Meeting Title: Brainforge Project Check-in and Updates Date: 2025-09-26 Meeting participants: Micah Gallardo, Samuel Roberts, Mustafa Raja
WEBVTT
1 00:00:22.530 ⇒ 00:00:23.630 Samuel Roberts: Hello?
2 00:00:23.860 ⇒ 00:00:25.480 Micah Gallardo: Hey, Sam, how’s it going?
3 00:00:25.540 ⇒ 00:00:27.820 Samuel Roberts: Good. How about yourself?
4 00:00:27.820 ⇒ 00:00:30.729 Micah Gallardo: Doing great. Doing good.
5 00:00:30.960 ⇒ 00:00:31.770 Samuel Roberts: Cuckoo.
6 00:00:31.990 ⇒ 00:00:32.990 Micah Gallardo: Yeah, man.
7 00:00:34.440 ⇒ 00:00:40.610 Samuel Roberts: I know Mustaf should be here in a minute. I was just on a call with him, so we were just finishing up, and…
8 00:00:40.830 ⇒ 00:00:43.059 Micah Gallardo: Okay, sounds good, sounds good.
9 00:00:43.060 ⇒ 00:00:46.090 Samuel Roberts: Yeah, how’s your week been?
10 00:00:46.390 ⇒ 00:00:50.789 Micah Gallardo: Good, it’s been a good week, a nice… Easy week, clean week.
11 00:00:50.790 ⇒ 00:00:51.530 Samuel Roberts: Oh, that’s…
12 00:00:51.770 ⇒ 00:01:06.079 Micah Gallardo: Yes. I have a, so I do these interlude hours sort of, like, part-time-ish. We’re sort of folding in for me. My full-time job is at a church right now, so I do a lot of different, very… I wear many hats.
13 00:01:06.080 ⇒ 00:01:06.780 Samuel Roberts: Yeah.
14 00:01:06.780 ⇒ 00:01:09.529 Micah Gallardo: It’s similar to being at a startup.
15 00:01:09.530 ⇒ 00:01:10.130 Samuel Roberts: Yeah.
16 00:01:10.130 ⇒ 00:01:16.020 Micah Gallardo: a lot of different things. So some weeks are busy and crazy, and other weeks are kind of like…
17 00:01:16.400 ⇒ 00:01:19.409 Micah Gallardo: Sitting, twiddling your thumbs, you’re like, alright, anything you’re gonna do?
18 00:01:19.410 ⇒ 00:01:21.809 Samuel Roberts: Yeah, you need a good balance of those weeks.
19 00:01:21.810 ⇒ 00:01:26.060 Micah Gallardo: Yeah, that’s for sure. So, yeah, it was a pretty chill week this week, it’s been good.
20 00:01:26.060 ⇒ 00:01:27.000 Samuel Roberts: Good, good.
21 00:01:27.000 ⇒ 00:01:29.159 Micah Gallardo: Yeah, yeah, how are you guys doing? How’s it going?
22 00:01:29.690 ⇒ 00:01:38.649 Samuel Roberts: It’s going well. This was a pretty busy week for me. I had a lot of, like, backlog cleaning stuff, because I joined Brainforge, like, mid-July?
23 00:01:38.650 ⇒ 00:01:39.670 Micah Gallardo: Oh, okay, okay.
24 00:01:39.870 ⇒ 00:01:48.359 Samuel Roberts: And so, there’s a lot of stuff in linear in the backlog that I was just trying to, like, sift through and figure out, like, is this still relevant now with the.
25 00:01:48.360 ⇒ 00:01:48.710 Micah Gallardo: Okay.
26 00:01:48.710 ⇒ 00:02:02.349 Samuel Roberts: And so, I spent a few hours the other night just going through, like, a hundred-something tickets and just figuring out if I knew it, and I was like, whoa, it’s how late now? Like, I just… I kind of lost track of time thing, but besides that, like, it’s been a good week.
27 00:02:02.350 ⇒ 00:02:02.700 Micah Gallardo: Nice.
28 00:02:02.700 ⇒ 00:02:04.319 Samuel Roberts: A lot of things moving, so…
29 00:02:04.320 ⇒ 00:02:07.559 Micah Gallardo: Nice, very good. Very, very good. What’s up, Mustafa?
30 00:02:08.410 ⇒ 00:02:09.850 Mustafa Raja: Yeah, doing good, how are you?
31 00:02:09.850 ⇒ 00:02:12.019 Micah Gallardo: I’m doing well, doing well, good to see you again, brother.
32 00:02:13.240 ⇒ 00:02:13.929 Mustafa Raja: Yeah, I think…
33 00:02:13.930 ⇒ 00:02:15.800 Micah Gallardo: Yeah, good, very good.
34 00:02:15.900 ⇒ 00:02:17.510 Samuel Roberts: So, so, stoked.
35 00:02:17.990 ⇒ 00:02:23.519 Samuel Roberts: Yeah, so let’s, let’s, I think Mustafa has a couple updates that he made based on your feedback.
36 00:02:23.960 ⇒ 00:02:26.130 Samuel Roberts: And then I’m just curious to, like.
37 00:02:26.460 ⇒ 00:02:29.750 Samuel Roberts: hear what else you’ve been… how it’s been looking? Yeah.
38 00:02:29.750 ⇒ 00:02:30.290 Micah Gallardo: Yeah, yeah, yeah.
39 00:02:30.290 ⇒ 00:02:35.189 Samuel Roberts: We have some other ideas about things, like the decks and stuff that I want to kind of talk through a little bit, but…
40 00:02:35.190 ⇒ 00:02:35.760 Micah Gallardo: Cool, okay.
41 00:02:35.760 ⇒ 00:02:37.879 Samuel Roberts: Would definitely just love to hear, like, you…
42 00:02:38.120 ⇒ 00:02:40.140 Samuel Roberts: talk about the experience, because, you know, I want to…
43 00:02:40.140 ⇒ 00:02:40.680 Micah Gallardo: Yeah.
44 00:02:40.830 ⇒ 00:02:43.480 Micah Gallardo: hear it from your, your, your voice. Yes, yes.
45 00:02:43.480 ⇒ 00:02:48.540 Samuel Roberts: So yeah, I guess, Mustafa, you want to let him know what’s new… what’s now live?
46 00:02:48.540 ⇒ 00:02:49.460 Mustafa Raja: Yeah, yeah, yeah.
47 00:02:49.810 ⇒ 00:02:52.349 Mustafa Raja: So let me share my screen.
48 00:02:53.970 ⇒ 00:02:55.970 Mustafa Raja: Okay.
49 00:02:56.330 ⇒ 00:03:00.509 Mustafa Raja: Okay, so the rubrics will now be as a heading over here.
50 00:03:01.930 ⇒ 00:03:06.300 Mustafa Raja: the notion when we go there. Now, I did see…
51 00:03:06.760 ⇒ 00:03:15.799 Mustafa Raja: that you have this hyperstructure mentioned, right? Yeah, yeah. So this is something… so do you want your rubrics in this structure?
52 00:03:16.720 ⇒ 00:03:17.500 Micah Gallardo: Yes.
53 00:03:17.500 ⇒ 00:03:19.120 Samuel Roberts: Right now, it’s just the… okay.
54 00:03:19.120 ⇒ 00:03:21.449 Micah Gallardo: It’s just, like, this summary, which is.
55 00:03:21.450 ⇒ 00:03:21.829 Samuel Roberts: Oh my god.
56 00:03:21.830 ⇒ 00:03:29.140 Micah Gallardo: too. I think having some of the feedback, kind of seeing it all right away, is also very good.
57 00:03:30.060 ⇒ 00:03:31.950 Micah Gallardo: So, yeah, something like that would be…
58 00:03:31.950 ⇒ 00:03:32.719 Samuel Roberts: Okay, okay.
59 00:03:32.720 ⇒ 00:03:34.140 Micah Gallardo: That exact structure, that’s gonna be perfect.
60 00:03:34.140 ⇒ 00:03:34.810 Mustafa Raja: Yeah.
61 00:03:35.000 ⇒ 00:03:39.539 Mustafa Raja: Yeah, it is currently, like, this biggest bill interest only returns its numbers, so we can…
62 00:03:39.540 ⇒ 00:03:40.050 Samuel Roberts: Yeah.
63 00:03:40.050 ⇒ 00:03:56.569 Mustafa Raja: some AI steps in our NHN workflow, and what that is going to do is it… it is going to enable us to, make the structure more towards this. Yeah. Yeah, so for context…
64 00:03:56.570 ⇒ 00:04:10.339 Samuel Roberts: NN runs this workflow, this, like, iterative process with several different agents, and then it goes to an eval platform called Brain Trust that we have, like, some other prompts to do that kind of grading system.
65 00:04:10.720 ⇒ 00:04:21.230 Samuel Roberts: So that’s why it’s just spitting out numbers, basically, right now. I think this would become probably a few other agents that do the rubric analysis more than just an eval analysis.
66 00:04:21.230 ⇒ 00:04:22.069 Micah Gallardo: Got it.
67 00:04:22.079 ⇒ 00:04:22.469 Samuel Roberts: So, just.
68 00:04:22.470 ⇒ 00:04:23.210 Mustafa Raja: Okay.
69 00:04:23.210 ⇒ 00:04:27.839 Samuel Roberts: We were just… the reason we were asking is because, like, we were just trying to prioritize, like, what to work on next, and is this…
70 00:04:27.840 ⇒ 00:04:28.360 Micah Gallardo: Right.
71 00:04:28.360 ⇒ 00:04:31.120 Samuel Roberts: Is this the thing you… that would, like, level this up again?
72 00:04:31.120 ⇒ 00:04:32.779 Micah Gallardo: Right, right, right, right.
73 00:04:32.780 ⇒ 00:04:33.739 Samuel Roberts: That makes sense.
74 00:04:33.960 ⇒ 00:04:34.560 Micah Gallardo: Yeah.
75 00:04:34.730 ⇒ 00:04:44.479 Micah Gallardo: I think what we liked about the, we’ve been running that deck eval just as, like, a separate prompt in, like, on the Cloud platform.
76 00:04:44.740 ⇒ 00:04:45.700 Samuel Roberts: in.
77 00:04:45.750 ⇒ 00:04:58.849 Micah Gallardo: We liked it for two reasons. The first thing we liked is… it’s, like, LLM as judge, like, built in right away. So before we even start looking at any of the copy, we get a rough idea of where it’s at, so that we can
78 00:04:59.100 ⇒ 00:05:13.109 Micah Gallardo: hit certain pieces, like, if there’s any assumptions that it’s flagged, or if it just… it’s… it’s giving weak output, but maybe just because there’s not enough data, or something like that, like, it gives us an idea of where that’s at. And…
79 00:05:13.170 ⇒ 00:05:25.430 Micah Gallardo: We’ve also used it to review decks that we’ve already sort of put copy out for. We wouldn’t do it in this case, because it’s built in to just getting a V1 out.
80 00:05:25.760 ⇒ 00:05:45.160 Micah Gallardo: But it was… in testing, that’s what we were using it for previously, and kind of ran some decks through it and saw, like, oh, this is actually really good for review as well. Yeah. But anyway, all that to say, it’s nice to see, like, where the output is at when it gets…
81 00:05:45.300 ⇒ 00:05:49.140 Micah Gallardo: when it’s all in Notion. It’s nice to see that kind of right away, so we don’t.
82 00:05:49.140 ⇒ 00:05:50.100 Samuel Roberts: Right, right.
83 00:05:50.100 ⇒ 00:05:58.359 Micah Gallardo: oh, this is all pretty good, like, it should be… it feels good, like, it’s done. Yeah. Like, it did its job, it’s never usually… it’s never that, and .
84 00:05:58.360 ⇒ 00:06:02.880 Samuel Roberts: That’s… that’s actually, like, really good insight for, like, general AI output. Like, sometimes it’s, like.
85 00:06:02.880 ⇒ 00:06:03.240 Micah Gallardo: Yeah.
86 00:06:03.240 ⇒ 00:06:10.760 Samuel Roberts: that seems good enough, but you miss something that it hallucinated, or something like that, so that’s a good… that’s a good step, yeah. Yeah, making that in would be…
87 00:06:11.000 ⇒ 00:06:12.600 Samuel Roberts: I can see how that’s very helpful.
88 00:06:12.810 ⇒ 00:06:13.400 Micah Gallardo: Yeah.
89 00:06:13.900 ⇒ 00:06:15.360 Samuel Roberts: Okay, sweet.
90 00:06:15.360 ⇒ 00:06:18.830 Micah Gallardo: So that’s the goal with the rubric. Cool. Go ahead, Mustafa.
91 00:06:19.170 ⇒ 00:06:22.339 Mustafa Raja: Yeah, so another example is to make this shorter, so this is.
92 00:06:22.340 ⇒ 00:06:23.250 Micah Gallardo: Yes.
93 00:06:23.760 ⇒ 00:06:24.310 Mustafa Raja: Yes.
94 00:06:24.310 ⇒ 00:06:24.980 Samuel Roberts: Oh, yeah.
95 00:06:24.980 ⇒ 00:06:27.890 Mustafa Raja: Anything else we did cover?
96 00:06:28.320 ⇒ 00:06:31.059 Mustafa Raja: Yeah, this one should be working now.
97 00:06:31.320 ⇒ 00:06:31.930 Micah Gallardo: Cool.
98 00:06:32.820 ⇒ 00:06:35.189 Samuel Roberts: Yeah, what was the issue with that, Mustafa? What was it?
99 00:06:35.190 ⇒ 00:06:41.819 Mustafa Raja: Oh, yeah, so… what happened here was that no files were attached in this.
100 00:06:41.920 ⇒ 00:06:43.010 Mustafa Raja: And…
101 00:06:43.010 ⇒ 00:06:43.880 Samuel Roberts: Right, right.
102 00:06:43.880 ⇒ 00:06:46.789 Mustafa Raja: I had missed… In the workflow.
103 00:06:46.920 ⇒ 00:06:49.880 Mustafa Raja: And so I built that pipeline.
104 00:06:49.880 ⇒ 00:06:50.390 Samuel Roberts: Cool.
105 00:06:50.390 ⇒ 00:06:54.649 Mustafa Raja: That if we don’t have, this, the things… Right, just to use the…
106 00:06:54.650 ⇒ 00:06:58.789 Samuel Roberts: That’s right, I remember you saying that, yeah. That was hopefully okay. Yeah.
107 00:06:58.790 ⇒ 00:07:02.489 Mustafa Raja: Yeah, this is okay now. So.
108 00:07:03.440 ⇒ 00:07:11.329 Samuel Roberts: What else? Yeah, rubrics, yeah, this is all. Okay, cool. Yeah, so I think all of those are now in, so definitely keep testing that.
109 00:07:11.330 ⇒ 00:07:11.820 Micah Gallardo: Yeah, that’s great.
110 00:07:11.820 ⇒ 00:07:16.650 Samuel Roberts: And we’ll… we’ll definitely work on the, like, improved rubric output.
111 00:07:16.650 ⇒ 00:07:17.330 Micah Gallardo: Okay.
112 00:07:18.310 ⇒ 00:07:19.040 Samuel Roberts: I don’t…
113 00:07:19.390 ⇒ 00:07:28.200 Samuel Roberts: I don’t have a good sense of, like, how much that… because right now, it’s just hitting brain trust and getting a number, and that’s partially to, like, track how good these are over time as well.
114 00:07:28.200 ⇒ 00:07:28.770 Micah Gallardo: Okay.
115 00:07:28.970 ⇒ 00:07:33.889 Samuel Roberts: So that’s, like, a number we can point to and be like, okay, we have improved when we’ve changed this prompt kind of thing.
116 00:07:33.890 ⇒ 00:07:35.230 Micah Gallardo: Yeah, yeah, yeah, yeah.
117 00:07:35.230 ⇒ 00:07:47.359 Samuel Roberts: But the feedback will be a little more complex than just hitting range, so we’ll see how that… I don’t know, Mousafa, if you have a good idea of how best to, like, get that structure going, but I imagine
118 00:07:47.740 ⇒ 00:07:53.140 Samuel Roberts: Yeah. It won’t be crazy, but it might be a little chunk of work for that, but worth it, probably, so…
119 00:07:53.140 ⇒ 00:07:53.810 Mustafa Raja: Yeah.
120 00:07:54.160 ⇒ 00:07:56.689 Samuel Roberts: Cool. Yeah.
121 00:07:56.960 ⇒ 00:07:59.700 Samuel Roberts: other… other… so that’s… that’s the update, I guess.
122 00:07:59.700 ⇒ 00:08:02.829 Micah Gallardo: Yeah, yeah, yeah, yeah, it’s a good update. I think, too, just to…
123 00:08:03.160 ⇒ 00:08:08.569 Micah Gallardo: give you some peace of mind. I think the rubric thing doesn’t… it’s not entirely urgent.
124 00:08:08.740 ⇒ 00:08:16.699 Micah Gallardo: That’s, like, it’s all text, and it’s a copy-paste into Claude, and then run the prompts that we already have. So, if that…
125 00:08:17.520 ⇒ 00:08:32.829 Micah Gallardo: I would… I would say having the… the positive-negative rationale that’s already there, and the rubric that’s already being run right now, I think is… is plenty, again, just to have, like, at a glance. You can…
126 00:08:32.830 ⇒ 00:08:33.719 Samuel Roberts: Yeah, okay, that’s.
127 00:08:33.720 ⇒ 00:08:37.620 Micah Gallardo: through that feedback, and that’s, like, pretty good, because I…
128 00:08:37.620 ⇒ 00:08:37.950 Samuel Roberts: Okay.
129 00:08:37.950 ⇒ 00:08:40.950 Micah Gallardo: I was running into some stuff this week with Claude that was, like.
130 00:08:41.289 ⇒ 00:09:00.940 Micah Gallardo: it was smearing a lot of stuff and needed… I just needed to know, like, okay, where is this at in general? Like, I didn’t need to know all the details quite yet, but I just needed to know, like, okay, where are you starting to smear some of the output? So if that’s all that it is right now, I think that’s perfectly fine, and then if we wanted to…
131 00:09:01.200 ⇒ 00:09:10.380 Micah Gallardo: maybe… like, what I can do is… is take a look at this and see, like, is this detail all that necessary? Because it is quite detailed. Yeah.
132 00:09:10.500 ⇒ 00:09:11.340 Micah Gallardo: And…
133 00:09:11.340 ⇒ 00:09:14.279 Samuel Roberts: Yeah, if there’s specific things that would be, like, more high leverage, like.
134 00:09:14.280 ⇒ 00:09:14.910 Micah Gallardo: Yeah.
135 00:09:14.910 ⇒ 00:09:20.640 Samuel Roberts: Oh, the numbers are okay, but we definitely want, like, to check this, or action items, or whatever.
136 00:09:20.640 ⇒ 00:09:21.850 Micah Gallardo: Yeah, yeah.
137 00:09:21.850 ⇒ 00:09:24.959 Samuel Roberts: We can prioritize what we start adding to the output.
138 00:09:24.960 ⇒ 00:09:25.290 Micah Gallardo: Yeah.
139 00:09:25.790 ⇒ 00:09:28.599 Micah Gallardo: Yeah, that’s the other thing I want to talk about. Now that we have the decks.
140 00:09:28.600 ⇒ 00:09:31.350 Samuel Roberts: Trying to probably figure out, like, where to… Where to focus.
141 00:09:31.350 ⇒ 00:09:32.020 Micah Gallardo: Yes.
142 00:09:32.020 ⇒ 00:09:32.430 Samuel Roberts: Yes.
143 00:09:32.430 ⇒ 00:09:33.290 Micah Gallardo: Yes, yes.
144 00:09:33.290 ⇒ 00:09:36.710 Samuel Roberts: Like, for… for improvement overall,
145 00:09:36.710 ⇒ 00:09:37.530 Micah Gallardo: Right.
146 00:09:39.660 ⇒ 00:09:43.719 Samuel Roberts: But I think, yeah, I think if you can, like, highlight a few things, we can definitely, like, start the…
147 00:09:44.150 ⇒ 00:09:46.770 Samuel Roberts: new pipeline for that sort of thing, and then add to it.
148 00:09:46.770 ⇒ 00:09:47.210 Micah Gallardo: That’s fine.
149 00:09:47.780 ⇒ 00:09:48.410 Samuel Roberts: Sure.
150 00:09:48.410 ⇒ 00:09:48.990 Micah Gallardo: Yeah.
151 00:09:49.700 ⇒ 00:09:55.920 Samuel Roberts: But yeah, that’s all. I would say, yeah, so what other, like, I mean, any other thoughts besides what you mentioned? Obviously, like…
152 00:09:55.920 ⇒ 00:10:04.450 Micah Gallardo: Yeah, well, the iterative… the iterating thing that you mentioned is… that… I think that’s the most important thing that we’re… we want to make sure that
153 00:10:05.240 ⇒ 00:10:19.550 Micah Gallardo: is happening as we use the agent, so that’s probably the most important thing to tackle. So, I’d say in my mind, like, I can definitely take a look and get this rubric sorted, and then we can take a look at
154 00:10:19.700 ⇒ 00:10:31.949 Micah Gallardo: the iterative process, but on that point, I checked up… I checked out LangFuse, and I was like, whoa, this thing looks super sick. And it was funny, because I was… what was I looking at? I’ve been…
155 00:10:32.550 ⇒ 00:10:45.670 Micah Gallardo: I don’t know if you go… know Lang Chain, and that, so I was… I was taking a look at some of that, and I was like, okay, I was doing it… I was in San Francisco, like, at the beginning of last week, and…
156 00:10:45.670 ⇒ 00:10:53.299 Micah Gallardo: scrolling around some things, I was like, oh, this is kind of cool, and so when you mentioned that you guys are using Langviews, I was like, oh, what? I was kind of just seeing some of this.
157 00:10:53.300 ⇒ 00:10:53.690 Samuel Roberts: Yeah.
158 00:10:53.690 ⇒ 00:11:00.899 Micah Gallardo: I’d love to see, like, get an idea of how it works, and, also, like.
159 00:11:01.690 ⇒ 00:11:16.409 Micah Gallardo: the… everything up to this point has been great, like, kind of like I mentioned, like, everything that I’m experiencing, like, using it, and… like, because the hardest thing for me, just to get it out into the open air, the hardest thing is taking all the output that we have.
160 00:11:16.510 ⇒ 00:11:19.060 Micah Gallardo: And formatting it, and then…
161 00:11:19.160 ⇒ 00:11:28.869 Micah Gallardo: being able to run sort of some… some sort of eval, like, during or after the process is all kind of sorted. So, I think that those two steps
162 00:11:29.050 ⇒ 00:11:37.389 Micah Gallardo: being able to, like, get this formatted into Notion right away is killer, it’s super great, and then because it’s all plain text.
163 00:11:37.770 ⇒ 00:11:53.589 Micah Gallardo: a quick copy-paste into a deck eval, that runs, it’s in Notion, and then having… you could look at those things side by side, and, like, we can do… I can do that for now, and that works perfectly fine. But yeah, I think the output is exactly, sort of, what we’ve…
164 00:11:54.260 ⇒ 00:12:03.629 Micah Gallardo: been aiming for when we’re making decks in the first place, and so I’m… I’m not… I’m not too pressed about it at all. I think the workflow is great.
165 00:12:03.740 ⇒ 00:12:10.670 Samuel Roberts: Okay, cool. Yeah, so, there’s a ton of these, like, AI tooling stuff out there.
166 00:12:10.670 ⇒ 00:12:11.670 Micah Gallardo: Right, right, right, right.
167 00:12:11.670 ⇒ 00:12:23.149 Samuel Roberts: The prompt management is a big thing that a number of platforms offer. We went with LangFuse partly because it has a hook right in N8N for pulling them, otherwise it involves, like, a little more manual.
168 00:12:23.150 ⇒ 00:12:23.990 Micah Gallardo: Got it.
169 00:12:23.990 ⇒ 00:12:35.369 Samuel Roberts: reaching out to things, but, if and when we ever, like, move off of N8N and, like, try to harden some of these up by, like, making them into code, that’s, like, a little more,
170 00:12:35.480 ⇒ 00:12:47.509 Samuel Roberts: observable and traceable, and, like, you can really dig into, like, every LLN call eventually. We’re still evaluating some frameworks in general for, like, what we think is good. They all kind of have the same features, but Langsmith.
171 00:12:47.510 ⇒ 00:12:52.030 Micah Gallardo: Lang Chain, LangGraph, they’re all kind of in that same ecosystem of, like, those tools.
172 00:12:52.030 ⇒ 00:12:53.270 Samuel Roberts: So…
173 00:12:53.670 ⇒ 00:13:01.710 Samuel Roberts: Yeah, LangFuse is really just, like, it’s there, it lets us do the prompt management for free with, I think, 2 users, so I hopped off so you could hop on, basically, but…
174 00:13:01.710 ⇒ 00:13:02.099 Micah Gallardo: No, no.
175 00:13:02.100 ⇒ 00:13:05.840 Samuel Roberts: But besides that, like, we shouldn’t have an issue with that.
176 00:13:05.840 ⇒ 00:13:06.260 Micah Gallardo: Nice.
177 00:13:06.260 ⇒ 00:13:09.620 Samuel Roberts: So… If it works, great,
178 00:13:10.060 ⇒ 00:13:20.799 Samuel Roberts: I think, you know, we can also find another tool. We’ve played around with doing some of our own stuff in Git, like having a GitHub repo for some of our own internal prompts, but that tracking is what’s important, I feel like, you know?
179 00:13:20.800 ⇒ 00:13:21.740 Micah Gallardo: Right, right.
180 00:13:21.740 ⇒ 00:13:25.620 Samuel Roberts: make a change in N8N and not realize what the output is.
181 00:13:25.620 ⇒ 00:13:26.130 Micah Gallardo: Right, right.
182 00:13:26.130 ⇒ 00:13:36.160 Samuel Roberts: And the overall eval is what we’re gonna start really, trying to, like, create a good loop for that, and that’s what these decks, I think, and the inputs of those decks can help with, because now we can…
183 00:13:36.840 ⇒ 00:13:50.570 Samuel Roberts: the idea with the decks is to be able to, like, take those inputs, potentially. Like, we have a lot of ideas to what to do with the decks. One is that we know what, like, good copy looks like, like, what good final copy looks like, so we can try to make it, like, a little more than V1, but, like.
184 00:13:50.750 ⇒ 00:14:02.489 Samuel Roberts: maybe not perfect, obviously, but just kind of get closer to that. And then the other thing is, if we know the inputs that went into making the final deck, we can test our process through that.
185 00:14:02.700 ⇒ 00:14:20.610 Samuel Roberts: and see, like, okay, it’s getting close, but it’s not there. What prompt is not doing it? And then we can make the changes. You can make the changes, we can make the changes in length views, and then test it again on, like, the kind of golden dataset of the good decks, and ideally their inputs, because…
186 00:14:20.610 ⇒ 00:14:21.180 Micah Gallardo: Yeah.
187 00:14:21.180 ⇒ 00:14:24.380 Samuel Roberts: Then we can kind of make that a whole system that, like, whenever we change a prompt.
188 00:14:24.760 ⇒ 00:14:32.089 Samuel Roberts: Rather than, like, run it and see how the output is, and kind of have the rubric, and have, like, your vibe check on it kind of thing, we can…
189 00:14:33.640 ⇒ 00:14:42.160 Samuel Roberts: Be like, is this closer or farther away from what the intended output is, now that we know we have good, like, human-generated output?
190 00:14:42.350 ⇒ 00:14:43.030 Micah Gallardo: Yeah.
191 00:14:43.450 ⇒ 00:14:45.389 Samuel Roberts: So on that, I actually,
192 00:14:45.980 ⇒ 00:14:49.139 Samuel Roberts: I saw the inputs for one of them,
193 00:14:49.450 ⇒ 00:14:54.870 Samuel Roberts: That’s… that’s huge, thank you. That will help us get this kind of system going. Definitely, like.
194 00:14:54.900 ⇒ 00:15:10.699 Samuel Roberts: whatever else you can get us for those inputs is great. Deck-wise, we’re trying to figure out the best way, because I want to do a few things with the deck. One is, like I’m talking about, getting closer to the copy. The other side, and I think we were talking about this last week,
195 00:15:11.590 ⇒ 00:15:14.740 Samuel Roberts: with Bhutam and Matthew was,
196 00:15:15.440 ⇒ 00:15:18.800 Samuel Roberts: What can we enable after this process a little bit?
197 00:15:18.800 ⇒ 00:15:19.359 Micah Gallardo: What’s this?
198 00:15:19.740 ⇒ 00:15:20.580 Samuel Roberts: like…
199 00:15:21.060 ⇒ 00:15:25.070 Samuel Roberts: and this is the other side of it, like, I don’t wanna… I don’t wanna take the human out of the loop, I don’t wanna, like.
200 00:15:25.370 ⇒ 00:15:28.720 Samuel Roberts: You know, the design and creativity stuff that goes into that is, like.
201 00:15:28.830 ⇒ 00:15:33.379 Samuel Roberts: the value add for Interlude, you know what I mean? Like, we don’t want to just, like, machine output everything.
202 00:15:33.380 ⇒ 00:15:34.240 Micah Gallardo: But…
203 00:15:34.250 ⇒ 00:15:38.089 Samuel Roberts: late, like, we actually saw, Sylvia, I think?
204 00:15:38.190 ⇒ 00:15:46.120 Samuel Roberts: Is that her name of the deck? She was doing the deck design, and he was… he brought us into Figma, we saw what she was doing, and we were just trying to figure out, like, what is a good
205 00:15:48.850 ⇒ 00:15:57.250 Samuel Roberts: output we can generate that will help her and not, like… I actually talked with our… one of our designers about, like, what… what would help you make a deck? Like, what…
206 00:15:57.250 ⇒ 00:15:57.850 Micah Gallardo: Hmm.
207 00:15:58.670 ⇒ 00:16:12.220 Samuel Roberts: Besides just the copy, and then having to figure out the layout yourself, would a lo-fi version be good? Would several lo-fi versions be good? Giving you options and ideas to start? I don’t want to box a designer in, but I also want to, like, give a few more constraints that might help guide.
208 00:16:12.590 ⇒ 00:16:12.910 Micah Gallardo: Yeah.
209 00:16:12.910 ⇒ 00:16:16.319 Samuel Roberts: So that’s good with the decks, and we’re gonna start figuring out, like.
210 00:16:16.590 ⇒ 00:16:23.459 Samuel Roberts: if we can get the LM to look at it and figure, like, classify that, like, size or something. The other question I had is, those are…
211 00:16:23.810 ⇒ 00:16:28.429 Samuel Roberts: PDFs that are… I can’t select the text in them when I’m looking at it, so they’re just…
212 00:16:28.950 ⇒ 00:16:34.939 Samuel Roberts: Which is not a huge problem, like, we can OCR, but it just kind of introduces another potential, like.
213 00:16:36.050 ⇒ 00:16:36.430 Micah Gallardo: Yeah.
214 00:16:36.430 ⇒ 00:16:37.410 Samuel Roberts: Area in the system.
215 00:16:37.410 ⇒ 00:16:38.070 Micah Gallardo: Yeah, seriously.
216 00:16:38.070 ⇒ 00:16:42.250 Samuel Roberts: So, are those things in Figma that can be generated differently, or…
217 00:16:42.250 ⇒ 00:16:46.969 Micah Gallardo: Yeah, I mean, so we can do… let me double check what’s possible in Figma.
218 00:16:46.970 ⇒ 00:16:47.520 Samuel Roberts: Okay, yeah.
219 00:16:47.520 ⇒ 00:16:53.240 Micah Gallardo: It’s like… it’s like JPEG, PNG, PDF, and… I think there’s one more that.
220 00:16:53.240 ⇒ 00:16:58.470 Samuel Roberts: Okay, so I… I guess my… yeah, so if the PDF is just an image, that’s… I mean, it seems like it’s just…
221 00:16:58.620 ⇒ 00:17:08.359 Samuel Roberts: there’s different levels of… there’s PDFs that are, like, generated from, like, a Word document, and usually that text is embedded in the file. These seem like they’re embedded as an image more.
222 00:17:08.359 ⇒ 00:17:11.469 Micah Gallardo: So I was just curious if there’s a way to make those…
223 00:17:12.260 ⇒ 00:17:21.879 Samuel Roberts: easier for us to ingest, because again, like, we can do stuff with it, I just… if there’s a way to, at some point in the process, like, spit out one that has
224 00:17:22.010 ⇒ 00:17:23.740 Samuel Roberts: text.
225 00:17:23.740 ⇒ 00:17:24.329 Micah Gallardo: Hmm…
226 00:17:24.339 ⇒ 00:17:28.489 Samuel Roberts: It would definitely, like, eliminate some potential like…
227 00:17:29.220 ⇒ 00:17:29.800 Micah Gallardo: Right.
228 00:17:30.180 ⇒ 00:17:35.210 Samuel Roberts: places where we’re gonna introduce, like, oh, is this word actually getting OCR’d right?
229 00:17:35.210 ⇒ 00:17:35.680 Micah Gallardo: Excellent.
230 00:17:35.680 ⇒ 00:17:36.360 Samuel Roberts: copy.
231 00:17:36.790 ⇒ 00:17:37.470 Micah Gallardo: Right.
232 00:17:37.470 ⇒ 00:17:47.290 Samuel Roberts: again, having them is already good, because we can start to do some stuff and figure out, like, how good is the OCI that we’re gonna use. Like, maybe it’s fine, it’s very clear, it’s not handwriting and stuff, so it’s not as big an issue, but…
233 00:17:47.290 ⇒ 00:17:48.240 Micah Gallardo: Right.
234 00:17:48.930 ⇒ 00:17:51.430 Samuel Roberts: I was just curious what… what outputs are available.
235 00:17:51.430 ⇒ 00:18:01.329 Micah Gallardo: Yeah, so, again, it’s got some PNGs, JPEGs, PDFs, which is what we chose, and then it also has SFG files.
236 00:18:01.470 ⇒ 00:18:04.259 Micah Gallardo: or SVG file, excuse me.
237 00:18:04.260 ⇒ 00:18:14.259 Samuel Roberts: SVG May be good… So, SVGs are, like.
238 00:18:15.110 ⇒ 00:18:32.099 Samuel Roberts: the text is probably still text in those, which might be helpful. That actually might be even easier, because SVGs are really just, like, a big code file that knows how to render the vector images, and so if the text is text, and not, like, a ton of little curves on the fonts and stuff, that would be probably
239 00:18:32.710 ⇒ 00:18:34.820 Samuel Roberts: very helpful. So I would say…
240 00:18:34.970 ⇒ 00:18:38.379 Samuel Roberts: If you could take a look at…
241 00:18:39.810 ⇒ 00:18:44.510 Samuel Roberts: which one do we get the inputs for, actually? That’s probably the best one to try to generate next.
242 00:18:44.510 ⇒ 00:18:51.169 Micah Gallardo: I don’t remember. I got Rails and another one. Yeah, Rails and Standard Practice.
243 00:18:51.850 ⇒ 00:18:56.660 Samuel Roberts: Yeah, so if there are the… If we could try…
244 00:18:56.820 ⇒ 00:19:06.290 Samuel Roberts: the SV… those also might be bigger files, which might be a little bit of an issue, but if it enables us to just, like, pull the copy, and even the SVGs might be easier to process from a…
245 00:19:06.820 ⇒ 00:19:13.270 Samuel Roberts: like, visual point, because we won’t have to rely on an LLM to Like…
246 00:19:13.520 ⇒ 00:19:15.930 Samuel Roberts: Not guess, but, like, look at it and say, like, oh.
247 00:19:15.930 ⇒ 00:19:16.260 Micah Gallardo: Yeah.
248 00:19:16.260 ⇒ 00:19:22.280 Samuel Roberts: kind of. We can actually just be like, oh, okay, we can see where the text is, we can see where these things are, like, reading an SVG is…
249 00:19:22.510 ⇒ 00:19:41.310 Samuel Roberts: is just code for us. So, if possible, to, like, at least generate one of those two, if not both, as the SVG, and let us test that out. Because then we can see, like, oh, the PDF’s fine, we’re not… I’m over… I’m just worrying about things that are not going to be huge deals, or, oh, no, the SVG is going to improve things in another way.
250 00:19:42.260 ⇒ 00:19:52.660 Samuel Roberts: But I would say, like, don’t worry about doing all of them at this point, but those two would be probably the most helpful for us to then figure out, like, okay, we want SVGs, it’s great, or no, PDFs are fine.
251 00:19:52.660 ⇒ 00:19:53.380 Micah Gallardo: For sure.
252 00:19:54.270 ⇒ 00:20:00.259 Samuel Roberts: Because, like, also, like, if it’s just, like, exporting something different from Figma, it’s not adding a ton on your end, I think.
253 00:20:00.260 ⇒ 00:20:00.860 Micah Gallardo: Not at all.
254 00:20:00.860 ⇒ 00:20:08.380 Samuel Roberts: But I don’t want to, like, make… I don’t want to overcomplicate it for you either. If it’s just… if PDFs are what you guys have, that’s fine, but if it’s something you can just click in Figma when you’re done.
255 00:20:08.540 ⇒ 00:20:10.790 Samuel Roberts: Hopefully that would be the best way. So.
256 00:20:10.790 ⇒ 00:20:11.120 Micah Gallardo: Yeah.
257 00:20:11.120 ⇒ 00:20:16.150 Samuel Roberts: If you could take a look at that, I’d appreciate it. We’ll try to figure out… what’s best?
258 00:20:16.260 ⇒ 00:20:18.679 Samuel Roberts: Mustafa, have you worked with SVGs before at all?
259 00:20:19.920 ⇒ 00:20:25.539 Mustafa Raja: For text extraction, I have not, so this will be new for me, too.
260 00:20:25.670 ⇒ 00:20:33.270 Samuel Roberts: Cool, cool. Yeah, I’ve done some stuff with SVGs, and playing around and generating logos and stuff, but text, I think, is gonna be…
261 00:20:33.920 ⇒ 00:20:37.970 Samuel Roberts: pretty good, so… cool. I think…
262 00:20:38.670 ⇒ 00:20:44.009 Samuel Roberts: That’s the only, like, ask I have right now besides, you know, whatever feedback, good or bad.
263 00:20:44.010 ⇒ 00:20:44.720 Micah Gallardo: Yeah, yeah.
264 00:20:44.780 ⇒ 00:20:51.680 Samuel Roberts: want to hear more, but, but you, yeah, you gave some feedback earlier, so I don’t know if there’s much more, this week besides that, but…
265 00:20:51.890 ⇒ 00:20:59.260 Micah Gallardo: Yeah, yeah, yeah. Other than what I put up in the thread, and what we’re talking about, about getting these decks sorted, that’s pretty much it.
266 00:20:59.320 ⇒ 00:21:00.520 Samuel Roberts: Okay, cool.
267 00:21:00.970 ⇒ 00:21:02.709 Micah Gallardo: Yeah, I think the…
268 00:21:03.320 ⇒ 00:21:15.810 Micah Gallardo: Matt’s question is good, like, meta-awareness for, like, everybody else, but as far as, like, my role, like, on the content, like, side, this is getting very, very close, so…
269 00:21:15.810 ⇒ 00:21:16.290 Samuel Roberts: Okay, good.
270 00:21:16.290 ⇒ 00:21:35.960 Micah Gallardo: I think just the… I think the… the iterative process that we want to build in is the most important thing, and then, like I said, like, I can run the rubric manually, it’s really not that big of a deal. I’m saving plenty of time getting… having this thing format and get generated in the first place, so it’s really.
271 00:21:35.960 ⇒ 00:21:36.420 Samuel Roberts: Yeah.
272 00:21:36.420 ⇒ 00:21:39.850 Micah Gallardo: big of a deal to get a rubric going.
273 00:21:39.850 ⇒ 00:21:41.209 Samuel Roberts: Okay. And then… Yeah, that’s…
274 00:21:41.210 ⇒ 00:22:00.609 Micah Gallardo: Yeah, I think the rest of it is, like, yeah, getting some more of the iterative stuff sorted, and getting you the good material, so I’ll focus on that, getting you some of those SVGs, and then we’ll keep using it this coming week to see, like, what we… what we can deliver. We have some other stuff in the pipeline that we can probably…
275 00:22:00.770 ⇒ 00:22:04.499 Micah Gallardo: Start to see, like, where this begins to really, really shine.
276 00:22:04.640 ⇒ 00:22:13.619 Micah Gallardo: But, just a quick note, Mustafa, you were talking about the… the text, so when I…
277 00:22:13.780 ⇒ 00:22:20.449 Micah Gallardo: when I… add it in Slack? Should I just upload a PDF? Is that what’s gonna be best, or…
278 00:22:20.450 ⇒ 00:22:22.389 Samuel Roberts: I think it’s good either way now, I think he fixed it.
279 00:22:22.390 ⇒ 00:22:22.870 Mustafa Raja: Yeah, yeah.
280 00:22:22.870 ⇒ 00:22:23.440 Samuel Roberts: Got it.
281 00:22:23.440 ⇒ 00:22:24.180 Mustafa Raja: Yeah, I mean…
282 00:22:24.180 ⇒ 00:22:25.979 Samuel Roberts: I think… go ahead, Ms. Apafari.
283 00:22:25.980 ⇒ 00:22:32.080 Mustafa Raja: We want you to use it the way you would prefer to use it. And we can cater it to that.
284 00:22:32.450 ⇒ 00:22:34.810 Micah Gallardo: Okay, got it. Yeah, I think…
285 00:22:34.810 ⇒ 00:22:38.950 Samuel Roberts: stuff I had, like, text files of the transcription questionnaires and stuff.
286 00:22:39.150 ⇒ 00:22:44.350 Samuel Roberts: So it was just… we were uploading them as testing, and we hadn’t tested it that way, but it should…
287 00:22:44.350 ⇒ 00:22:47.250 Mustafa Raja: It should be able to handle both. This is exactly what happened.
288 00:22:47.910 ⇒ 00:22:50.699 Samuel Roberts: Whatever, whatever’s easiest for you, whatever’s, like…
289 00:22:50.830 ⇒ 00:22:56.530 Samuel Roberts: you know, doesn’t mess up, like, you don’t… I don’t want you to, like, prepare files a certain way or anything like that.
290 00:22:56.530 ⇒ 00:22:56.970 Micah Gallardo: Sure.
291 00:22:56.970 ⇒ 00:23:01.619 Samuel Roberts: I want you to be able to take what you have that, like, is part of your normal process and dump it in.
292 00:23:01.830 ⇒ 00:23:02.489 Micah Gallardo: So… Okay.
293 00:23:02.490 ⇒ 00:23:06.640 Samuel Roberts: If something else comes up, if another file format is good, we had that earlier, I think the…
294 00:23:06.870 ⇒ 00:23:14.479 Samuel Roberts: the rich text format, so that should be handled now, and the PDFs weren’t handled initially, like, it was all text to start. So, like, that’s the stuff we need to keep learning, but if anything else.
295 00:23:14.770 ⇒ 00:23:23.130 Samuel Roberts: you know, if you try something and it doesn’t work because it’s the wrong file format, let us know. If it’s just pasting text in, that’s great too. Okay.
296 00:23:23.360 ⇒ 00:23:26.370 Samuel Roberts: I think, yeah, again, we want to try to make this as, like.
297 00:23:26.640 ⇒ 00:23:30.150 Samuel Roberts: seamless as possible, and not add more work for you.
298 00:23:30.150 ⇒ 00:23:30.660 Micah Gallardo: If it’s 6.
299 00:23:30.660 ⇒ 00:23:34.800 Samuel Roberts: Even if it’s saving you time, we don’t want to, like, make you have to do work to save that time, you know?
300 00:23:34.800 ⇒ 00:23:39.619 Micah Gallardo: Right. Right, okay, that makes sense. That makes sense.
301 00:23:40.010 ⇒ 00:23:43.179 Micah Gallardo: Okay, cool, yeah. Other than that…
302 00:23:43.910 ⇒ 00:23:55.309 Micah Gallardo: Feeling pretty good about the whole thing. Again, the formatting looks great. I love, like, how all the slides are laid out in Notion. It’s really easy and simple to edit. Cool. I think…
303 00:23:59.090 ⇒ 00:23:59.780 Micah Gallardo: Excuse me.
304 00:23:59.880 ⇒ 00:24:09.200 Micah Gallardo: Yeah, no, this is… yeah, I think it’s… I think it’s fair, and I think we’ll work on getting you those SVGs, and then start to see what we can…
305 00:24:09.400 ⇒ 00:24:11.109 Micah Gallardo: Stress test it with, like, this.
306 00:24:11.110 ⇒ 00:24:11.660 Samuel Roberts: Yeah.
307 00:24:12.120 ⇒ 00:24:12.680 Samuel Roberts: Okay, cool.
308 00:24:12.680 ⇒ 00:24:15.840 Micah Gallardo: that are needing decks and stuff like that. We’ll start to see.
309 00:24:15.840 ⇒ 00:24:21.119 Samuel Roberts: I’d also say, like, not to, like, keep a log of the changes you’re making in Notion, but, like.
310 00:24:21.120 ⇒ 00:24:21.710 Micah Gallardo: Sure, sure, sure.
311 00:24:21.710 ⇒ 00:24:26.089 Samuel Roberts: If there are, like, major things that you’re, like, updating, definitely let us know that.
312 00:24:26.090 ⇒ 00:24:26.460 Micah Gallardo: Got it.
313 00:24:26.460 ⇒ 00:24:30.340 Samuel Roberts: Because obviously, like, it’s a V1, but, like, we want it to be as good a V1 as possible, so…
314 00:24:30.340 ⇒ 00:24:31.040 Micah Gallardo: Right, right, right.
315 00:24:31.040 ⇒ 00:24:40.080 Samuel Roberts: There’s something where you’re, like, constantly changing the same type of thing, or just, like, a big thing that it totally missed, if it’s a source or a hallucination or something, like…
316 00:24:40.080 ⇒ 00:24:40.860 Micah Gallardo: Sure, sure, sure.
317 00:24:40.860 ⇒ 00:24:51.830 Samuel Roberts: Just, like, ping us with those, just to, like, let us know. And as they… if they pile up in a certain way, then we’ll know, like, okay, it’s consistently doing the same thing, or we can tweak the prompt to make sure it doesn’t do one thing, kind of thing.
318 00:24:51.830 ⇒ 00:24:52.320 Micah Gallardo: Sure, sure, sure.
319 00:24:52.320 ⇒ 00:24:55.339 Samuel Roberts: So definitely, yeah, keep us in the loop with that.
320 00:24:55.650 ⇒ 00:24:56.549 Samuel Roberts: I think…
321 00:24:57.040 ⇒ 00:25:05.979 Samuel Roberts: like, we can take a look at your back and forths with it, too, because you can iterate in Slack, but if you’re doing things on Notion, we won’t see that, so… Got it.
322 00:25:05.980 ⇒ 00:25:06.490 Micah Gallardo: Okay.
323 00:25:06.490 ⇒ 00:25:13.869 Samuel Roberts: Again, it doesn’t need to be every little thing, because obviously, again, the human loop’s important here, but if it’s something that can be tweaked, and not
324 00:25:14.200 ⇒ 00:25:17.180 Samuel Roberts: You know, make that same change every time, definitely let us know.
325 00:25:17.730 ⇒ 00:25:19.529 Micah Gallardo: Okay, sounds good, no problem.
326 00:25:19.800 ⇒ 00:25:26.080 Samuel Roberts: Cool, alright, yeah, I think that’s… that’s, good on my end. Gustav, you got anything?
327 00:25:29.510 ⇒ 00:25:32.029 Micah Gallardo: Yeah, this is, this is all from YouTube.
328 00:25:32.370 ⇒ 00:25:47.339 Samuel Roberts: Okay, great. Cool. Yeah, if you’re good, I think we’re good. Yeah, all good. …taking the time today. Of course, absolutely. We have a call today, later, I see Tom got one on the calendar. But I think he’s also doing, like, a overall project check-in.
329 00:25:47.690 ⇒ 00:26:03.189 Samuel Roberts: Some days. Okay. Yeah, okay. So there’s another one who told Matthew. Oh, Justin, the new PM. Yeah. Oh, no, Sylvia said no. Okay, we gotta get… I think we’re getting in touch with her directly, though, now, because we want to start seeing what we can do visually for her, design-wise, but…
330 00:26:03.340 ⇒ 00:26:04.710 Micah Gallardo: Cool, cool, cool, cool, cool.
331 00:26:05.040 ⇒ 00:26:05.999 Samuel Roberts: Cool, so yeah.
332 00:26:06.000 ⇒ 00:26:06.320 Micah Gallardo: as.
333 00:26:06.320 ⇒ 00:26:16.489 Samuel Roberts: Yeah, that meeting’s a little more… I’m there sometimes, so I can give a little bit of, like, tech input, but it’s usually, like, Utum and Matthew just talking stuff through and making sure everything’s…
334 00:26:17.010 ⇒ 00:26:25.899 Samuel Roberts: going along, so anything comes out of that, and we need to make any changes or do anything differently, we’ll see, but… Okay. This is… this is the really helpful meeting for our, like.
335 00:26:25.900 ⇒ 00:26:26.299 Micah Gallardo: Yes, sir.
336 00:26:26.300 ⇒ 00:26:28.849 Samuel Roberts: So, I appreciate the time, and this is coming.
337 00:26:28.850 ⇒ 00:26:30.589 Micah Gallardo: Of course, absolutely, no problem.
338 00:26:30.590 ⇒ 00:26:30.920 Samuel Roberts: All right.
339 00:26:30.920 ⇒ 00:26:33.999 Micah Gallardo: Good to chat, we’ll, we’ll see you again next week.
340 00:26:34.240 ⇒ 00:26:34.660 Samuel Roberts: Sounds good.
341 00:26:34.660 ⇒ 00:26:35.170 Mustafa Raja: Thank you.
342 00:26:35.170 ⇒ 00:26:37.489 Samuel Roberts: Okay. Cheers, guys. Alright, bye.
343 00:26:37.720 ⇒ 00:26:38.460 Mustafa Raja: Mike.