Meeting Title: AI Interview Agent POC Planning Date: 2025-11-10 Meeting participants: Samuel Roberts, Mustafa Raja, Gabriel Lam
WEBVTT
1 00:00:47.530 ⇒ 00:00:48.440 Samuel Roberts: Hello.
2 00:00:51.410 ⇒ 00:00:52.419 Mustafa Raja: Hey, how are you?
3 00:00:53.120 ⇒ 00:00:55.779 Samuel Roberts: Doing alright. Sorry about that, I’m gonna step away for a minute.
4 00:00:56.520 ⇒ 00:00:57.800 Samuel Roberts: You’re back fast enough.
5 00:01:00.280 ⇒ 00:01:01.349 Samuel Roberts: How’s your day going?
6 00:01:02.210 ⇒ 00:01:10.670 Mustafa Raja: Yeah, they’re doing good. I did the, foreign key thing, and then… Oh, great!
7 00:01:10.990 ⇒ 00:01:17.419 Mustafa Raja: I also did the foreign key for the other table we had for… What’s your goal?
8 00:01:19.090 ⇒ 00:01:21.800 Mustafa Raja: The week, the weekly goals.
9 00:01:22.530 ⇒ 00:01:23.150 Samuel Roberts: Perfect.
10 00:01:23.730 ⇒ 00:01:27.240 Mustafa Raja: It also has the client ID, so I’m just going to do…
11 00:01:27.240 ⇒ 00:01:27.950 Samuel Roberts: Great.
12 00:01:28.720 ⇒ 00:01:30.150 Samuel Roberts: Yeah, that’s excellent.
13 00:01:31.960 ⇒ 00:01:33.920 Samuel Roberts: Okay, cool. Thank you for that.
14 00:01:34.350 ⇒ 00:01:35.500 Mustafa Raja: Huh.
15 00:01:37.820 ⇒ 00:01:40.459 Samuel Roberts: Yeah, should we hold on for Gabe, or…
16 00:01:40.460 ⇒ 00:01:47.409 Mustafa Raja: Yeah, we might be just looking for the link in…
17 00:01:47.410 ⇒ 00:01:48.650 Samuel Roberts: Yay.
18 00:01:49.500 ⇒ 00:01:54.349 Samuel Roberts: I’m just gonna… I’m gonna tag you both there so it doesn’t look like…
19 00:01:55.910 ⇒ 00:01:56.630 Mustafa Raja: Yay.
20 00:02:03.140 ⇒ 00:02:04.799 Samuel Roberts: I wanna share it, what am I doing?
21 00:02:10.440 ⇒ 00:02:12.030 Samuel Roberts: Cool.
22 00:02:14.830 ⇒ 00:02:23.790 Samuel Roberts: Yeah, so I did a little bit of diving into, like, the speech-to-speech stuff, and it looks like Mastra can do some of it, so… hopefully we can just work with those agents.
23 00:02:23.790 ⇒ 00:02:24.510 Mustafa Raja: Thanks.
24 00:02:25.070 ⇒ 00:02:25.860 Samuel Roberts: Yeah.
25 00:02:27.190 ⇒ 00:02:30.539 Mustafa Raja: I’m wondering what the interface would look like.
26 00:02:30.830 ⇒ 00:02:32.060 Samuel Roberts: Yeah, that’s… we gotta…
27 00:02:32.060 ⇒ 00:02:33.520 Mustafa Raja: Where you will live.
28 00:02:34.070 ⇒ 00:02:34.910 Samuel Roberts: Say that again?
29 00:02:35.800 ⇒ 00:02:42.450 Mustafa Raja: I’m wondering where is this integration going to live, in terms of, is it going to be a meeting?
30 00:02:43.540 ⇒ 00:02:45.570 Mustafa Raja: What it is. What is it going to be?
31 00:02:45.920 ⇒ 00:02:47.020 Samuel Roberts: Oh, yeah.
32 00:02:48.530 ⇒ 00:02:51.740 Samuel Roberts: I think we can think about that a little bit, and I… I…
33 00:02:51.960 ⇒ 00:02:53.849 Samuel Roberts: I think it could just be an interaction.
34 00:02:54.240 ⇒ 00:02:54.980 Samuel Roberts: Go ahead.
35 00:02:55.570 ⇒ 00:02:59.439 Mustafa Raja: Yeah, I think Zoom meeting would be cool to do, if we can.
36 00:02:59.710 ⇒ 00:03:04.999 Samuel Roberts: I thought about that, because it’s going to record and transcribe everything, but I actually like the idea of it…
37 00:03:05.150 ⇒ 00:03:08.319 Samuel Roberts: Being an agent that you chat with that’s doing the interview.
38 00:03:09.360 ⇒ 00:03:12.090 Mustafa Raja: Chat as in… as in talking, right?
39 00:03:12.270 ⇒ 00:03:15.170 Samuel Roberts: Yeah, and it looks like it can do it.
40 00:03:15.380 ⇒ 00:03:20.509 Samuel Roberts: I just… we gotta maybe do a little proof of concept of that part of it as we build out the rest of it, maybe.
41 00:03:21.850 ⇒ 00:03:27.270 Mustafa Raja: I’d be interested in… interested in… Happy OC.
42 00:03:27.790 ⇒ 00:03:35.439 Samuel Roberts: Yeah. Yeah, yeah, we’ll have to dig in and see what… because I think there’s a few things we can get done this week.
43 00:03:35.690 ⇒ 00:03:41.770 Samuel Roberts: But I… I don’t wanna… overextend the scope here, but I think what we can do,
44 00:03:42.090 ⇒ 00:03:45.450 Samuel Roberts: Just chatting a little bit with cursor… where is my cursor?
45 00:03:46.240 ⇒ 00:03:50.099 Samuel Roberts: But I think… I’m gonna give you the PRD app, actually, that would be more useful right now.
46 00:03:53.430 ⇒ 00:03:57.000 Mustafa Raja: We also have the bug thing, bug report thing.
47 00:03:58.120 ⇒ 00:04:03.789 Samuel Roberts: Yes, I looked at that, I looked at that this morning, I didn’t see any comment, but yeah, I think we could probably just throw one of those in there, the, what was it, the Sentry one?
48 00:04:04.400 ⇒ 00:04:10.440 Mustafa Raja: Yeah, yeah, I feel comfortable with Centrink. They have better documentation.
49 00:04:10.700 ⇒ 00:04:11.320 Samuel Roberts: Okay.
50 00:04:11.820 ⇒ 00:04:19.130 Samuel Roberts: Yeah, I think my only… I gotta dig in a little bit more. I just kind of glanced at it, but I want to make sure that we’re not going too…
51 00:04:19.459 ⇒ 00:04:19.980 Samuel Roberts: Out there.
52 00:04:19.980 ⇒ 00:04:24.080 Mustafa Raja: Yeah, I actually… yeah, I’d actually love that, too.
53 00:04:24.410 ⇒ 00:04:25.050 Samuel Roberts: Yeah.
54 00:04:25.460 ⇒ 00:04:27.630 Mustafa Raja: Yeah, the certified is always good.
55 00:04:27.870 ⇒ 00:04:31.279 Samuel Roberts: I’ll definitely take a look at that later in more depth, but…
56 00:04:34.630 ⇒ 00:04:35.700 Samuel Roberts: Yeah.
57 00:04:35.840 ⇒ 00:04:39.060 Samuel Roberts: Let me… where do I have it? I don’t have it open, maybe?
58 00:04:41.970 ⇒ 00:04:43.270 Samuel Roberts: Maybe I need to open.
59 00:04:46.770 ⇒ 00:04:54.219 Samuel Roberts: Oh, I have it in… or sorry, yeah, okay. So I think what we need to focus on initially is… where’s the PRD? Here it is.
60 00:04:55.550 ⇒ 00:05:02.000 Samuel Roberts: We want to put together, basically, an interview agent, is my thought.
61 00:05:02.710 ⇒ 00:05:03.390 Mustafa Raja: Yeah.
62 00:05:05.150 ⇒ 00:05:11.569 Samuel Roberts: And then… but with that, we should create the… Kind of case study table.
63 00:05:13.560 ⇒ 00:05:17.710 Samuel Roberts: that… We’ll take that interview…
64 00:05:21.560 ⇒ 00:05:26.329 Samuel Roberts: as one thing that stores there, but I don’t know exactly what that data structure looks like yet.
65 00:05:34.640 ⇒ 00:05:39.140 Samuel Roberts: Yeah… Sorry, I’m just re-scrolling through the PRD right now.
66 00:05:43.860 ⇒ 00:05:48.950 Samuel Roberts: Yeah, I think what we’re gonna wanna see is…
67 00:06:01.570 ⇒ 00:06:04.029 Samuel Roberts: What did he have for the, kind of, weekly…
68 00:06:08.090 ⇒ 00:06:13.120 Samuel Roberts: Yeah, so are you… do you have the PRD up? Can you look at it? Or you want me to share a screen, maybe?
69 00:06:14.920 ⇒ 00:06:20.980 Mustafa Raja: I can, yeah, I can share. Yeah, that’s…
70 00:06:20.980 ⇒ 00:06:25.750 Samuel Roberts: Yeah, it should be in the main branch and cursor now, too. Oh, no, it’s not, actually, it’s not merged in, but there should be a branch.
71 00:06:25.750 ⇒ 00:06:33.139 Mustafa Raja: Yeah. Utam is asking if we have, EMV for Insomnia, or do you have it? I don’t have it.
72 00:06:35.370 ⇒ 00:06:37.570 Mustafa Raja: It’s on Slack.
73 00:06:37.570 ⇒ 00:06:40.370 Samuel Roberts: Yeah, I don’t… what is the Insomnia repo?
74 00:06:41.190 ⇒ 00:06:43.549 Mustafa Raja: I think they shifted the code.
75 00:06:43.670 ⇒ 00:06:44.889 Mustafa Raja: There or something.
76 00:06:45.310 ⇒ 00:06:57.220 Samuel Roberts: Oh, oh, no, is that running somewhere, besides locally? Like, can we pull it from anywhere?
77 00:06:58.170 ⇒ 00:07:04.420 Mustafa Raja: I don’t have much context either on this.
78 00:07:04.600 ⇒ 00:07:06.470 Samuel Roberts: Yeah, let me, let me check something real quick.
79 00:07:06.470 ⇒ 00:07:08.760 Gabriel Lam: Hey guys, sorry about that, mock.
80 00:07:08.760 ⇒ 00:07:09.300 Samuel Roberts: Oh, you’re good.
81 00:07:09.300 ⇒ 00:07:12.789 Gabriel Lam: Just decided to conk itself out.
82 00:07:14.350 ⇒ 00:07:18.039 Samuel Roberts: My Slack was being real weird earlier, too. Oh, man, okay.
83 00:07:19.330 ⇒ 00:07:20.100 Samuel Roberts: Yeah.
84 00:07:20.100 ⇒ 00:07:21.730 Gabriel Lam: I’m on my bus, though.
85 00:07:21.730 ⇒ 00:07:26.589 Samuel Roberts: Okay, okay, no worries. We were just taking a look… Utom was asked about something for the…
86 00:07:26.810 ⇒ 00:07:28.620 Gabriel Lam: another repo, but I don’t…
87 00:07:32.710 ⇒ 00:07:34.750 Samuel Roberts: I don’t even… yeah, I don’t have it at all.
88 00:07:39.730 ⇒ 00:07:42.709 Samuel Roberts: I’m trying to think what that even is, like, what are the…
89 00:07:43.160 ⇒ 00:07:48.429 Mustafa Raja: If it were on… what’s it called here, who we could have pulled it on, but I haven’t seen the…
90 00:07:48.430 ⇒ 00:07:53.450 Samuel Roberts: Okay, yeah, if it’s not in Heroku, then I think it’s probably just a… Okay.
91 00:07:53.720 ⇒ 00:07:54.440 Samuel Roberts: Okay.
92 00:07:57.020 ⇒ 00:08:06.159 Samuel Roberts: Let’s jump back to what we’re doing then. Okay, so yeah, we were just talking through, a little bit of the PRD and just some, like… I think what we can do…
93 00:08:07.620 ⇒ 00:08:11.620 Samuel Roberts: I was looking at it here…
94 00:08:13.310 ⇒ 00:08:21.020 Samuel Roberts: I think there’s a couple things. Obviously, we need to do… well, that’s not obvious, hold on. The interviewing side of it, I think, is a big…
95 00:08:22.180 ⇒ 00:08:24.570 Samuel Roberts: a big thing. We kind of talked about that.
96 00:08:24.680 ⇒ 00:08:33.819 Samuel Roberts: I don’t necessarily know exactly the best way to do it. My thought was a, like, like it was in here, a kind of chat thing with an agent,
97 00:08:34.480 ⇒ 00:08:37.400 Samuel Roberts: I don’t know if we need more than that.
98 00:08:37.669 ⇒ 00:08:39.779 Samuel Roberts: Like, I suppose we could do something in Zoom.
99 00:08:43.320 ⇒ 00:08:45.580 Mustafa Raja: Yeah, I think this is…
100 00:08:46.370 ⇒ 00:08:48.960 Samuel Roberts: I don’t know if we need video, yeah.
101 00:08:49.630 ⇒ 00:08:54.320 Mustafa Raja: It just needs, sort of questions and… And that should be…
102 00:08:54.930 ⇒ 00:08:55.440 Samuel Roberts: Right.
103 00:08:55.440 ⇒ 00:08:55.890 Mustafa Raja: Cool.
104 00:08:55.890 ⇒ 00:08:56.999 Samuel Roberts: So I think… If it has…
105 00:08:59.110 ⇒ 00:09:04.480 Samuel Roberts: I did a little bit of research into the, like, speech-to-speech stuff,
106 00:09:05.890 ⇒ 00:09:10.739 Samuel Roberts: And Mastra, which is the, like, framework that we’re using, Can handle it.
107 00:09:11.830 ⇒ 00:09:16.100 Samuel Roberts: So I think… I think that’s gonna be the way to go about working towards it.
108 00:09:16.460 ⇒ 00:09:19.480 Samuel Roberts: The question becomes a little more…
109 00:09:20.210 ⇒ 00:09:25.490 Gabriel Lam: what needs to be stored in this case study object that we’re gonna create, and then what does the UI kind of look like?
110 00:09:27.600 ⇒ 00:09:31.330 Samuel Roberts: I think those are kind of the big, outstanding questions that I have.
111 00:09:32.480 ⇒ 00:09:35.480 Samuel Roberts: Oh, do we lose Gabe? Oh, nope, alright, I don’t know, it’s… yeah.
112 00:09:35.480 ⇒ 00:09:36.440 Gabriel Lam: We’re good, we’re directing.
113 00:09:36.440 ⇒ 00:09:40.309 Samuel Roberts: Okay, cool, cool. You’re good, I just saw it flicker, and I didn’t know what was, what was where.
114 00:09:44.720 ⇒ 00:09:47.270 Samuel Roberts: Yeah, so I think…
115 00:09:50.590 ⇒ 00:09:56.450 Samuel Roberts: I think the things I want to understand best is maybe we need to do, like, a think-through of
116 00:09:56.760 ⇒ 00:09:58.470 Samuel Roberts: what is important?
117 00:10:00.180 ⇒ 00:10:07.019 Samuel Roberts: for these case studies to be stored. I chatted with Cursor a little bit, and they put together a little bit of a plan that I can share.
118 00:10:07.140 ⇒ 00:10:07.790 Samuel Roberts: But it’s a little.
119 00:10:07.790 ⇒ 00:10:08.430 Mustafa Raja: Yes.
120 00:10:08.820 ⇒ 00:10:17.729 Samuel Roberts: It’s not… not quite what I want, and I just kind of fed up the PRD and said, like, help me plan this, and I think it’s going a little too deep on some things, but,
121 00:10:17.980 ⇒ 00:10:20.870 Samuel Roberts: Let me show you what I’ve got here.
122 00:10:24.560 ⇒ 00:10:25.300 Samuel Roberts: Alright.
123 00:10:26.040 ⇒ 00:10:33.689 Samuel Roberts: So, it’s… I had pointed it towards Monster because it started out with some other stuff.
124 00:10:34.140 ⇒ 00:10:39.819 Samuel Roberts: But the question it also had was whether or not we wanted things server-side or client-side.
125 00:10:43.860 ⇒ 00:10:55.040 Samuel Roberts: I think with Monster… then I pointed it towards Monster, and it said, okay, we can do it server-side, because the question was whether or not to just go a straight WebSocket to OpenAI, but I think we want to be able to manage it through the platform better, so…
126 00:10:57.940 ⇒ 00:11:04.279 Samuel Roberts: where it got a little in the weeds with some of this, like, I don’t know if we need all these API routes to figure this all out initially, but…
127 00:11:04.510 ⇒ 00:11:12.089 Samuel Roberts: If we’re gonna think of it in kind of two pieces, one is, like, managing…
128 00:11:12.200 ⇒ 00:11:17.950 Samuel Roberts: Other things for the assets and stuff, and one thing is… Doing the interview.
129 00:11:22.860 ⇒ 00:11:29.310 Samuel Roberts: I think that’s kind of… basically what I’m trying to get at is that I don’t have a ton of certainty around the actual interview piece yet.
130 00:11:31.040 ⇒ 00:11:36.570 Samuel Roberts: And so I’m thinking we should take a swing at, like, a… V0, V1 of that.
131 00:11:38.140 ⇒ 00:11:39.520 Gabriel Lam: I’m just wondering.
132 00:11:39.990 ⇒ 00:11:40.610 Samuel Roberts: Yeah.
133 00:11:41.080 ⇒ 00:11:45.489 Mustafa Raja: I’m just wondering how are we going to trigger… trigger this?
134 00:11:45.930 ⇒ 00:11:49.119 Mustafa Raja: Meeting, or whatever we’re going to call it.
135 00:11:51.720 ⇒ 00:11:52.540 Samuel Roberts: Yeah, okay, maybe…
136 00:11:52.540 ⇒ 00:11:53.470 Mustafa Raja: marketing.
137 00:11:54.070 ⇒ 00:11:54.740 Samuel Roberts: Go ahead.
138 00:11:55.010 ⇒ 00:12:01.460 Mustafa Raja: I guess marketing should have an interface to, tag, tag a…
139 00:12:01.660 ⇒ 00:12:13.009 Mustafa Raja: member of the platform. Yes. And then give some, context, and then whenever the member is available, they can just take the interview.
140 00:12:13.560 ⇒ 00:12:20.869 Samuel Roberts: Yeah, that’s kind of what I… yes, I think that’s a good… you’re right, I think what I should probably do is back up out of this a little bit, and maybe we should fig jam a little bit.
141 00:12:20.870 ⇒ 00:12:22.190 Mustafa Raja: Oh yeah, I love that.
142 00:12:22.190 ⇒ 00:12:39.530 Samuel Roberts: I was just… my brain was like, I think this interview thing is the big question mark here, and the rest of it is kind of like, yeah, like, even it points out here, like, this is CRUD operations, so we can create a simple thing to do that, but I think you’re right, the UI is probably where we want to go. So let’s fig jam a little bit first, let’s do that.
143 00:12:39.530 ⇒ 00:12:40.180 Mustafa Raja: Yes.
144 00:12:41.480 ⇒ 00:12:46.350 Samuel Roberts: Let me get Figma open and find… we have our AI…
145 00:12:48.390 ⇒ 00:12:51.689 Samuel Roberts: Ai team whiteboard, we’ll go there.
146 00:12:52.760 ⇒ 00:12:57.139 Samuel Roberts: I’m gonna make a new page for case study, I guess?
147 00:12:59.320 ⇒ 00:13:01.729 Samuel Roberts: Is this even the right one? Yeah, whatever, we would do it here.
148 00:13:04.700 ⇒ 00:13:05.960 Samuel Roberts: Okay.
149 00:13:06.530 ⇒ 00:13:09.229 Samuel Roberts: Okay, cool, Gabe, you’re already here. Excellent.
150 00:13:09.630 ⇒ 00:13:11.190 Samuel Roberts: Good.
151 00:13:13.810 ⇒ 00:13:15.950 Gabriel Lam: Share that there so we can all hop in. Okay.
152 00:13:16.030 ⇒ 00:13:25.800 Samuel Roberts: So I think the first thing, I normally… Where are the… diagram.
153 00:13:28.320 ⇒ 00:13:30.770 Samuel Roberts: Diagram basics.
154 00:13:31.520 ⇒ 00:13:36.760 Samuel Roberts: So I dropped this guy in, and then we can… sometimes… what we did last time was, like, copy and paste things, right? So…
155 00:13:36.760 ⇒ 00:13:37.260 Gabriel Lam: Yep.
156 00:13:37.260 ⇒ 00:13:39.690 Samuel Roberts: Let’s think about this. So, yeah.
157 00:13:40.150 ⇒ 00:13:50.330 Samuel Roberts: couple… Couple things to start with, so we’re gonna have a… case study…
158 00:13:51.260 ⇒ 00:13:56.149 Samuel Roberts: object. So, basically, something will get created, like a…
159 00:13:56.950 ⇒ 00:14:00.899 Samuel Roberts: A case study object kind of gets created, right?
160 00:14:04.670 ⇒ 00:14:13.449 Samuel Roberts: I’m just gonna talk a little bit before I start actually doing things, so that I need to get it out of my… out of my brain a little bit. So, case study gets created, that’s either through…
161 00:14:13.700 ⇒ 00:14:18.579 Samuel Roberts: Something automated, or just someone clicks, like, new case study, they pick the client, they pick…
162 00:14:18.740 ⇒ 00:14:21.420 Samuel Roberts: You know, they give it a description, maybe.
163 00:14:21.650 ⇒ 00:14:27.490 Samuel Roberts: What other information do we need to start that, I guess, is kind of my first thing.
164 00:14:29.290 ⇒ 00:14:38.619 Samuel Roberts: Or is that enough for now? And then they gotta go through this process of, like, setting up the interview with the right people, the right people get triggered to do an interview.
165 00:14:40.970 ⇒ 00:14:46.709 Samuel Roberts: The interview gets processed along with whatever other kind of assets there might be.
166 00:14:47.920 ⇒ 00:14:49.320 Samuel Roberts: And then…
167 00:14:54.030 ⇒ 00:14:56.809 Samuel Roberts: We get an output that is, like, a rough…
168 00:14:57.990 ⇒ 00:15:00.000 Samuel Roberts: version, I guess? I don’t know.
169 00:15:00.200 ⇒ 00:15:04.470 Samuel Roberts: what are we looking at here? Okay, yeah, flow, creates new case study, contact members to interview, yeah.
170 00:15:05.640 ⇒ 00:15:11.180 Mustafa Raja: Yeah, this… this should then notify the members whenever they want to be able to pass.
171 00:15:11.180 ⇒ 00:15:11.530 Gabriel Lam: Yep.
172 00:15:11.530 ⇒ 00:15:12.839 Mustafa Raja: Right, right.
173 00:15:13.010 ⇒ 00:15:14.149 Samuel Roberts: Totally, totally.
174 00:15:15.000 ⇒ 00:15:15.910 Samuel Roberts: Okay.
175 00:15:17.790 ⇒ 00:15:19.760 Mustafa Raja: This sort of initiates it, right?
176 00:15:20.390 ⇒ 00:15:26.479 Samuel Roberts: Yeah, yeah, so… We will have kind of this… alright, so, like.
177 00:15:26.790 ⇒ 00:15:28.270 Mustafa Raja: So let’s do this, okay.
178 00:15:29.350 ⇒ 00:15:30.130 Mustafa Raja: Yes.
179 00:15:32.180 ⇒ 00:15:37.560 Samuel Roberts: So, like, we’re gonna have a, new case study, right?
180 00:15:37.560 ⇒ 00:15:38.320 Mustafa Raja: Yes.
181 00:15:40.870 ⇒ 00:15:43.700 Samuel Roberts: That’s gonna do a few things, right?
182 00:15:44.460 ⇒ 00:15:46.920 Samuel Roberts: It’s gonna create a new case study object.
183 00:15:48.240 ⇒ 00:16:00.420 Samuel Roberts: Maybe this diagram isn’t the best way to… I don’t know. The problem I have right now is when I’m thinking of this stuff, there’s a few different, like… there’s the process of it, and the UI of it, and the data structure of it, and I want to kind of get a few of these down on…
184 00:16:01.060 ⇒ 00:16:01.820 Samuel Roberts: At least.
185 00:16:02.730 ⇒ 00:16:05.250 Samuel Roberts: losing my head in the weeds of this a little bit, I think.
186 00:16:05.250 ⇒ 00:16:08.160 Mustafa Raja: I’m thinking, let’s talk about flow.
187 00:16:08.380 ⇒ 00:16:09.380 Mustafa Raja: Yeah, I agree.
188 00:16:09.380 ⇒ 00:16:09.730 Samuel Roberts: Okay.
189 00:16:09.730 ⇒ 00:16:13.350 Mustafa Raja: And then, we can then, for each flow, lay out the UI thing.
190 00:16:13.350 ⇒ 00:16:13.850 Gabriel Lam: Yeah.
191 00:16:13.850 ⇒ 00:16:18.059 Samuel Roberts: Okay, alright, so the case study’s gonna get created. What is…
192 00:16:19.360 ⇒ 00:16:22.370 Samuel Roberts: what flows from that? We need something that’s gonna…
193 00:16:23.280 ⇒ 00:16:25.180 Mustafa Raja: Notify the user, or…
194 00:16:25.180 ⇒ 00:16:25.920 Gabriel Lam: Yep.
195 00:16:25.920 ⇒ 00:16:30.109 Mustafa Raja: A place for users to initiate the interview.
196 00:16:33.480 ⇒ 00:16:40.460 Samuel Roberts: Right. So, okay, I guess that’s gonna be… alright, so then we’re gonna have… Yeah, it’s gonna go to…
197 00:16:44.230 ⇒ 00:16:45.220 Samuel Roberts: Process, bro.
198 00:16:45.220 ⇒ 00:16:47.520 Gabriel Lam: I don’t think it really matters at this point, like, what…
199 00:16:47.520 ⇒ 00:16:53.070 Samuel Roberts: No, no, no, let’s just rent it out, let’s just write it out. Yeah, then we can do, this is it, yeah.
200 00:16:55.680 ⇒ 00:16:58.370 Samuel Roberts: Go ahead and drag that, yeah.
201 00:16:59.050 ⇒ 00:17:07.259 Samuel Roberts: Is it working? Okay. So, new case study gets created, as part of that, there’ll be a…
202 00:17:07.640 ⇒ 00:17:11.680 Samuel Roberts: selection of… who’s the interviewee, I guess?
203 00:17:13.280 ⇒ 00:17:16.250 Samuel Roberts: And then that person does the interview.
204 00:17:17.960 ⇒ 00:17:22.369 Samuel Roberts: Which is a big black box right now, but that is kind of the biggest box, I think.
205 00:17:23.040 ⇒ 00:17:23.650 Mustafa Raja: Yes.
206 00:17:23.650 ⇒ 00:17:28.559 Samuel Roberts: Gabe, when you talk to Hannah.
207 00:17:28.569 ⇒ 00:17:29.219 Gabriel Lam: Yes.
208 00:17:29.710 ⇒ 00:17:34.580 Samuel Roberts: the… Process of taking that… transcript.
209 00:17:34.950 ⇒ 00:17:37.279 Samuel Roberts: to an output. Is that…
210 00:17:38.670 ⇒ 00:17:40.619 Gabriel Lam: The transcript from the interview?
211 00:17:40.900 ⇒ 00:17:41.670 Samuel Roberts: Yes.
212 00:17:41.670 ⇒ 00:17:45.009 Gabriel Lam: Yeah, it basically just runs through two different.
213 00:17:45.010 ⇒ 00:17:45.979 Samuel Roberts: That’s what… okay.
214 00:17:45.980 ⇒ 00:17:49.770 Gabriel Lam: two different… Openai prompts.
215 00:17:50.220 ⇒ 00:17:55.469 Samuel Roberts: Got it, okay, that’s kind of what I was wondering. So there’s two different prompts… what did… do you know what those prompts were? Or, like, what their purpose were?
216 00:17:55.470 ⇒ 00:18:04.339 Gabriel Lam: The first one is to… Create, like, a… a copy?
217 00:18:04.480 ⇒ 00:18:07.500 Gabriel Lam: Like, a… a narrative, basically.
218 00:18:07.500 ⇒ 00:18:07.850 Samuel Roberts: Okay.
219 00:18:07.850 ⇒ 00:18:10.219 Gabriel Lam: Exact words that are gonna be used.
220 00:18:10.690 ⇒ 00:18:16.570 Gabriel Lam: And that’s split into, like, the different sections. So there’s, like, context, challenge, solution, result, impact.
221 00:18:16.950 ⇒ 00:18:23.379 Gabriel Lam: And then the next prompt is really to turn those words into something that aligns with the brand messaging.
222 00:18:23.970 ⇒ 00:18:29.270 Gabriel Lam: So, like, it’s sort of the first pass is, like, categorize it, and then the second pass is, like, polish.
223 00:18:29.270 ⇒ 00:18:31.009 Samuel Roberts: the word. Got it. Okay, cool.
224 00:18:31.150 ⇒ 00:18:39.879 Samuel Roberts: That’s a good fact. So that’s a process we have pretty well, understood, compared at least to the… well, the interview process also has questions and stuff, but I think…
225 00:18:39.880 ⇒ 00:18:40.350 Gabriel Lam: Right.
226 00:18:40.350 ⇒ 00:18:45.969 Samuel Roberts: how… how we go about making that work is the big question, but most of this is pretty well defined, then. Okay.
227 00:18:46.470 ⇒ 00:18:49.340 Samuel Roberts: Create the case study, add whatever context, yes.
228 00:18:55.050 ⇒ 00:18:57.950 Mustafa Raja: Yeah, I’m just laying out what kind of UI we would need.
229 00:18:58.440 ⇒ 00:18:59.300 Samuel Roberts: Yeah.
230 00:19:00.630 ⇒ 00:19:06.569 Samuel Roberts: Right, okay, yeah. So, I feel like my brain is just all over the place right now, and I apologize for that.
231 00:19:06.570 ⇒ 00:19:08.239 Mustafa Raja: Hmm, this is…
232 00:19:09.560 ⇒ 00:19:12.880 Samuel Roberts: So then the interview will happen… .
233 00:19:18.380 ⇒ 00:19:29.059 Mustafa Raja: Okay, now… Here comes the… here comes the interview object, right? Yes. For my straw agent.
234 00:19:29.830 ⇒ 00:19:33.959 Mustafa Raja: He’s the… Here’s the link to it if anyone wants to…
235 00:19:33.960 ⇒ 00:19:35.490 Samuel Roberts: Yeah, that’s what I was… yeah, yeah, yeah.
236 00:19:36.120 ⇒ 00:19:37.780 Samuel Roberts: So, then there’s a big…
237 00:19:38.130 ⇒ 00:19:44.980 Samuel Roberts: back and forth. Well, that’s the kind of box we’ll figure out, so let’s just say, like, Notify users…
238 00:19:44.980 ⇒ 00:19:45.490 Mustafa Raja: So…
239 00:19:45.750 ⇒ 00:19:48.210 Samuel Roberts: Sure, it doesn’t matter right now, let’s just get it down there, yeah.
240 00:19:48.350 ⇒ 00:19:49.310 Mustafa Raja: Yeah, yeah.
241 00:19:51.080 ⇒ 00:20:00.030 Samuel Roberts: Okay, so, like, interview… Interview, trend… script…
242 00:20:00.890 ⇒ 00:20:07.209 Samuel Roberts: comes out of this, right? Or the actual interview happens, right? And then the transcript is an output of that, so let’s just say…
243 00:20:08.560 ⇒ 00:20:12.839 Samuel Roberts: Interface to initiate, alright, good enough. Interview transcript, so then that’s gonna go into…
244 00:20:14.620 ⇒ 00:20:16.710 Samuel Roberts: Well, so, out of this, also…
245 00:20:16.910 ⇒ 00:20:19.490 Samuel Roberts: We’re gonna have outputs from this, too, right?
246 00:20:22.120 ⇒ 00:20:25.389 Mustafa Raja: There’s gonna be other outputs. Well, actually, let me ask that.
247 00:20:25.780 ⇒ 00:20:26.420 Samuel Roberts: Go ahead.
248 00:20:27.180 ⇒ 00:20:31.870 Mustafa Raja: Yeah, I think this, this would just, go towards this thing.
249 00:20:32.790 ⇒ 00:20:33.630 Mustafa Raja: You know?
250 00:20:34.200 ⇒ 00:20:36.370 Samuel Roberts: Okay, I guess my question is, are there other…
251 00:20:36.650 ⇒ 00:20:42.369 Samuel Roberts: assets that go into this generation step besides the interview? Or is the interview the big, like.
252 00:20:43.290 ⇒ 00:20:48.029 Mustafa Raja: Yeah, I, I remember, I remember…
253 00:20:49.140 ⇒ 00:20:52.599 Mustafa Raja: Hannah’s saying that we should… we can tag.
254 00:20:52.710 ⇒ 00:20:55.459 Mustafa Raja: What’s it called, meetings?
255 00:20:55.640 ⇒ 00:20:58.450 Mustafa Raja: If we have those regarding that particular.
256 00:20:58.450 ⇒ 00:21:01.999 Samuel Roberts: Okay, so there’d be other meetings to pass into this object?
257 00:21:02.590 ⇒ 00:21:04.479 Mustafa Raja: If we can see if there are…
258 00:21:05.690 ⇒ 00:21:10.400 Mustafa Raja: Even those for the meetings. So, essentially, we can say transcripts.
259 00:21:11.810 ⇒ 00:21:13.619 Mustafa Raja: Supporting transit folder.
260 00:21:16.210 ⇒ 00:21:16.970 Samuel Roberts: Okay.
261 00:21:22.980 ⇒ 00:21:23.960 Samuel Roberts: I guess…
262 00:21:27.550 ⇒ 00:21:32.899 Samuel Roberts: No, that makes sense. Okay, so then other things that come in here could be, referenced meetings, is what you’re saying?
263 00:21:36.790 ⇒ 00:21:38.100 Mustafa Raja: Yeah, the…
264 00:21:40.150 ⇒ 00:21:47.529 Mustafa Raja: The assigned users or the assigned members would be… we should be able to, add some reference and things.
265 00:21:48.660 ⇒ 00:21:52.470 Mustafa Raja: Okay, and then… when to do that, when to do creating that content.
266 00:21:53.280 ⇒ 00:21:57.300 Samuel Roberts: Oh, I see, so, like, okay, so… Wait, what was that? Sorry.
267 00:21:57.730 ⇒ 00:22:03.950 Gabriel Lam: So, like, a list of meetings, and then all the transcripts from that list of meetings goes into context and stories.
268 00:22:03.950 ⇒ 00:22:04.790 Mustafa Raja: little coffee.
269 00:22:04.790 ⇒ 00:22:07.189 Samuel Roberts: Okay. Are there other things that need to be?
270 00:22:07.530 ⇒ 00:22:10.459 Mustafa Raja: From there, we can prioritize the interview.
271 00:22:11.400 ⇒ 00:22:12.049 Mustafa Raja: And then…
272 00:22:12.050 ⇒ 00:22:12.440 Samuel Roberts: Yeah.
273 00:22:13.410 ⇒ 00:22:13.810 Samuel Roberts: Okay.
274 00:22:14.200 ⇒ 00:22:15.609 Gabriel Lam: Oh, I see, I see, yeah.
275 00:22:16.290 ⇒ 00:22:29.289 Samuel Roberts: Yeah, and I think this is where, like, a multi-agent kind of monster flow might be helpful, where it does the pass of the narrative, it does a pass of the brand, it scans the meetings and sees what relevant stuff can be pulled from there. Like, we could do a…
276 00:22:30.580 ⇒ 00:22:35.160 Samuel Roberts: a pretty… And this could be an interesting, complex kind of…
277 00:22:35.370 ⇒ 00:22:41.710 Samuel Roberts: non-deterministic kind of thing, and get the output. We’ll see, we’ll see what goes well for that.
278 00:22:43.010 ⇒ 00:22:43.840 Samuel Roberts: Okay.
279 00:22:46.090 ⇒ 00:22:49.279 Samuel Roberts: And then there’s usually some iteration on this, I assume, right?
280 00:22:49.280 ⇒ 00:22:50.030 Mustafa Raja: Yes.
281 00:22:51.870 ⇒ 00:22:53.619 Samuel Roberts: So is that something that…
282 00:22:53.620 ⇒ 00:22:58.740 Mustafa Raja: I’m wondering if this is going to be,
283 00:22:59.100 ⇒ 00:23:03.800 Mustafa Raja: what’s it called? A proper PDF, or just some text?
284 00:23:04.670 ⇒ 00:23:06.740 Samuel Roberts: I think it… it would be…
285 00:23:08.910 ⇒ 00:23:15.040 Samuel Roberts: Well, I suppose there’s a few different ways to go with it, you’re right. My thought was that it would be some text.
286 00:23:16.000 ⇒ 00:23:21.180 Mustafa Raja: Yeah, I guess for now, we can do text, and then… We can iterate on it.
287 00:23:21.180 ⇒ 00:23:25.080 Samuel Roberts: But I think… I think this… this is gonna be a big… a big thing to build.
288 00:23:25.220 ⇒ 00:23:32.719 Samuel Roberts: Yes, sir. Already, in addition to, like, the UI for managing the case study. So there’s… there’s two main UIs that we have here at this point, right?
289 00:23:33.220 ⇒ 00:23:33.720 Mustafa Raja: this…
290 00:23:33.720 ⇒ 00:23:35.570 Gabriel Lam: There’s an interview UI and a…
291 00:23:36.400 ⇒ 00:23:37.259 Samuel Roberts: Yes. To study that.
292 00:23:37.260 ⇒ 00:23:37.860 Gabriel Lam: Okay.
293 00:23:38.010 ⇒ 00:23:44.930 Samuel Roberts: Well, the case study UI already sort of exists, right, no? On the marketing assets page on the platform?
294 00:23:45.250 ⇒ 00:23:49.409 Samuel Roberts: So that is, like, the completed assets page. I see.
295 00:23:49.410 ⇒ 00:23:49.870 Mustafa Raja: Bye.
296 00:23:49.870 ⇒ 00:23:54.509 Samuel Roberts: That evolved recently from just being files stored on GitHub.
297 00:23:54.930 ⇒ 00:23:57.669 Samuel Roberts: it still is file stored in GitHub.
298 00:23:57.810 ⇒ 00:24:02.170 Samuel Roberts: a few weeks ago, Hannah asked me to make, like,
299 00:24:02.350 ⇒ 00:24:04.210 Samuel Roberts: She wanted to be able to, like, sort
300 00:24:04.530 ⇒ 00:24:07.970 Samuel Roberts: Add tags to it, and so we added some metadata.
301 00:24:07.970 ⇒ 00:24:09.739 Gabriel Lam: a metadata table.
302 00:24:10.210 ⇒ 00:24:12.839 Samuel Roberts: It’s still stored in GitHub right now.
303 00:24:13.410 ⇒ 00:24:19.779 Samuel Roberts: But we’re gonna eventually move that to, like, its own storage place. I think that’s gonna be separate from this.
304 00:24:19.980 ⇒ 00:24:23.960 Samuel Roberts: interface for now. I think of that, because there could be other assets there.
305 00:24:24.560 ⇒ 00:24:30.290 Samuel Roberts: I think this is gonna be the place to, like, manage the case study creation process?
306 00:24:30.760 ⇒ 00:24:32.289 Samuel Roberts: And then, for now…
307 00:24:32.290 ⇒ 00:24:33.230 Gabriel Lam: to the assets later.
308 00:24:33.230 ⇒ 00:24:39.380 Samuel Roberts: Exactly. Once there’s, like, a done PDF going to the assets, perhaps we link back to…
309 00:24:39.520 ⇒ 00:24:41.949 Samuel Roberts: This case study, if things need to change, but…
310 00:24:41.950 ⇒ 00:24:42.480 Gabriel Lam: Yeah.
311 00:24:42.670 ⇒ 00:24:49.600 Samuel Roberts: I still… I don’t think every… not everyone’s gonna go to this case study page, but people who don’t would still go to the marketing assets page for, like, sharing.
312 00:24:49.600 ⇒ 00:24:49.930 Gabriel Lam: Right.
313 00:24:49.930 ⇒ 00:24:51.230 Samuel Roberts: To leads and stuff.
314 00:24:52.290 ⇒ 00:24:55.480 Samuel Roberts: Okay.
315 00:24:57.690 ⇒ 00:24:58.870 Samuel Roberts: I think this is…
316 00:24:59.510 ⇒ 00:25:08.449 Samuel Roberts: I mean, it seems… it looks very simple, but I feel like there’s gonna be a lot here already that I don’t want to go beyond, like, the final completion step yet. Yeah. I think maybe…
317 00:25:08.870 ⇒ 00:25:11.250 Samuel Roberts: Maybe that would be, like, a next week thing.
318 00:25:11.250 ⇒ 00:25:13.029 Mustafa Raja: Yeah, I think this looks good.
319 00:25:13.620 ⇒ 00:25:15.249 Samuel Roberts: Okay, so then…
320 00:25:15.310 ⇒ 00:25:17.309 Mustafa Raja: Let’s break this down to a few…
321 00:25:17.310 ⇒ 00:25:19.420 Samuel Roberts: Things like tickets that we can do.
322 00:25:19.890 ⇒ 00:25:25.080 Samuel Roberts: What do I do with this? I just say, like.
323 00:25:25.080 ⇒ 00:25:28.309 Mustafa Raja: I’m thinking of splitting it in two parts, so a UI.
324 00:25:28.310 ⇒ 00:25:29.369 Samuel Roberts: Yes, exactly, yeah.
325 00:25:29.370 ⇒ 00:25:32.820 Mustafa Raja: And then, what a POC for this?
326 00:25:33.550 ⇒ 00:25:36.910 Samuel Roberts: Yeah, -oh, I just hit back. Hold on.
327 00:25:36.910 ⇒ 00:25:37.680 Mustafa Raja: Yeah, yeah.
328 00:25:38.560 ⇒ 00:25:45.800 Samuel Roberts: Where’d I go? Case study. Okay. Yeah, so I think, basically, we’re gonna need to build that CRUD interface for
329 00:25:47.020 ⇒ 00:26:00.750 Samuel Roberts: right, so there’s, like, over here we have the… CRUD UI for… Case study… Right.
330 00:26:06.080 ⇒ 00:26:07.499 Samuel Roberts: I don’t want that, what’d I do?
331 00:26:09.850 ⇒ 00:26:12.789 Samuel Roberts: Alright, so that’s, like, over here, and then there’ll be a big…
332 00:26:16.570 ⇒ 00:26:19.130 Samuel Roberts: And then this is the actual, like, interview UI.
333 00:26:19.430 ⇒ 00:26:20.499 Mustafa Raja: There you go.
334 00:26:22.670 ⇒ 00:26:25.949 Samuel Roberts: UI and AI agent.
335 00:26:27.870 ⇒ 00:26:28.700 Samuel Roberts: Right.
336 00:26:28.700 ⇒ 00:26:33.110 Gabriel Lam: And then there’s… then there’s… I mean, realistically, there’s this third piece, which is the actual, like.
337 00:26:33.200 ⇒ 00:26:34.350 Samuel Roberts: processing.
338 00:26:36.160 ⇒ 00:26:39.130 Samuel Roberts: Of, our, generation.
339 00:26:40.110 ⇒ 00:26:41.560 Samuel Roberts: Okay, that’s actually…
340 00:26:41.870 ⇒ 00:26:42.430 Gabriel Lam: Right.
341 00:26:42.660 ⇒ 00:26:45.090 Samuel Roberts: Three pretty good chunks that we can break down, you know?
342 00:26:45.780 ⇒ 00:26:46.500 Mustafa Raja: It’s…
343 00:26:47.220 ⇒ 00:26:54.440 Samuel Roberts: The first one… Honestly, like…
344 00:26:54.820 ⇒ 00:27:02.570 Samuel Roberts: is basically, like, an exercise in figuring out what elements need to be stored for case studies. So, you know…
345 00:27:02.750 ⇒ 00:27:05.889 Samuel Roberts: let’s just think about that real quick while we’re all here.
346 00:27:07.550 ⇒ 00:27:15.050 Samuel Roberts: We have… what? Title, like a, you know, rough title… title?
347 00:27:17.120 ⇒ 00:27:23.290 Samuel Roberts: title, client… Project, maybe you would call it?
348 00:27:23.720 ⇒ 00:27:26.120 Samuel Roberts: The title might come out of the generation step, but…
349 00:27:26.120 ⇒ 00:27:26.950 Mustafa Raja: Yes.
350 00:27:27.140 ⇒ 00:27:32.530 Samuel Roberts: So, there’ll be a client, project, Interviewee…
351 00:27:33.770 ⇒ 00:27:35.660 Samuel Roberts: That’s not how you spell any of that, Sam.
352 00:27:38.660 ⇒ 00:27:47.720 Samuel Roberts: interviewee, I guess? Interviewee will even say… could be plural at some point, because right now it’s just one person, but theoretically we could have multiple people.
353 00:27:48.960 ⇒ 00:27:52.169 Samuel Roberts: What else do we want there? Meetings?
354 00:27:52.790 ⇒ 00:27:54.159 Samuel Roberts: Store it on there.
355 00:27:55.250 ⇒ 00:28:01.440 Samuel Roberts: Other assets, perhaps? Like, images?
356 00:28:03.380 ⇒ 00:28:05.199 Samuel Roberts: Screenshots kind of thing.
357 00:28:06.260 ⇒ 00:28:10.699 Samuel Roberts: But eventually it’s gonna, you know, we want this to be, like, a one-stop shop for doing this, right?
358 00:28:12.270 ⇒ 00:28:15.829 Gabriel Lam: In this… wait, can I… can I step back for a little bit?
359 00:28:15.830 ⇒ 00:28:16.210 Samuel Roberts: Yeah.
360 00:28:16.210 ⇒ 00:28:23.749 Gabriel Lam: when… when you’re including meetings and images into this CRUD UI, are you suggesting that, like, this is gonna be a new
361 00:28:23.880 ⇒ 00:28:29.359 Gabriel Lam: Essentially, like, a platform for you to upload all your… Existing assets onto.
362 00:28:32.910 ⇒ 00:28:33.270 Samuel Roberts: Yeah.
363 00:28:33.270 ⇒ 00:28:41.549 Gabriel Lam: My question really is, like, Is this more of a… tool to… send out
364 00:28:42.610 ⇒ 00:28:46.659 Gabriel Lam: basically, like, a little dialogue to be like, hey, I have a new case study, I need to send this out.
365 00:28:47.040 ⇒ 00:28:49.449 Gabriel Lam: And notify, and then get the copy.
366 00:28:49.790 ⇒ 00:28:55.610 Gabriel Lam: To put into our existing flow, or are we… or it sounds more like we’re gonna totally revamp it.
367 00:28:55.610 ⇒ 00:28:58.769 Samuel Roberts: I think… I think I’m thinking of it as a revamp, and if we need to.
368 00:28:58.770 ⇒ 00:28:59.220 Gabriel Lam: Yeah.
369 00:28:59.220 ⇒ 00:29:08.039 Samuel Roberts: pare it down, I would go from there. My thought, from, like, a very high level, is, you know, just like the platform, you go to, like, an AI tools, and it says stand-up assistant.
370 00:29:08.790 ⇒ 00:29:25.120 Samuel Roberts: it’ll have, like, case studies, or case study assistant. And you go there, and there you can manage, you know, there might be a list of case studies. You click on one of those, it brings you to a page that lets you, like, input stuff. Kind of like the marketing assets page, like that list. Yeah.
371 00:29:25.420 ⇒ 00:29:28.319 Samuel Roberts: But this is for, like, the pre-completed…
372 00:29:29.770 ⇒ 00:29:36.449 Samuel Roberts: And so, this is not necessarily all things that will get uploaded over… like, but this will be stuff that has to get stored on this object in Superbase.
373 00:29:36.650 ⇒ 00:29:37.060 Gabriel Lam: Yep.
374 00:29:37.060 ⇒ 00:29:37.670 Mustafa Raja: Okay.
375 00:29:37.950 ⇒ 00:29:46.240 Samuel Roberts: Is my thought. So, like, as part of this interview process, they might say, oh, I’ll give you a screenshot, and part of that might be, okay, make sure to get a screenshot of that, or something, you know?
376 00:29:46.240 ⇒ 00:29:46.940 Gabriel Lam: Yep.
377 00:29:46.940 ⇒ 00:29:52.299 Samuel Roberts: Doesn’t necessarily mean that when you’re creating a case study, you’re gonna add all this stuff.
378 00:29:52.540 ⇒ 00:29:53.859 Samuel Roberts: But at some point.
379 00:29:54.000 ⇒ 00:29:59.729 Samuel Roberts: These things will be associated with that case study, which will all probably go into this output at some point.
380 00:30:00.910 ⇒ 00:30:02.500 Samuel Roberts: Is my… is my thought.
381 00:30:02.790 ⇒ 00:30:03.310 Mustafa Raja: Yes, yes.
382 00:30:03.310 ⇒ 00:30:17.589 Samuel Roberts: Maybe I’m overdoing it, and again, we can pare it down to just, like, a… if by the time we do, like, a first pass on the interview, we’re like, hey, this is gonna be a lot more work this week than we thought, I think we prioritize that part of it.
383 00:30:17.810 ⇒ 00:30:24.400 Samuel Roberts: But I think… from, like, the full PRD standpoint, this is kind of the full picture.
384 00:30:27.100 ⇒ 00:30:27.900 Mustafa Raja: Yes.
385 00:30:28.570 ⇒ 00:30:33.550 Samuel Roberts: But I’m just thinking, like, so database… like, maybe the DB… table…
386 00:30:34.260 ⇒ 00:30:39.999 Samuel Roberts: maybe that’s a better way to say it. And then, like, whatever the UI looks like is that. But for now, like, it’s just, like, that’s what’s the important…
387 00:30:40.550 ⇒ 00:30:46.550 Samuel Roberts: data that we’re gonna need for this. Are there other things that I’m missing that might go into this eventual final case study?
388 00:30:49.200 ⇒ 00:30:51.549 Samuel Roberts: I mean, eventually there’ll be a transcript, I guess, too.
389 00:30:53.190 ⇒ 00:30:54.889 Samuel Roberts: That’ll be stored on there, right?
390 00:30:54.890 ⇒ 00:30:55.630 Mustafa Raja: Yeah.
391 00:30:57.310 ⇒ 00:30:57.920 Samuel Roberts: Come on.
392 00:30:58.520 ⇒ 00:31:02.740 Samuel Roberts: Any other… any other things, that Anna might have mentioned, or anything like that?
393 00:31:04.440 ⇒ 00:31:06.040 Mustafa Raja: I think this looks good.
394 00:31:07.030 ⇒ 00:31:08.940 Gabriel Lam: I think it’s fine, I think there’s…
395 00:31:09.520 ⇒ 00:31:11.880 Gabriel Lam: For Hannah, it was really more about…
396 00:31:13.240 ⇒ 00:31:20.680 Gabriel Lam: the interviewing. I think the design part seems a little… Separate at the moment.
397 00:31:22.350 ⇒ 00:31:30.729 Samuel Roberts: Yes, no, I agree, I agree. This is just… I’m trying to think what references will need to be for the generation of the copy. Beyond that, the design stuff, I don’t necessarily want to touch yet.
398 00:31:30.730 ⇒ 00:31:32.910 Gabriel Lam: Okay, okay, got it. I see, makes sense.
399 00:31:32.910 ⇒ 00:31:36.920 Samuel Roberts: No, no, no, this is just, like, if, like, what would…
400 00:31:37.560 ⇒ 00:31:40.239 Samuel Roberts: So we’ll get fed in here. Yeah.
401 00:31:40.540 ⇒ 00:31:47.810 Samuel Roberts: That’s my thought. Yeah. I don’t wanna… I want this to… the output of this should be the output of her current agents.
402 00:31:47.810 ⇒ 00:31:48.620 Gabriel Lam: Yes.
403 00:31:49.050 ⇒ 00:31:52.280 Samuel Roberts: And so I just don’t know if there’s any other things that go into that that I’m missing here.
404 00:31:54.310 ⇒ 00:31:56.099 Mustafa Raja: But I think it’s fine.
405 00:31:56.290 ⇒ 00:31:59.769 Mustafa Raja: once we have the UI, we can… we can… Yeah.
406 00:31:59.770 ⇒ 00:32:00.710 Samuel Roberts: So then…
407 00:32:00.710 ⇒ 00:32:02.050 Mustafa Raja: Hannah’s opinion on it.
408 00:32:03.610 ⇒ 00:32:07.499 Gabriel Lam: Maybe one thing is, I see a lot of the questions that they have.
409 00:32:07.760 ⇒ 00:32:21.769 Gabriel Lam: are something like, you know, what specific problems were people experiencing? Yeah. What was the solution? What metrics changed? And oftentimes, like, maybe as an interviewee, you’re also just like, where do I find this information? Maybe it exists already.
410 00:32:22.090 ⇒ 00:32:29.960 Gabriel Lam: We have Slack messages, right? So, would any of those Will that information exist?
411 00:32:30.240 ⇒ 00:32:32.600 Gabriel Lam: in Slack messages already, that we could…
412 00:32:34.190 ⇒ 00:32:34.900 Samuel Roberts: I just want to…
413 00:32:34.900 ⇒ 00:32:36.020 Mustafa Raja: Might exist.
414 00:32:37.470 ⇒ 00:32:40.040 Mustafa Raja: But we’ll have to classify them somehow.
415 00:32:40.040 ⇒ 00:32:40.430 Gabriel Lam: Okay.
416 00:32:40.430 ⇒ 00:32:45.819 Samuel Roberts: Yeah. My other thought is that, I think, at least when I did one of these interviews a week or two ago.
417 00:32:46.030 ⇒ 00:32:49.559 Samuel Roberts: She kind of wanted to hear it in my voice.
418 00:32:49.560 ⇒ 00:32:50.280 Gabriel Lam: Yeah.
419 00:32:50.280 ⇒ 00:32:51.110 Samuel Roberts: So that…
420 00:32:51.280 ⇒ 00:33:06.689 Samuel Roberts: there’s something to this, like, you know, like, a blog post doesn’t just come off as, like, you know, an AI-generated summary of, like, Slack messages. So I think there’s something where, like, there’s definitely some of that, and maybe we can add that in. But I think the questions in the transcript is kind of going to be the…
421 00:33:06.930 ⇒ 00:33:07.660 Samuel Roberts: Mike.
422 00:33:09.550 ⇒ 00:33:13.600 Samuel Roberts: I think that’s okay that the person might have to answer that question and give more of a, like.
423 00:33:14.130 ⇒ 00:33:21.299 Samuel Roberts: you know, explain like I’m 5 kind of answer for something that might be already available somewhere else.
424 00:33:21.300 ⇒ 00:33:22.320 Gabriel Lam: Fair enough.
425 00:33:23.290 ⇒ 00:33:25.460 Samuel Roberts: So, okay, let’s… let’s think about…
426 00:33:25.940 ⇒ 00:33:31.570 Samuel Roberts: the transcript, then, if this is enough for, like… this can be, like, a very basic UI that we can put together, right?
427 00:33:32.630 ⇒ 00:33:39.410 Samuel Roberts: this interview UI and AI agent, we need things like…
428 00:33:40.480 ⇒ 00:33:43.150 Samuel Roberts: Question list, or what do we call that, like, interview?
429 00:33:44.300 ⇒ 00:33:47.620 Gabriel Lam: There’s just, like, an interview guide, yeah.
430 00:33:47.620 ⇒ 00:33:48.940 Samuel Roberts: And, yeah.
431 00:33:49.130 ⇒ 00:33:54.870 Samuel Roberts: There could be a little bit of, like, Well, I don’t know.
432 00:33:55.590 ⇒ 00:33:59.299 Samuel Roberts: I mean, there’s definitely some, like, background info that could be supplied to this as well.
433 00:34:00.650 ⇒ 00:34:02.740 Samuel Roberts: I don’t know what that looks like at this point.
434 00:34:03.520 ⇒ 00:34:10.110 Samuel Roberts: Like… Client info… I don’t know.
435 00:34:10.489 ⇒ 00:34:12.420 Samuel Roberts: Maybe that’s wrong at this point, but…
436 00:34:12.420 ⇒ 00:34:16.460 Gabriel Lam: It’s part of the guide, like, the first, I can… if I…
437 00:34:16.469 ⇒ 00:34:22.839 Samuel Roberts: If that’s in there, okay, that’s… yeah, no, this is… I’m just not sure what… I don’t remember what was in there, but interview guide, and then…
438 00:34:23.719 ⇒ 00:34:31.509 Samuel Roberts: That… I mean, that might be the big chunk of it. Like, there will be, other things that we need to start, like, other prompts?
439 00:34:32.919 ⇒ 00:34:37.319 Samuel Roberts: to dig deeper. I don’t know what that looked like at this point, though.
440 00:34:38.049 ⇒ 00:34:42.369 Samuel Roberts: Until we dig in and start building this, like, you know…
441 00:34:43.239 ⇒ 00:34:47.959 Samuel Roberts: Speech-to-speech kind of thing, and see how good the prompts are?
442 00:34:48.799 ⇒ 00:34:51.719 Samuel Roberts: We’re gonna have to really… you know, there’s gonna be a bunch of, like.
443 00:34:52.729 ⇒ 00:34:55.899 Samuel Roberts: Maybe not a bunch, but at least one big system prompt.
444 00:34:57.579 ⇒ 00:35:03.619 Samuel Roberts: Right, like, interview… interviewer system.
445 00:35:04.119 ⇒ 00:35:04.929 Samuel Roberts: Prompt.
446 00:35:05.649 ⇒ 00:35:06.489 Samuel Roberts: Right.
447 00:35:08.660 ⇒ 00:35:10.140 Samuel Roberts: Then the guide…
448 00:35:10.420 ⇒ 00:35:14.659 Samuel Roberts: kind of goes into that, but I think there’s also, like, a list of questions that are… oops.
449 00:35:15.230 ⇒ 00:35:17.259 Samuel Roberts: A list of questions that are…
450 00:35:19.260 ⇒ 00:35:23.190 Samuel Roberts: And I… this is where I’m losing it a little bit, because I don’t know how much we need to, like…
451 00:35:23.380 ⇒ 00:35:28.190 Samuel Roberts: Tell the agent what to ask, or give it a list of questions to make sure it asks, or…
452 00:35:29.620 ⇒ 00:35:30.739 Gabriel Lam: I think the interview…
453 00:35:30.740 ⇒ 00:35:32.510 Mustafa Raja: Mission pretty much has it.
454 00:35:33.410 ⇒ 00:35:33.800 Samuel Roberts: Say that again?
455 00:35:33.800 ⇒ 00:35:37.410 Mustafa Raja: donation dog that Anna shared with Gabe.
456 00:35:38.330 ⇒ 00:35:43.799 Samuel Roberts: Oh, no, I understand that. I think… well, yeah, there is a list of questions. What I’m trying to dig into now is…
457 00:35:45.000 ⇒ 00:35:47.120 Samuel Roberts: When we build this agent.
458 00:35:48.010 ⇒ 00:35:48.780 Mustafa Raja: How does it be?
459 00:35:49.670 ⇒ 00:35:52.950 Samuel Roberts: How does it dig in beyond those questions, you know?
460 00:35:52.950 ⇒ 00:35:53.849 Mustafa Raja: Yeah.
461 00:35:53.850 ⇒ 00:36:04.949 Samuel Roberts: like, if we just give it a big system prompt that includes those questions, will it get to those? Or does it… do we need to build an agent that cycles through those questions and gets answers and then digs deeper? You know what I mean? It’s either…
462 00:36:04.950 ⇒ 00:36:05.740 Mustafa Raja: Yes.
463 00:36:06.160 ⇒ 00:36:10.860 Samuel Roberts: A big thing we tell it, and then the agent is smart enough to know that as it goes back and forth.
464 00:36:11.540 ⇒ 00:36:12.040 Mustafa Raja: Yes.
465 00:36:12.040 ⇒ 00:36:16.919 Samuel Roberts: Or… It’s a multi-agent kind of flow.
466 00:36:18.040 ⇒ 00:36:22.019 Samuel Roberts: Where it’s checking boxes of questions that have been asked, and maybe goes deeper on a few.
467 00:36:22.650 ⇒ 00:36:29.130 Mustafa Raja: It’s like a cursor plan. When we give it something, it asks us more questions.
468 00:36:29.630 ⇒ 00:36:30.680 Samuel Roberts: Right.
469 00:36:30.680 ⇒ 00:36:34.610 Mustafa Raja: on getting it right, I feel this is… this is where we want it to be.
470 00:36:35.680 ⇒ 00:36:43.499 Samuel Roberts: Yeah, yeah, so it’s gonna have to have some, like, clarifying questions, and I’m just not sure yet how to… how this is gonna work, that’s just kind of a big question mark.
471 00:36:44.730 ⇒ 00:36:50.329 Samuel Roberts: So I don’t want to over-prescribe it here either, because I think that’s something we’ll have to build a little proof of concept and see how it does.
472 00:36:50.380 ⇒ 00:36:51.320 Mustafa Raja: We’ll speak.
473 00:36:51.320 ⇒ 00:36:51.660 Samuel Roberts: issues.
474 00:36:51.660 ⇒ 00:36:53.770 Mustafa Raja: Definitely going to help us figure this out.
475 00:36:53.770 ⇒ 00:37:00.640 Samuel Roberts: Yeah, and I think that’s probably the first thing we want to tackle, because, like, this… this we can do. This is the third piece that is an AI step, right?
476 00:37:01.650 ⇒ 00:37:09.450 Samuel Roberts: And here we have, two prompts for generation, is that what we call it? Like, two,
477 00:37:10.810 ⇒ 00:37:12.379 Samuel Roberts: pre-made prompts, I guess?
478 00:37:12.380 ⇒ 00:37:13.100 Gabriel Lam: Yeah.
479 00:37:13.590 ⇒ 00:37:19.030 Samuel Roberts: made prompts… That big.
480 00:37:19.460 ⇒ 00:37:26.889 Samuel Roberts: Transcript as input. Okay, so I think for now, we’ll just assume we’re gonna use those and make a step that does
481 00:37:27.410 ⇒ 00:37:31.450 Samuel Roberts: It’s like a… is it a two-step process, then, where it takes one… one takes the output of the other?
482 00:37:31.450 ⇒ 00:37:32.490 Gabriel Lam: Yeah, pretty much.
483 00:37:32.650 ⇒ 00:37:42.279 Samuel Roberts: Okay, then yeah, that should be… that should be pretty simple to process off once the interview. So I think that’s… you know, there might be some tuning that has to happen here, because we’re not…
484 00:37:42.720 ⇒ 00:37:47.100 Samuel Roberts: gonna use ChatGPT, we’re gonna use, like, GPT.
485 00:37:47.650 ⇒ 00:37:54.089 Samuel Roberts: like, the models? So, like, there might be a few other things we have to add so that it does it as good as it does there, but I’m not that worried about that.
486 00:37:54.090 ⇒ 00:37:54.720 Gabriel Lam: Yeah.
487 00:37:55.100 ⇒ 00:38:01.999 Samuel Roberts: this is definitely the big question mark, because I think… and like I said on the call earlier, I think that’s also the one that even if we only get the interviews to be automated.
488 00:38:02.490 ⇒ 00:38:04.410 Samuel Roberts: Will be a huge, like…
489 00:38:04.770 ⇒ 00:38:05.700 Mustafa Raja: Yeah, big win.
490 00:38:05.700 ⇒ 00:38:12.330 Samuel Roberts: levered. Yeah, exactly. The rest of it could be, like, next week, or whatever. Like, this is the one I want to focus on nailing, kind of the way we nailed the…
491 00:38:12.330 ⇒ 00:38:12.910 Mustafa Raja: Yes.
492 00:38:13.120 ⇒ 00:38:14.750 Samuel Roberts: the stand-up…
493 00:38:15.100 ⇒ 00:38:24.660 Samuel Roberts: thing before we added historical data and stuff. And I just don’t know, like, again, this is where it could be multiple days, it could be, you know.
494 00:38:25.400 ⇒ 00:38:31.149 Samuel Roberts: cursor nails it in one shot, and we just find it. Like, I just don’t know, it’s such a, you know…
495 00:38:31.430 ⇒ 00:38:36.240 Samuel Roberts: unknown thing at this point, but I think that’s… that’s the plan. Does that sound good to everyone?
496 00:38:36.240 ⇒ 00:38:43.400 Gabriel Lam: Yeah, I have a more general question about, like, we are talking about automating these interviews.
497 00:38:43.750 ⇒ 00:38:44.370 Samuel Roberts: Yeah.
498 00:38:45.600 ⇒ 00:38:50.930 Gabriel Lam: For lack of a better word, like, what exactly happens when you are an interviewee?
499 00:38:51.990 ⇒ 00:38:52.870 Samuel Roberts: Yes.
500 00:38:53.110 ⇒ 00:38:56.470 Samuel Roberts: That is a good question.
501 00:38:56.470 ⇒ 00:39:02.370 Gabriel Lam: Are we automating? Like, we’re gonna still schedule it with a real person, and the problems are just helping us
502 00:39:02.530 ⇒ 00:39:06.850 Gabriel Lam: move it along? Are we talking to, like, essentially a bot? .
503 00:39:06.850 ⇒ 00:39:12.739 Samuel Roberts: we’re talking to a bot, is the kind of way I’ve been envisioning it, and I kind of think that’s how Huntan was envisioning it, to make it so that…
504 00:39:12.740 ⇒ 00:39:17.490 Gabriel Lam: Yeah, that’s… yeah. I’m just trying to call it out as it is, to be like, okay, we’re all…
505 00:39:17.830 ⇒ 00:39:20.450 Gabriel Lam: We all know what… we’re all thinking the same thing.
506 00:39:20.730 ⇒ 00:39:26.320 Samuel Roberts: Yeah, because, I mean, okay, so what I was thinking initially when he mentioned this PRD last week was…
507 00:39:26.320 ⇒ 00:39:27.130 Gabriel Lam: a little bit.
508 00:39:27.430 ⇒ 00:39:39.439 Samuel Roberts: Yeah, I remember doing, like, interviews to, like, a camera where there were questions, and I just had, like, you know, here’s a question, and in 10 seconds, it’s gonna record your answer to it. Yeah. Which is, like, the very, like.
509 00:39:39.690 ⇒ 00:39:42.150 Samuel Roberts: lo-fi, non-AI version of this.
510 00:39:43.370 ⇒ 00:39:46.099 Samuel Roberts: But I think what we can do is…
511 00:39:47.020 ⇒ 00:39:55.340 Samuel Roberts: Make this so that you’re gonna have a… Conversation with the bot, Like, a voice conversation.
512 00:39:56.230 ⇒ 00:40:01.490 Samuel Roberts: with a bot, asking you the questions that Hannah is asking us right now during the process.
513 00:40:01.490 ⇒ 00:40:01.990 Gabriel Lam: Yo.
514 00:40:02.310 ⇒ 00:40:04.090 Samuel Roberts: Which…
515 00:40:04.290 ⇒ 00:40:10.460 Samuel Roberts: I think is gonna… you know, it frees up Hannah, basically, which I think is kind of the biggest thing here.
516 00:40:11.590 ⇒ 00:40:14.600 Samuel Roberts: What makes that…
517 00:40:15.240 ⇒ 00:40:20.570 Samuel Roberts: you know, one session kind of thing is where I’m not 100% sure how we do that, like…
518 00:40:21.720 ⇒ 00:40:29.110 Samuel Roberts: You know, notifying users, initiating the interview, you know, setting, like, a time box or something on it, like, that becomes a little bit of a…
519 00:40:29.490 ⇒ 00:40:31.560 Samuel Roberts: Creative, kind of, figuring it out.
520 00:40:31.560 ⇒ 00:40:34.780 Mustafa Raja: Notification can be some… Of course. Slack.
521 00:40:35.460 ⇒ 00:40:41.249 Samuel Roberts: Oh, yeah, yeah. I’m thinking, like, if I sit down and I click, like, start interview.
522 00:40:41.710 ⇒ 00:40:42.960 Gabriel Lam: No, no, no, all good, all good.
523 00:40:42.960 ⇒ 00:40:44.349 Samuel Roberts: Did we time box that?
524 00:40:45.610 ⇒ 00:40:46.070 Mustafa Raja: Do we…
525 00:40:46.070 ⇒ 00:40:56.069 Samuel Roberts: do we say this is, like, a one-time thing? Like, we want it to be kind of synchronous so that there’s a conversation happening. Like, I don’t want it to be like, oh, I’ll come back and finish this later and forget where I was.
526 00:40:57.600 ⇒ 00:40:58.390 Gabriel Lam: I mean.
527 00:40:58.390 ⇒ 00:41:00.040 Samuel Roberts: He’s not sure how we do all that yet.
528 00:41:00.040 ⇒ 00:41:00.950 Gabriel Lam: a second time.
529 00:41:00.950 ⇒ 00:41:08.010 Samuel Roberts: even matters, maybe I’m overthinking it, but I think there’s something there where the kind of synchronous nature of having a
530 00:41:08.230 ⇒ 00:41:12.159 Samuel Roberts: 15-30 minute chat about this project.
531 00:41:12.160 ⇒ 00:41:13.059 Gabriel Lam: Yeah, and then I.
532 00:41:13.060 ⇒ 00:41:15.420 Samuel Roberts: Is the… is the good input, you know?
533 00:41:15.460 ⇒ 00:41:16.140 Gabriel Lam: And then…
534 00:41:16.410 ⇒ 00:41:21.160 Samuel Roberts: Like, it’s one… because imagine if Hannah, like, just sent you the questions, and you just sent back answers.
535 00:41:21.520 ⇒ 00:41:24.309 Gabriel Lam: It’s different than the kind of… Yeah, it was…
536 00:41:24.310 ⇒ 00:41:25.580 Samuel Roberts: Slightly more…
537 00:41:25.580 ⇒ 00:41:26.620 Gabriel Lam: 20 incense that way.
538 00:41:26.620 ⇒ 00:41:28.310 Samuel Roberts: Reformed nature of a conversation.
539 00:41:28.310 ⇒ 00:41:46.530 Gabriel Lam: is like, I feel bad for you. Yeah. I’m curious what has been the biggest difference between, like, if you were to just send back a bunch of answers versus, like, a voice interview for these case studies, like, are we looking for, like, a human voice to the copy, or…
540 00:41:47.150 ⇒ 00:41:53.580 Samuel Roberts: That’s a good question. That’s kind of the impression that I got. Yeah. Like, if it’s a blog post, you want it… it wants to read, like, a.
541 00:41:53.880 ⇒ 00:41:56.239 Gabriel Lam: Like, someone wrote it on it. Yeah.
542 00:41:57.080 ⇒ 00:42:02.520 Gabriel Lam: But that’s definitely a good question for Hannah, then. I think…
543 00:42:02.680 ⇒ 00:42:06.749 Samuel Roberts: I think Tom’s also right that, like, the voice is just a faster way to do things sometimes.
544 00:42:06.750 ⇒ 00:42:07.720 Gabriel Lam: Yeah, I agree.
545 00:42:08.060 ⇒ 00:42:11.029 Samuel Roberts: And so, like, there’s definitely an element where if we can just have someone
546 00:42:11.220 ⇒ 00:42:14.939 Samuel Roberts: Ask you a question and you answer it, rather than, like, have to type out an answer.
547 00:42:14.940 ⇒ 00:42:15.640 Gabriel Lam: -
548 00:42:15.960 ⇒ 00:42:16.870 Samuel Roberts: Even if there’s not
549 00:42:17.260 ⇒ 00:42:27.169 Samuel Roberts: even if the benefits of the kind of top-of-the-head talking, top-of-mind kind of thing isn’t important, the, like, fact that I was able to do that in
550 00:42:27.410 ⇒ 00:42:31.950 Samuel Roberts: 20 seconds instead of typing out an answer for a minute or two, and then going, you know?
551 00:42:31.950 ⇒ 00:42:33.330 Gabriel Lam: The, the exact…
552 00:42:33.330 ⇒ 00:42:34.759 Samuel Roberts: I think that’s a benefit either way.
553 00:42:34.760 ⇒ 00:42:38.129 Gabriel Lam: She was like, in a real-life situation, is this how you’re gonna…
554 00:42:39.120 ⇒ 00:42:39.460 Samuel Roberts: Okay.
555 00:42:39.460 ⇒ 00:42:41.100 Gabriel Lam: It’s like.
556 00:42:41.100 ⇒ 00:42:45.419 Samuel Roberts: other thoughts or questions? This is really good, I’m enjoying this, I want to keep digging in more.
557 00:42:46.580 ⇒ 00:42:47.849 Gabriel Lam: The testing room.
558 00:42:48.100 ⇒ 00:42:49.790 Samuel Roberts: What… what else?
559 00:42:49.790 ⇒ 00:42:52.920 Gabriel Lam: I like the time boxing that you mentioned.
560 00:42:52.920 ⇒ 00:42:53.270 Samuel Roberts: Yeah.
561 00:42:53.270 ⇒ 00:42:55.040 Gabriel Lam: As a total.
562 00:42:55.900 ⇒ 00:42:56.480 Samuel Roberts: Oh, how do I…
563 00:42:56.480 ⇒ 00:43:00.139 Gabriel Lam: overall time box instead of, like, a per question, I think when it’s…
564 00:43:00.140 ⇒ 00:43:03.180 Samuel Roberts: Definitely, I think, I think it’s, like, time box, like, as if it was a meeting.
565 00:43:03.180 ⇒ 00:43:04.130 Gabriel Lam: Yeah.
566 00:43:04.280 ⇒ 00:43:05.040 Gabriel Lam: No.
567 00:43:05.350 ⇒ 00:43:12.510 Gabriel Lam: And I think it also sets expectations for people. It’s like, hey, okay, this is gonna take, like, 30 minutes max. It could be earlier, but…
568 00:43:12.710 ⇒ 00:43:13.420 Gabriel Lam: Yeah.
569 00:43:13.980 ⇒ 00:43:14.610 Samuel Roberts: No, you’re absolutely right.
570 00:43:14.610 ⇒ 00:43:18.090 Mustafa Raja: I want the agent to be aware of the time, also.
571 00:43:18.390 ⇒ 00:43:23.350 Samuel Roberts: Yeah, this is where… and this is exactly where I’m like, I don’t know how best to handle that, I don’t know how… like, there’s gonna be a lot.
572 00:43:23.350 ⇒ 00:43:24.340 Mustafa Raja: Yes.
573 00:43:24.670 ⇒ 00:43:31.420 Samuel Roberts: keeping a timer, maybe, is important, and then feeding that into it. I don’t know, we’ll find out, but I think that’s an important step that is, like…
574 00:43:31.680 ⇒ 00:43:37.639 Samuel Roberts: The interviewer, i.e. the bot, and the interviewee knows that there’s a set
575 00:43:38.490 ⇒ 00:43:41.480 Samuel Roberts: I don’t necessarily want a countdown, but, like, a, you know, a set end time.
576 00:43:41.880 ⇒ 00:43:42.600 Gabriel Lam: Yeah.
577 00:43:42.600 ⇒ 00:43:48.850 Samuel Roberts: Whether or not that’s a countdown or not, you know, I don’t want people to be rushing. Okay, I think that’s good.
578 00:43:52.980 ⇒ 00:44:00.040 Gabriel Lam: Are we imagining a sort of… you know how sometimes if you have, like, a virtual interview, or, like, an async interview, you get, like.
579 00:44:00.300 ⇒ 00:44:03.470 Gabriel Lam: You know, the question, and then a set time to answer.
580 00:44:05.190 ⇒ 00:44:11.839 Gabriel Lam: Or is it gonna be more like, you’re gonna see a message real time, and then you keep going? Is it, like, one long conversation? Is it gonna be split up into…
581 00:44:12.200 ⇒ 00:44:14.730 Gabriel Lam: Multiple sets of conversations.
582 00:44:14.880 ⇒ 00:44:22.079 Samuel Roberts: Yeah, I’m… picturing… One long conversation?
583 00:44:24.610 ⇒ 00:44:29.829 Samuel Roberts: And I think that’s partially because I’m picturing… I want to make this as, like.
584 00:44:30.870 ⇒ 00:44:32.910 Samuel Roberts: sitting down with Hannah as possible.
585 00:44:34.110 ⇒ 00:44:38.329 Samuel Roberts: I want to take, like, what Hannah does in that interview and make a bot do it.
586 00:44:38.650 ⇒ 00:44:44.600 Samuel Roberts: And if that’s the right way to do it or not, I don’t know, but that’s the current way we do it, you know?
587 00:44:45.150 ⇒ 00:44:45.990 Gabriel Lam: Yeah.
588 00:44:45.990 ⇒ 00:45:05.690 Samuel Roberts: Maybe there’s a better way, maybe it is, like, okay, let’s go through the first step of the process, let’s go through the second step. You know, like, there could be a way that we break it out a little bit more and say there’s, like, okay, here’s the five sections of your interview, click into one, do the… you know what I mean? Like, there could be something where we break it up differently, because it’s not a human.
589 00:45:06.000 ⇒ 00:45:06.620 Gabriel Lam: tomorrow.
590 00:45:06.620 ⇒ 00:45:08.919 Samuel Roberts: But my… my gut is to try to…
591 00:45:09.260 ⇒ 00:45:12.790 Samuel Roberts: Replicate what we currently have, and just be able to scale it out.
592 00:45:12.790 ⇒ 00:45:15.790 Gabriel Lam: Then, also, Trying to… Yes.
593 00:45:15.790 ⇒ 00:45:17.369 Mustafa Raja: Yes, I’m thinking the same.
594 00:45:17.370 ⇒ 00:45:18.590 Gabriel Lam: Kind of pushbacker.
595 00:45:18.590 ⇒ 00:45:23.700 Samuel Roberts: Whether or not that works is another question. Whether or not it’s as good is one thing. Whether or not it’s…
596 00:45:24.420 ⇒ 00:45:36.380 Samuel Roberts: I don’t want to say easy to implement, but implementable is one thing, I guess, too, you know? Like, if we dig into this speech-to-speech thing, and it’s like, okay, it can only handle so much context at a time.
597 00:45:36.380 ⇒ 00:45:37.440 Gabriel Lam: You can bring it up.
598 00:45:37.440 ⇒ 00:45:44.329 Samuel Roberts: There’s… there’s… I don’t know, I just… I don’t know yet how best to handle that, and I think that’s gonna be where, if we can get something.
599 00:45:44.330 ⇒ 00:45:45.399 Gabriel Lam: and then.
600 00:45:45.400 ⇒ 00:45:48.220 Samuel Roberts: Like, a simple chat.
601 00:45:48.220 ⇒ 00:45:48.760 Gabriel Lam: Me neither.
602 00:45:48.760 ⇒ 00:45:56.239 Samuel Roberts: Maybe not… maybe not with all the given questions, but even just any kind of… Chat, interview…
603 00:45:57.050 ⇒ 00:46:02.570 Samuel Roberts: that records the transcript and everything out. We can kind of see, like, what it’s capable of.
604 00:46:02.570 ⇒ 00:46:04.280 Gabriel Lam: Yep.
605 00:46:04.820 ⇒ 00:46:08.049 Samuel Roberts: Because at this point, it’s still just, like, there’s so many questions.
606 00:46:08.050 ⇒ 00:46:08.560 Gabriel Lam: Top of your thoughts.
607 00:46:08.560 ⇒ 00:46:12.939 Samuel Roberts: that I don’t know what will dictate the kind of rest of the design decisions.
608 00:46:12.940 ⇒ 00:46:13.440 Gabriel Lam: on.
609 00:46:13.440 ⇒ 00:46:16.590 Samuel Roberts: I think Best case scenario, it’s a…
610 00:46:16.590 ⇒ 00:46:17.240 Gabriel Lam: some of it I have.
611 00:46:17.240 ⇒ 00:46:18.280 Samuel Roberts: conversation.
612 00:46:18.280 ⇒ 00:46:19.290 Gabriel Lam: with.
613 00:46:19.290 ⇒ 00:46:25.690 Samuel Roberts: you know, you answer a question, the bot says, okay, great, how about, like, tell me more about this? And it just keeps going like that.
614 00:46:25.690 ⇒ 00:46:26.079 Gabriel Lam: Let’s see if there’s.
615 00:46:26.080 ⇒ 00:46:32.770 Samuel Roberts: I think next step down is something like… there’s, like, 3 questions, and it asks you those in a certain section, and…
616 00:46:33.170 ⇒ 00:46:35.850 Samuel Roberts: Then you jump to the next section so we don’t overload the context.
617 00:46:36.640 ⇒ 00:46:39.449 Samuel Roberts: I don’t know. That’s… this is… this is a big, like.
618 00:46:39.590 ⇒ 00:46:40.889 Gabriel Lam: been, so…
619 00:46:40.890 ⇒ 00:46:49.049 Samuel Roberts: this is why it’s kind of cool to figure out, but until we do it, I don’t know. So I don’t want to make too many assumptions there. But I would say, yeah, let’s… let’s…
620 00:46:49.640 ⇒ 00:46:51.850 Samuel Roberts: Let’s start digging into it.
621 00:46:51.850 ⇒ 00:46:56.710 Mustafa Raja: To me, to me, this project is really exciting. I haven’t worked with voice before.
622 00:46:56.710 ⇒ 00:46:57.429 Gabriel Lam: the action.
623 00:46:57.430 ⇒ 00:46:57.900 Samuel Roberts: Yeah.
624 00:46:57.900 ⇒ 00:46:58.599 Mustafa Raja: a new thing.
625 00:46:59.300 ⇒ 00:47:02.170 Mustafa Raja: And, yeah, I’m just excited about…
626 00:47:02.350 ⇒ 00:47:04.050 Mustafa Raja: How this is going to turn out.
627 00:47:04.700 ⇒ 00:47:10.439 Samuel Roberts: Yeah, me too, me too. I’m trying to think how we can break this down and be smaller, but I think the…
628 00:47:10.440 ⇒ 00:47:10.820 Gabriel Lam: Valley.
629 00:47:10.820 ⇒ 00:47:13.930 Samuel Roberts: I think the first step is probably, like, if you want to take a stab at it…
630 00:47:13.930 ⇒ 00:47:15.540 Gabriel Lam: I’m trying to distingu.
631 00:47:15.540 ⇒ 00:47:20.519 Samuel Roberts: just some… kind of the way Casey put together that, like, the back end with a very mock front end.
632 00:47:20.520 ⇒ 00:47:22.189 Gabriel Lam: I’m with this sunlight, like, walking.
633 00:47:22.190 ⇒ 00:47:27.970 Samuel Roberts: you’re excited enough about it and want to try to dig into the master stuff and throw together, like, a interview page.
634 00:47:28.380 ⇒ 00:47:29.619 Samuel Roberts: And see what it can do.
635 00:47:31.880 ⇒ 00:47:45.089 Samuel Roberts: I would say that’s probably the best first step, whether or not it even is, like, gonna directly be the code we use, or at least just the proof of concept that it can work this way, or we need to break it up because of the context, or whatever we can learn from that by tomorrow would be great.
636 00:47:45.300 ⇒ 00:47:46.960 Gabriel Lam: If that’s something.
637 00:47:46.960 ⇒ 00:47:51.930 Samuel Roberts: want to try, Mustafa, I’m totally down to let you run with it, but… what was that game?
638 00:47:51.930 ⇒ 00:47:52.590 Mustafa Raja: Nevertheless.
639 00:47:52.590 ⇒ 00:47:54.010 Gabriel Lam: I’m… I’m just, like…
640 00:47:54.670 ⇒ 00:48:06.980 Gabriel Lam: I was just thinking what it might look like. I was, like, imagining almost something as simple as, like, a chat GPT, you know, chat box, where you’re just able to talk, and once you pause for, like, 5 seconds, 5-10 seconds.
641 00:48:06.980 ⇒ 00:48:07.320 Samuel Roberts: Exactly.
642 00:48:07.320 ⇒ 00:48:13.390 Gabriel Lam: It’ll know you stop, and then the next… it’ll count that as, like, your response and give you the next answer.
643 00:48:13.850 ⇒ 00:48:17.949 Samuel Roberts: Yeah, yeah, there was some cool stuff on that link that kind of talked about it.
644 00:48:18.480 ⇒ 00:48:19.860 Gabriel Lam: their phones, I think.
645 00:48:19.860 ⇒ 00:48:20.400 Samuel Roberts: No.
646 00:48:20.720 ⇒ 00:48:23.839 Samuel Roberts: I see. Maybe it’s not this one, but there was a few different voice things that it had.
647 00:48:23.840 ⇒ 00:48:25.459 Gabriel Lam: of educating, but.
648 00:48:25.460 ⇒ 00:48:29.760 Samuel Roberts: That, like, waits for a pause, and you can configure that stuff, and… Not quite.
649 00:48:29.930 ⇒ 00:48:31.379 Gabriel Lam: Our conversation last time was.
650 00:48:31.380 ⇒ 00:48:36.670 Samuel Roberts: Yeah, I mean, I think there’s a lot… I mean, that’s kind of what I’m picturing, like, it’ll show you the log of the chat.
651 00:48:36.670 ⇒ 00:48:40.590 Gabriel Lam: Differentiator should be people who have had coached experience versus people.
652 00:48:40.590 ⇒ 00:48:49.539 Samuel Roberts: Yeah, just like when you hit, you know, chat with voice on ChatGPT or something, and it just kind of has that natural back and forth as much as AI can do it.
653 00:48:49.940 ⇒ 00:48:51.410 Samuel Roberts: I said, I think that’s probably it.
654 00:48:52.950 ⇒ 00:48:57.760 Samuel Roberts: The question becomes, like, yeah, the context window, and whether or not we need to break that up.
655 00:48:58.950 ⇒ 00:49:00.829 Gabriel Lam: How quickly we can do this stuff.
656 00:49:00.830 ⇒ 00:49:03.389 Samuel Roberts: You know, what is the… how real time does it feel?
657 00:49:04.490 ⇒ 00:49:11.079 Samuel Roberts: Which, again, is not the biggest thing, because we’re… this is internal, and everyone’s using it, so we can kind of, like… this doesn’t need to be…
658 00:49:11.210 ⇒ 00:49:16.329 Samuel Roberts: Perfect, but… As realistic as possible, we’ll get good interviews and things.
659 00:49:16.330 ⇒ 00:49:17.030 Gabriel Lam: Yep.
660 00:49:19.560 ⇒ 00:49:21.180 Samuel Roberts: Yeah, other thoughts?
661 00:49:27.250 ⇒ 00:49:33.729 Gabriel Lam: I think that’s it for now, but I can always… if I find something online that I’m like, oh, I think this might be helpful, I can always just put it in the Slack.
662 00:49:34.060 ⇒ 00:49:36.159 Samuel Roberts: Definitely, yeah, yeah, yeah. Totally, totally.
663 00:49:38.320 ⇒ 00:49:39.280 Samuel Roberts: Cool.
664 00:49:40.310 ⇒ 00:49:42.340 Samuel Roberts: Mustafa, what’s the rest of your day look like?
665 00:49:43.300 ⇒ 00:49:44.000 Gabriel Lam: And…
666 00:49:44.620 ⇒ 00:49:45.580 Mustafa Raja: EOC.
667 00:49:47.400 ⇒ 00:49:48.220 Samuel Roberts: Sorry?
668 00:49:48.230 ⇒ 00:49:50.099 Mustafa Raja: I’ll be working on this POC.
669 00:49:50.300 ⇒ 00:49:53.670 Samuel Roberts: Okay, cool, yeah, this is… I just wanted to make sure I wasn’t, like, taking you away from anything else, but yeah.
670 00:49:53.670 ⇒ 00:49:54.300 Mustafa Raja: I would say.
671 00:49:54.300 ⇒ 00:49:55.730 Samuel Roberts: Hey, yeah, just.
672 00:49:56.320 ⇒ 00:49:59.919 Gabriel Lam: Yeah, try running with it in cursor. If you have any questions, like, feel free to…
673 00:49:59.990 ⇒ 00:50:01.319 Mustafa Raja: Oh, yeah. They need.
674 00:50:01.600 ⇒ 00:50:08.469 Samuel Roberts: But yeah, if you can just, you know, anything stood up and demoed or whatever would be cool to know, and whatever learnings you can get from that.
675 00:50:08.630 ⇒ 00:50:09.730 Samuel Roberts: Would be.
676 00:50:10.270 ⇒ 00:50:16.530 Samuel Roberts: Great, I guess, yeah, I think that’s the best way to do it, and then we can… running, you know, maybe we can do some magic patterns for this later.
677 00:50:16.530 ⇒ 00:50:16.880 Gabriel Lam: Yeah.
678 00:50:17.820 ⇒ 00:50:42.740 Samuel Roberts: And then this step should be just kind of an automatic thing that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transient
679 00:50:42.740 ⇒ 00:50:51.170 Samuel Roberts: that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the transiting that happens, once the trans…
680 00:50:51.170 ⇒ 00:50:52.559 Gabriel Lam: Do users feel comfortable?
681 00:50:52.560 ⇒ 00:50:57.389 Samuel Roberts: Transing that happens, once the trans thing that happens, once the transiting that happens, once the transiting that happens, once the trans…
682 00:51:05.210 ⇒ 00:51:06.040 Mustafa Raja: Hey.
683 00:51:06.990 ⇒ 00:51:08.000 Samuel Roberts: Can you guys hear me?
684 00:51:08.740 ⇒ 00:51:11.859 Gabriel Lam: I wasn’t sure if, it was me or you.
685 00:51:11.860 ⇒ 00:51:18.440 Samuel Roberts: No, no, okay. Honestly, the only way I know it does something is that when I’m not talking, I see my audio spiking in the corner.
686 00:51:19.350 ⇒ 00:51:22.119 Mustafa Raja: I was wondering if it’s my internet acting up.
687 00:51:22.120 ⇒ 00:51:22.929 Samuel Roberts: No, no.
688 00:51:22.930 ⇒ 00:51:24.139 Gabriel Lam: I got the meeting, and it came back.
689 00:51:24.140 ⇒ 00:51:43.020 Samuel Roberts: I was wondering why you disappeared. Okay, see, the problem is I don’t hear it, what it’s doing, something about my mic on my webcam and my computer don’t play well together, and it just kind of loops something, so I have no idea when that started, I don’t hear it, though, at all, so I don’t know. Okay, I don’t know what that started with, so…
690 00:51:43.020 ⇒ 00:51:44.740 Gabriel Lam: It was, like, a minute ago.
691 00:51:45.140 ⇒ 00:51:45.990 Samuel Roberts: Okay.
692 00:51:46.320 ⇒ 00:51:53.219 Samuel Roberts: I was just saying, I think… I think the last thing I was saying was, Mustafa, is there anything else you need in order to get started, or are you good to…
693 00:51:53.840 ⇒ 00:51:54.460 Samuel Roberts: Start planning.
694 00:51:54.460 ⇒ 00:52:02.599 Mustafa Raja: Yeah, One thing I’m seeing here is… what do we want to connect this master agent with?
695 00:52:02.760 ⇒ 00:52:07.530 Mustafa Raja: Oh, like, which model? We have, yeah, we have…
696 00:52:08.010 ⇒ 00:52:12.570 Mustafa Raja: Azure, but I don’t think we have the right models deployed.
697 00:52:12.570 ⇒ 00:52:13.180 Samuel Roberts: Yeah.
698 00:52:13.590 ⇒ 00:52:15.620 Samuel Roberts: That’s a good question. I would say…
699 00:52:15.900 ⇒ 00:52:25.009 Samuel Roberts: Let’s just go OpenAI Voice, because I think we want to… If they do have questions about this. Speech-to-speech, yeah, OpenAI real-time voice. Yeah.
700 00:52:26.440 ⇒ 00:52:29.539 Samuel Roberts: and use… what does it say here? GPT-40 looks like it’s good.
701 00:52:29.540 ⇒ 00:52:35.199 Gabriel Lam: Yeah, I would say just try that, play with… if that’s a pretty…
702 00:52:35.200 ⇒ 00:52:36.370 Samuel Roberts: work.
703 00:52:36.370 ⇒ 00:52:36.969 Gabriel Lam: Play with something.
704 00:52:36.970 ⇒ 00:52:37.530 Samuel Roberts: else, but…
705 00:52:37.530 ⇒ 00:52:37.980 Gabriel Lam: It’s not the.
706 00:52:37.980 ⇒ 00:52:39.760 Samuel Roberts: Yeah, let me know if that works, and then…
707 00:52:39.760 ⇒ 00:52:40.180 Gabriel Lam: Yeah.
708 00:52:40.180 ⇒ 00:52:42.029 Samuel Roberts: I would just say, for proof of concept.
709 00:52:42.030 ⇒ 00:52:46.990 Gabriel Lam: as worried about that yet, but… Either connect or set a reminder later.
710 00:52:47.320 ⇒ 00:52:51.169 Gabriel Lam: In that case, maybe we don’t need the static suite. Okay.
711 00:52:51.610 ⇒ 00:52:54.679 Samuel Roberts: Yeah, let me know if there’s any other questions there about it, but if for some reason that doesn’t work.
712 00:52:54.680 ⇒ 00:52:55.350 Gabriel Lam: Basically…
713 00:52:55.350 ⇒ 00:52:57.539 Samuel Roberts: Spin something up, or you can spin something up, let me know.
714 00:52:57.540 ⇒ 00:52:58.349 Gabriel Lam: It’s fine.
715 00:52:58.820 ⇒ 00:52:59.800 Gabriel Lam: Again, my June.
716 00:52:59.800 ⇒ 00:53:04.099 Samuel Roberts: And yeah, I would say the other thing is.
717 00:53:04.100 ⇒ 00:53:04.690 Gabriel Lam: First database.
718 00:53:04.690 ⇒ 00:53:06.359 Samuel Roberts: The sooner you can get something.
719 00:53:06.360 ⇒ 00:53:06.939 Gabriel Lam: And that’s like…
720 00:53:06.940 ⇒ 00:53:07.450 Samuel Roberts: They’re a loop.
721 00:53:07.450 ⇒ 00:53:08.090 Gabriel Lam: to,
722 00:53:08.090 ⇒ 00:53:10.969 Samuel Roberts: Or live, for everyone to test out, will be great to, like.
723 00:53:10.970 ⇒ 00:53:12.190 Gabriel Lam: Proof of concepted.
724 00:53:12.190 ⇒ 00:53:14.129 Samuel Roberts: But also, keep in mind, like.
725 00:53:14.130 ⇒ 00:53:16.379 Gabriel Lam: The context window stuff, and…
726 00:53:16.380 ⇒ 00:53:19.049 Samuel Roberts: How we’re gonna make sure that it answers… asks the right questions.
727 00:53:19.050 ⇒ 00:53:31.440 Gabriel Lam: The same goes with movement. I might create a end of day… end of baseline day, like, cat-catch, if having a dog, you say.
728 00:53:31.440 ⇒ 00:53:33.570 Mustafa Raja: How do you work with OpenAI real-time rules?
729 00:53:33.570 ⇒ 00:53:34.690 Samuel Roberts: It’s too close to me.
730 00:53:34.910 ⇒ 00:53:37.760 Samuel Roberts: Yeah, I have no idea how well this, I’m… yeah, let’s, let’s figure it out.
731 00:53:37.760 ⇒ 00:53:42.169 Gabriel Lam: And then as soon as 24 hours later, you’ll receive a notification.
732 00:53:42.170 ⇒ 00:53:42.810 Samuel Roberts: Good.
733 00:53:42.810 ⇒ 00:53:44.629 Gabriel Lam: Sounds good, thanks guys.
734 00:53:44.630 ⇒ 00:53:45.060 Samuel Roberts: anything.
735 00:53:45.060 ⇒ 00:53:50.199 Gabriel Lam: Thank you so much. Yep. Thanks, guys, bye.