Casie Aviles: Okay.
450 00:33:04.120 ⇒ 00:33:10.370 Samuel Roberts: We also have logs that we could maybe Create batches of testing with?
451 00:33:11.180 ⇒ 00:33:12.300 Casie Aviles: Yes, yes.
452 00:33:12.650 ⇒ 00:33:24.770 Samuel Roberts: So, I think, like, we have… now that, you know, we’re starting in a different place here than we were when you’re starting Andy, because we know what kinds of questions are asked, we know what was good and bad, we can test against that. So, I think…
453 00:33:25.900 ⇒ 00:33:32.209 Samuel Roberts: I think the first step is to get it, you know, get an Andy equivalent stood up in Mastra.
454 00:33:33.960 ⇒ 00:33:43.589 Samuel Roberts: Then would be, okay, we have something, it looks like in, you know, in your testing at your building that it’s doing the same kind of job.
455 00:33:44.670 ⇒ 00:33:46.569 Samuel Roberts: Let’s put together some…
456 00:33:46.690 ⇒ 00:33:53.539 Samuel Roberts: Testing suites that have, you know, questions we know it can answer, questions we know it couldn’t answer, questions we know it’s having issues with.
457 00:33:54.290 ⇒ 00:34:00.759 Samuel Roberts: and, and, yeah, so I think, I think basically the, the process is, like, let’s…
458 00:34:01.110 ⇒ 00:34:03.130 Samuel Roberts: get out of N8N.
459 00:34:03.670 ⇒ 00:34:09.249 Samuel Roberts: In… As quick as we can, so that we have, kind of.
460 00:34:10.239 ⇒ 00:34:14.440 Samuel Roberts: The ability to do more testing, and observability, and notifications, and…
461 00:34:14.610 ⇒ 00:34:17.409 Samuel Roberts: You know, performance measurement, things like that.
462 00:34:17.770 ⇒ 00:34:18.909 Samuel Roberts: then…
463 00:34:19.210 ⇒ 00:34:28.440 Samuel Roberts: We use those capabilities, basically, to make sure that it’s on par, or obviously better than what we currently have, and then switch over, kind of thing.
464 00:34:31.090 ⇒ 00:34:31.790 Casie Aviles: Okay.
465 00:34:32.510 ⇒ 00:34:35.879 Samuel Roberts: But yeah, the faster we can at least get, like, the basic.
466 00:34:35.889 ⇒ 00:34:36.959 Mustafa Raja: I’m just wondering…
467 00:34:37.510 ⇒ 00:34:38.060 Samuel Roberts: Yeah.
468 00:34:44.100 ⇒ 00:34:46.369 Mustafa Raja: I’m just wondering what the f-
469 00:34:46.710 ⇒ 00:34:53.639 Mustafa Raja: what the framework is going to be, you know? Is it just going to be a plain Node application or something?
470 00:34:54.510 ⇒ 00:35:01.110 Samuel Roberts: Yeah, I think… I think the best strategy is just to spin up a Mastra app using their… their… their, like, Quick Start.
471 00:35:02.560 ⇒ 00:35:05.620 Mustafa Raja: Oh, yeah, yeah, yeah, we, yeah, oh, sorry, I forgot, I forgot we were using.
472 00:35:05.620 ⇒ 00:35:06.450 Samuel Roberts: Yeah, I think…
473 00:35:06.450 ⇒ 00:35:06.910 Mustafa Raja: Huh.
474 00:35:06.910 ⇒ 00:35:14.430 Samuel Roberts: I think for now, we just can do that, because, you know, it has the chat interface on the testing environment,
475 00:35:14.690 ⇒ 00:35:18.389 Samuel Roberts: You can even, if you do it right, you can swap models and things.
476 00:35:19.350 ⇒ 00:35:24.130 Samuel Roberts: So I think, just creating a new repo with that, or,
477 00:35:24.510 ⇒ 00:35:35.409 Samuel Roberts: Oh, okay. And just… and just creating a master… like, you can use a, create master CLI. We’re not looking to put it behind a front end or anything right now.
478 00:35:35.700 ⇒ 00:35:40.500 Samuel Roberts: So I think, for now, for our testing, we can just get it running locally.
479 00:35:40.750 ⇒ 00:35:43.619 Samuel Roberts: connecting it to Supabase, connecting it,
480 00:35:44.110 ⇒ 00:35:49.480 Samuel Roberts: Not even to a front end, but just using the… master dev environment.
481 00:35:50.280 ⇒ 00:36:01.340 Samuel Roberts: to, like, chat with it just like normal, and then once we get that in a good place, we can start running tests against it, then we can get it in a way that will connect, because I think it… it gives, you know.
482 00:36:01.590 ⇒ 00:36:04.940 Samuel Roberts: APIs for being able to access and everything, so…
483 00:36:06.230 ⇒ 00:36:11.950 Samuel Roberts: I’m not too worried about that. I think that’s the best path forward for that.
484 00:36:15.060 ⇒ 00:36:22.510 Samuel Roberts: Yeah, I think that’s… that’s pretty good. The only other thing is, I know… There are other…
485 00:36:22.830 ⇒ 00:36:26.680 Samuel Roberts: things out there that we could host as well, but I think they’re more…
486 00:36:27.380 ⇒ 00:36:31.059 Samuel Roberts: I don’t know if you guys looked at Leta, I think Gabe had mentioned something.
487 00:36:31.530 ⇒ 00:36:37.300 Samuel Roberts: That has, like, a little more memory… Directly?
488 00:36:39.920 ⇒ 00:36:41.659 Casie Aviles: Oh, yeah, I haven’t tried that one.
489 00:36:42.200 ⇒ 00:36:45.650 Samuel Roberts: Yeah, I haven’t… I’ve seen things about it, but I have not tried it, so I’m… I’m…
490 00:36:46.040 ⇒ 00:36:51.560 Samuel Roberts: A little hesitant to just say, like, let’s not just run with what we know right now.
491 00:36:51.750 ⇒ 00:36:52.620 Samuel Roberts: But…
492 00:36:54.600 ⇒ 00:37:04.110 Casie Aviles: Yeah, that’s actually another thing we have to think about now. I’m not sure how Master handles it yet, though, but with Andy, the N8N version.
493 00:37:04.320 ⇒ 00:37:09.019 Casie Aviles: It basically just has, like, a chat history node, and…
494 00:37:09.750 ⇒ 00:37:13.570 Casie Aviles: it… I think it handles it in memory, the chat.
495 00:37:15.170 ⇒ 00:37:15.770 Casie Aviles: Or, like…
496 00:37:15.770 ⇒ 00:37:16.600 Samuel Roberts: Yeah, you’re right.
497 00:37:16.600 ⇒ 00:37:21.670 Casie Aviles: There are… there are two… options, basically. One is in-memory, and then the other is
498 00:37:21.960 ⇒ 00:37:26.790 Casie Aviles: Via Postgres, so we can store the history there.
499 00:37:27.330 ⇒ 00:37:33.610 Samuel Roberts: Yeah, I think Monster basically has, like, a conversation history that you can plug in different,
500 00:37:35.410 ⇒ 00:37:38.019 Samuel Roberts: different things into… yeah, memory, and…
501 00:37:38.020 ⇒ 00:37:39.640 Casie Aviles: Okay, I’ll try that one.
502 00:37:40.210 ⇒ 00:37:45.709 Samuel Roberts: But yeah, take a look at the docs. Yeah, definitely, like, message me, you know, we can hop on and do whatever.
503 00:37:46.460 ⇒ 00:37:50.840 Samuel Roberts: But, I would say spin up a quick Mastra, and then try to get, like, the basic.
504 00:37:51.210 ⇒ 00:37:57.850 Samuel Roberts: You know, The Andy functionality connected,
505 00:37:58.660 ⇒ 00:38:01.749 Samuel Roberts: in the Mastra environment, and then…
506 00:38:02.970 ⇒ 00:38:09.000 Samuel Roberts: you know, we can work from there to optimize, like, all the little pieces that N88 might have given us.
507 00:38:10.900 ⇒ 00:38:11.590 Casie Aviles: Okay.
508 00:38:11.900 ⇒ 00:38:13.600 Samuel Roberts: But I believe that, you know, there’s just…
509 00:38:13.770 ⇒ 00:38:19.100 Mustafa Raja: Yeah, there’s storage, memory with… yeah, memory with Postgres, we can do all that, okay.
510 00:38:19.720 ⇒ 00:38:24.880 Samuel Roberts: conversation history… By default, each request includes the last 10 messages. Yeah, so I think…
511 00:38:25.120 ⇒ 00:38:34.259 Samuel Roberts: it’s all here, we just gotta, you know, we have to start using slightly different pieces from Monster than from any to end, but I think this will let us, you know.
512 00:38:34.780 ⇒ 00:38:37.419 Samuel Roberts: Get observability and stuff a lot better.
513 00:38:38.440 ⇒ 00:38:38.830 Mustafa Raja: Yeah.
514 00:38:38.830 ⇒ 00:38:43.000 Samuel Roberts: Okay, yeah, we probably need to… we also need to think about, like, the… Go ahead.
515 00:38:44.490 ⇒ 00:38:48.280 Mustafa Raja: Yeah, also feel free to, you know, assign tasks to me. I haven’t…
516 00:38:48.560 ⇒ 00:38:53.379 Mustafa Raja: I don’t have any task with, ABC, so yeah, I can help there also.
517 00:38:53.380 ⇒ 00:38:55.350 Samuel Roberts: Okay. Yeah, I would say,
518 00:38:55.350 ⇒ 00:38:57.379 Mustafa Raja: And I have time assigned, so…
519 00:38:58.120 ⇒ 00:39:05.810 Samuel Roberts: Okay, I’m not sure… I haven’t been in linear, really, since I got back, so I haven’t looked a ton at, like, what the tickets are like right now.
520 00:39:05.990 ⇒ 00:39:08.270 Samuel Roberts: What do you have open here? Any conversational agent?
521 00:39:08.890 ⇒ 00:39:11.640 Samuel Roberts: Pate the Monster Adrian, focus on conversation, yeah, Testus.
522 00:39:13.830 ⇒ 00:39:16.269 Casie Aviles: Is this, yeah, this is an old ticket that I had.
523 00:39:16.890 ⇒ 00:39:17.490 Samuel Roberts: Okay.
524 00:39:17.490 ⇒ 00:39:20.050 Casie Aviles: I just pulled it up, if you ever have to make.
525 00:39:20.050 ⇒ 00:39:25.979 Samuel Roberts: Yeah, I would say, I would say get that, let’s just create a new…
526 00:39:26.640 ⇒ 00:39:31.960 Samuel Roberts: repo for now. I don’t really want to put it in with other ones until we get a…
527 00:39:32.320 ⇒ 00:39:34.010 Samuel Roberts: Candle on how this is working.
528 00:39:34.010 ⇒ 00:39:36.500 Casie Aviles: So… I mean, unless we have one that’s good, is this…
529 00:39:37.220 ⇒ 00:39:38.810 Samuel Roberts: Andy, Google Trend.
530 00:39:40.340 ⇒ 00:39:46.189 Samuel Roberts: Is this… I just want to get something running locally, but go ahead.
531 00:39:46.560 ⇒ 00:39:53.530 Casie Aviles: Yeah, this is the repo that we use for… for, like, the dashboard, which is real.
532 00:39:53.930 ⇒ 00:40:01.850 Casie Aviles: And then also, Dim is… Basically, listening to this, repo whenever we make changes, so…
533 00:40:04.460 ⇒ 00:40:09.409 Casie Aviles: It’s not really hooked up to anything, just the dashboard thing, for real.
534 00:40:09.410 ⇒ 00:40:12.520 Samuel Roberts: Yeah, actually, that’s probably smart. I would say make a new folder in here.
535 00:40:13.480 ⇒ 00:40:17.090 Casie Aviles: Okay, yeah, I can visualize And run… Make a new folder. Yeah.
536 00:40:17.200 ⇒ 00:40:22.329 Samuel Roberts: That’s perfect, yeah, and then, and do all the NPM install and all the master stuff in there.
537 00:40:24.790 ⇒ 00:40:29.089 Samuel Roberts: I didn’t realize, yeah, I wasn’t thinking about what we had, but that’s definitely keeping everything together, makes sense.
538 00:40:31.070 ⇒ 00:40:31.570 Samuel Roberts: Okay.
539 00:40:31.570 ⇒ 00:40:32.300 Casie Aviles: you know.
540 00:40:32.300 ⇒ 00:40:34.620 Samuel Roberts: Yeah, I would say get that,
541 00:40:35.470 ⇒ 00:40:39.750 Samuel Roberts: Get that started, and then, let me pull up the whiteboard, because what do we have?
542 00:40:39.880 ⇒ 00:40:42.080 Samuel Roberts: There’s different pieces, right?
543 00:40:42.540 ⇒ 00:40:48.589 Samuel Roberts: So, we have the routing agent, which we might not worry about, because it’s not turned on right now, right?
544 00:40:50.260 ⇒ 00:40:55.060 Samuel Roberts: In any of that? Yes, yes, yes, it’s not… Okay. So we have the context agent.
545 00:40:55.750 ⇒ 00:41:01.730 Samuel Roberts: And then we have the Supabase Rag, tool calls… Basically.
546 00:41:04.220 ⇒ 00:41:06.530 Samuel Roberts: So I think, yeah, if we can get a basic…
547 00:41:06.650 ⇒ 00:41:12.670 Samuel Roberts: master project set up, then, like, one of you could work on the context agent, one of you could get the RAG set up.
548 00:41:12.920 ⇒ 00:41:19.270 Samuel Roberts: Probably, Mustafa, you could get the rag set up, because you did that with, the client hub.
549 00:41:20.020 ⇒ 00:41:27.410 Mustafa Raja: Yeah, yeah, I can do the… I can do the… what’s it called? The agent. Did you… did you get the chance to look at…
550 00:41:28.210 ⇒ 00:41:29.929 Mustafa Raja: What’s it called?
551 00:41:30.520 ⇒ 00:41:31.999 Mustafa Raja: The PG thing?
552 00:41:32.840 ⇒ 00:41:36.750 Samuel Roberts: Not yet, no, I have not had a chance to even get into any code, really, yet.
553 00:41:36.750 ⇒ 00:41:37.560 Mustafa Raja: Okay.
554 00:41:37.970 ⇒ 00:41:39.730 Samuel Roberts: But yeah, we’ll take a look.
555 00:41:39.730 ⇒ 00:41:40.620 Mustafa Raja: Yeah, it’s okay.
556 00:41:41.230 ⇒ 00:41:47.600 Samuel Roberts: Yeah, I would say, kind of build out those two pieces, because the other, like I said, the other side of this is we could probably…
557 00:41:47.830 ⇒ 00:42:03.029 Samuel Roberts: think about this a little different, like, we could put a routing agent that’s a little bit different in front of everything, and then it knows where to go to get the right context, or, you know, we can think about it a little bit broader. But I want to just get something out that is gonna be,
558 00:42:04.890 ⇒ 00:42:08.469 Samuel Roberts: You know, a start, at least, of this, so…
559 00:42:09.800 ⇒ 00:42:10.450 Mustafa Raja: Yep.
560 00:42:11.470 ⇒ 00:42:12.050 Samuel Roberts: Okay
561 00:42:13.100 ⇒ 00:42:23.760 Samuel Roberts: Does that sound like a plan from here? Because I think that, yeah, the Q&A stuff can all happen once we have this code kind of started, and we can actually, like, make requests to our own internal MOSTRA thing that we can…
562 00:42:24.530 ⇒ 00:42:28.600 Samuel Roberts: test against Andy, or test as Andy, too, you know?
563 00:42:29.570 ⇒ 00:42:30.200 Casie Aviles: Totally.
564 00:42:31.890 ⇒ 00:42:32.810 Samuel Roberts: Sound good?
565 00:42:33.460 ⇒ 00:42:35.080 Mustafa Raja: Yes, I just wanna ask.
566 00:42:35.080 ⇒ 00:42:35.440 Casie Aviles: Yeah.
567 00:42:35.880 ⇒ 00:42:41.360 Casie Aviles: Mustafa, for the, like, the… this is the same work that you did, right? Which…
568 00:42:41.530 ⇒ 00:42:46.619 Casie Aviles: part of the platform can I take a look at, just in case if I want?
569 00:42:47.120 ⇒ 00:42:47.890 Samuel Roberts: Oh yeah, sweet.
570 00:42:48.850 ⇒ 00:42:51.360 Mustafa Raja: For… for the…
571 00:42:52.710 ⇒ 00:42:53.349 Casie Aviles: for demonstration.
572 00:42:53.350 ⇒ 00:42:54.890 Mustafa Raja: For the general drag, right?
573 00:42:54.890 ⇒ 00:42:55.430 Casie Aviles: Yes, yes.
574 00:42:56.010 ⇒ 00:42:57.020 Mustafa Raja: Generally…
575 00:42:57.220 ⇒ 00:42:59.449 Samuel Roberts: There’s a PR that should have all the content.
576 00:42:59.900 ⇒ 00:43:00.750 Samuel Roberts: I think.
577 00:43:00.860 ⇒ 00:43:01.590 Casie Aviles: I see.
578 00:43:01.590 ⇒ 00:43:09.989 Mustafa Raja: Yeah, if you go into the pull request, Yeah, and then in the…
579 00:43:10.430 ⇒ 00:43:11.200 Casie Aviles: This one?
580 00:43:14.680 ⇒ 00:43:15.690 Samuel Roberts: It got a weird name.
581 00:43:15.690 ⇒ 00:43:16.140 Mustafa Raja: And then…
582 00:43:16.140 ⇒ 00:43:17.870 Samuel Roberts: Stuff that, they’ve added.
583 00:43:17.870 ⇒ 00:43:25.710 Mustafa Raja: Yeah, so I work, worked… It’s… Hmm, this is…
584 00:43:29.170 ⇒ 00:43:30.040 Mustafa Raja: Okay.
585 00:43:33.760 ⇒ 00:43:35.970 Casie Aviles: This one, right? You created this one.
586 00:43:36.400 ⇒ 00:43:37.290 Casie Aviles: Or…
587 00:43:37.480 ⇒ 00:43:38.320 Samuel Roberts: The master agent.
588 00:43:38.320 ⇒ 00:43:39.140 Mustafa Raja: Yeah.
589 00:43:40.090 ⇒ 00:43:45.009 Casie Aviles: Okay, yeah, I’ll just use… I’ll look at this as an example.
590 00:43:45.970 ⇒ 00:43:53.350 Casie Aviles: I think, yeah, yeah, that definitely gives some clarity on what needs to be prioritized.
591 00:43:53.840 ⇒ 00:44:04.450 Casie Aviles: I’m like, yeah, So I’ll work on getting the… the master… Agent… What else?
592 00:44:05.080 ⇒ 00:44:07.670 Casie Aviles: I think, yeah, for, for, like, the…
593 00:44:07.870 ⇒ 00:44:11.699 Casie Aviles: the actual… the stuff that we need from ABC’s side.
594 00:44:12.370 ⇒ 00:44:19.909 Casie Aviles: I think… There might be just some, areas there where I’m not super sure yet.
595 00:44:21.580 ⇒ 00:44:26.820 Casie Aviles: But at the minimum, like, I think what I got out of the call with them was…
596 00:44:26.820 ⇒ 00:44:27.490 Samuel Roberts: Oh, yeah.
597 00:44:27.910 ⇒ 00:44:33.460 Casie Aviles: getting… at least, like, different environments…
598 00:44:33.760 ⇒ 00:44:36.400 Casie Aviles: For, like, staging, dev, and prod.
599 00:44:37.670 ⇒ 00:44:44.299 Casie Aviles: I can follow up with him, as needed, but… Yeah, I think…
600 00:44:45.090 ⇒ 00:44:49.950 Casie Aviles: If there’s anything I missed there, yeah, let me know.
601 00:44:50.520 ⇒ 00:44:51.940 Samuel Roberts: Okay.
602 00:44:51.990 ⇒ 00:44:57.330 Casie Aviles: I also kind of talked briefly about What else,
603 00:44:57.520 ⇒ 00:45:05.810 Casie Aviles: the, like, the database, or, like, the data warehouse, the data warehouse being for, like, the one Utam’s working on, the discovery.
604 00:45:05.810 ⇒ 00:45:06.410 Samuel Roberts: Yep.
605 00:45:07.720 ⇒ 00:45:13.790 Casie Aviles: And then the database, potentially, we could move… It has Cloud SQL…
606 00:45:14.350 ⇒ 00:45:17.090 Casie Aviles: I think that’s what Google has.
607 00:45:17.450 ⇒ 00:45:21.490 Casie Aviles: I think the preference, basically, is that each…
608 00:45:21.680 ⇒ 00:45:27.010 Casie Aviles: a team prefers it to be, like, in… within their Google environment as much as possible.
609 00:45:27.010 ⇒ 00:45:27.620 Samuel Roberts: Yeah.
610 00:45:27.870 ⇒ 00:45:28.930 Samuel Roberts: That makes sense.
611 00:45:29.290 ⇒ 00:45:34.249 Casie Aviles: So I’m not… Yeah. Yeah, I just wasn’t sure, like, how…
612 00:45:34.650 ⇒ 00:45:43.049 Casie Aviles: how hosting was gonna work, like, I told them, like, internally, what we did was we had Heroku, we use Heroku to deploy.
613 00:45:43.810 ⇒ 00:45:49.329 Casie Aviles: So I wasn’t sure, like, how… that part, I wasn’t sure in… with GCP, how that.
614 00:45:49.330 ⇒ 00:45:53.389 Samuel Roberts: Yeah, I think… I think we’re… we’ll probably… I mean…
615 00:45:56.460 ⇒ 00:46:03.410 Samuel Roberts: I think we’ll eventually have our own GCP for, like, Dev, probably?
616 00:46:04.490 ⇒ 00:46:09.690 Samuel Roberts: And then we’ll also get access to theirs for, like, a dev and staging and prod.
617 00:46:09.820 ⇒ 00:46:13.980 Samuel Roberts: But for now, I’m less worried about that until we get something.
618 00:46:13.980 ⇒ 00:46:14.350 Casie Aviles: Okay.
619 00:46:14.350 ⇒ 00:46:25.079 Samuel Roberts: is, like, working. Because, like, if we can all run it locally, or we push to Heroku for now, just for testing, I’m not as worried about getting it to eventually run on GCP.
620 00:46:25.080 ⇒ 00:46:25.640 Casie Aviles: Okay.
621 00:46:26.490 ⇒ 00:46:36.130 Samuel Roberts: But I, I mean, it’s definitely something to be aware. I’m glad we had the conversation with Tim, because I know that can sometimes take a minute, because, like, there’s, you know, he’s got to jump through hoops, too, to get stuff out, so…
622 00:46:36.770 ⇒ 00:46:44.390 Samuel Roberts: I would say for now, don’t stress about that too much. Worry about getting this, like, locally running as, as,
623 00:46:45.170 ⇒ 00:46:49.579 Samuel Roberts: you know, on par with what we have, if not better for now. Make sure that this is…
624 00:46:49.790 ⇒ 00:46:51.270 Samuel Roberts: So I have to go down.
625 00:46:51.470 ⇒ 00:46:53.310 Samuel Roberts: But yeah, I think… I think…
626 00:46:53.550 ⇒ 00:47:01.320 Samuel Roberts: I’m gonna catch up on that meeting you had with him later today, and just get myself up to speed, but I’m not too worried, I mean…
627 00:47:01.460 ⇒ 00:47:04.790 Samuel Roberts: I think… I think we can get there pretty…
628 00:47:05.130 ⇒ 00:47:07.770 Samuel Roberts: Well, it just might be a little bit of,
629 00:47:08.510 ⇒ 00:47:12.179 Samuel Roberts: you know, I’m not as familiar with GCP as I am AWS stuff, so we’ll see.
630 00:47:13.510 ⇒ 00:47:14.209 Casie Aviles: Okay, no problem.
631 00:47:14.210 ⇒ 00:47:18.030 Samuel Roberts: It’s all the same stuff, it’s just named different things, you know, at the end of the day.
632 00:47:21.070 ⇒ 00:47:30.529 Samuel Roberts: I’m not too worried about moving everything over if we need to, but for now, I want to just worry about running this locally, getting the master dev stuff, chatting with that, and dealing with it that way.
633 00:47:31.580 ⇒ 00:47:40.549 Casie Aviles: Alright, and you’ll also be able to test it, right? Even if it’s not, like, if it’s just local for now? Like, if I just push it on to GitHub.
634 00:47:40.550 ⇒ 00:47:44.960 Samuel Roberts: Yeah, I think, yeah, for now, we’ll just pull it down as long as we have the right keys set up.
635 00:47:45.320 ⇒ 00:47:53.110 Samuel Roberts: For whatever model we’re using, and the right connection strings and stuff, which we can make sure to share on 1Password will be fine.
636 00:47:53.540 ⇒ 00:47:56.239 Casie Aviles: Okay, alright, I’ll focus on that.
637 00:47:56.680 ⇒ 00:48:01.670 Samuel Roberts: Yeah, I think for at least, like, the next week or two, that’s the best way to go with it, until we get something that’s, like…
638 00:48:03.380 ⇒ 00:48:07.749 Samuel Roberts: better than Andy, you know what I mean? Like, then we can be like, alright, now let’s host it, and
639 00:48:08.120 ⇒ 00:48:16.759 Samuel Roberts: get a staging environment set up that, you know, Janiece can test it, kind of thing, and Yvette can test it, and show them that it’s improved, you know?
640 00:48:18.140 ⇒ 00:48:18.930 Casie Aviles: Alright.
641 00:48:19.670 ⇒ 00:48:21.770 Casie Aviles: Yes, thank you, thank you. This is…
642 00:48:21.770 ⇒ 00:48:22.120 Samuel Roberts: Mexico.
643 00:48:22.120 ⇒ 00:48:22.740 Casie Aviles: fool.
644 00:48:23.370 ⇒ 00:48:29.089 Samuel Roberts: Yeah, I’m gonna do a little bit of research, maybe, into contextual, or maybe some other pools that might…
645 00:48:29.250 ⇒ 00:48:33.120 Samuel Roberts: Do, like, a little bit of this for us.
646 00:48:33.330 ⇒ 00:48:36.420 Samuel Roberts: Without having to reinvent the wheel,
647 00:48:37.180 ⇒ 00:48:40.710 Samuel Roberts: I’m wondering… I don’t know exactly what Monster does in terms of, like…
648 00:48:42.960 ⇒ 00:48:47.190 Samuel Roberts: Rag stuff out of the box for, like, the ingestion, at least, but…
649 00:48:47.490 ⇒ 00:48:51.250 Samuel Roberts: I’m gonna look up a little bit and just see what… what plays well together here.
650 00:48:53.750 ⇒ 00:48:54.530 Casie Aviles: Oh boy.
651 00:48:55.450 ⇒ 00:48:58.920 Samuel Roberts: Because the other side of it is… you know.
652 00:48:59.490 ⇒ 00:49:00.800 Samuel Roberts: If we have a…
653 00:49:02.910 ⇒ 00:49:09.449 Samuel Roberts: Right now, it’s kind of, like, it’s the routing agent, obviously, the master agent, and then the rag, but if we break that up into different tools that are, like.
654 00:49:10.090 ⇒ 00:49:16.849 Samuel Roberts: it knows what to look up from which document or something. Like, we could… we could structure it a little bit differently, we have a little more flexibility with Mastra.
655 00:49:17.130 ⇒ 00:49:21.299 Samuel Roberts: And then, even letting that figure out what,
656 00:49:21.540 ⇒ 00:49:26.729 Samuel Roberts: Which tools to call, giving it a, you know… it doesn’t need to be a linear process, it can be a little bit of a, like…
657 00:49:29.760 ⇒ 00:49:31.460 Samuel Roberts: what’s the word? You know?
658 00:49:32.910 ⇒ 00:49:40.000 Samuel Roberts: The agent can make the determination of, like, which… which tool to call for which document, or blah blah blah, like, we can figure all that out.
659 00:49:40.120 ⇒ 00:49:42.899 Samuel Roberts: But for now, I just want to get the basic agent, like.
660 00:49:43.280 ⇒ 00:49:50.630 Samuel Roberts: Into Mastra, and then run from there and make sure it’s working, because if for some reason we hit a roadblock here, I’d like to know that sooner rather than later.
661 00:49:52.280 ⇒ 00:49:53.000 Casie Aviles: Okay.
662 00:49:54.410 ⇒ 00:49:55.110 Samuel Roberts: Alright.
663 00:49:57.410 ⇒ 00:49:59.199 Casie Aviles: I think that’s all from me.
664 00:49:59.200 ⇒ 00:49:59.590 Samuel Roberts: Okay.
665 00:49:59.740 ⇒ 00:50:01.060 Casie Aviles: This was helpful.
666 00:50:03.110 ⇒ 00:50:10.370 Samuel Roberts: Yeah, don’t be afraid to, like, ping me or Mustafa, who’s done a little bit with the monster now, like, whatever questions you have, get it up and running, like…
667 00:50:10.660 ⇒ 00:50:11.440 Samuel Roberts: You know.
668 00:50:11.690 ⇒ 00:50:14.179 Samuel Roberts: Also, lean on the monster docs, they look pretty good.
669 00:50:14.290 ⇒ 00:50:17.000 Samuel Roberts: And when I played with it, it was very helpful, so…
670 00:50:18.720 ⇒ 00:50:19.340 Casie Aviles: Okay.
671 00:50:20.070 ⇒ 00:50:20.860 Samuel Roberts: Alright.
672 00:50:23.490 ⇒ 00:50:24.130 Samuel Roberts: Anything wrong?
673 00:50:24.130 ⇒ 00:50:25.200 Casie Aviles: Thank you, guys.
674 00:50:27.180 ⇒ 00:50:28.890 Casie Aviles: Yeah, that’s all from me.
675 00:50:29.450 ⇒ 00:50:29.990 Samuel Roberts: Okay.
676 00:50:30.150 ⇒ 00:50:31.240 Samuel Roberts: Yeah, once you get…
677 00:50:31.240 ⇒ 00:50:32.559 Mustafa Raja: Riku.
678 00:50:33.910 ⇒ 00:50:34.570 Samuel Roberts: Sorry?
679 00:50:34.960 ⇒ 00:50:40.199 Mustafa Raja: Yeah, I was saying, you could please assign me… assign me tasks regarding this.
680 00:50:40.200 ⇒ 00:50:46.410 Samuel Roberts: Yeah, that’s what I was saying, once you get, like, the basic Mastra, stuff scaffolded out, maybe…
681 00:50:46.870 ⇒ 00:50:50.239 Samuel Roberts: Push it up and message us, and we can, you know, maybe.
682 00:50:50.240 ⇒ 00:50:50.830 Mustafa Raja: Yeah.
683 00:50:50.830 ⇒ 00:50:53.920 Samuel Roberts: jump in. And figure it out that way.
684 00:50:54.440 ⇒ 00:50:55.000 Mustafa Raja: Yeah.
685 00:50:55.420 ⇒ 00:50:56.100 Casie Aviles: Okay.
686 00:50:58.200 ⇒ 00:50:59.010 Mustafa Raja: Thank you!
687 00:50:59.620 ⇒ 00:51:00.909 Samuel Roberts: Alright, yeah, thank you both.
688 00:51:01.480 ⇒ 00:51:02.040 Mustafa Raja: Yep.
689 00:51:02.460 ⇒ 00:51:03.210 Casie Aviles: Thank you.
690 00:51:03.240 ⇒ 00:51:04.380 Samuel Roberts: Alright, talk to y’all later.
691 00:51:04.710 ⇒ 00:51:05.620 Mustafa Raja: Yeah, bye!