Meeting Title: AI Team | Daily Standup Date: 2024-03-27 Meeting participants: Janna Wong, Amber Lin, Patrik, Miguel De Veyra, Casie Aviles
WEBVTT
1 00:01:26.360 ⇒ 00:01:27.620 Amber Lin: Hi team.
2 00:01:30.010 ⇒ 00:01:30.550 Casie Aviles: Hey! Amber.
3 00:01:30.550 ⇒ 00:01:31.579 Amber Lin: I am going to
4 00:01:31.580 ⇒ 00:01:31.980 Janna Wong: Number
5 00:01:31.980 ⇒ 00:01:33.000 Amber Lin: Share my screen.
6 00:01:33.180 ⇒ 00:01:34.730 Miguel de Veyra: Hey? Everyone! Good morning!
7 00:01:35.310 ⇒ 00:01:36.540 Amber Lin: Good morning!
8 00:01:39.640 ⇒ 00:01:40.800 Miguel de Veyra: Sorry can I be heard
9 00:01:41.810 ⇒ 00:01:42.460 Casie Aviles: Yes.
10 00:01:42.460 ⇒ 00:01:43.570 Miguel de Veyra: Okay. Goodness.
11 00:01:47.180 ⇒ 00:01:48.490 Miguel de Veyra: is Janice coming
12 00:01:49.778 ⇒ 00:01:51.870 Amber Lin: I don’t know. It’s okay.
13 00:01:51.870 ⇒ 00:01:54.230 Miguel de Veyra: Cause. I think, she said. She has meetings today right
14 00:01:55.040 ⇒ 00:01:56.600 Amber Lin: Oh! Did she
15 00:01:56.600 ⇒ 00:02:01.240 Miguel de Veyra: Yeah, she has trainings, although I think, she said, after she’s gonna have training
16 00:02:01.240 ⇒ 00:02:01.750 Amber Lin: Yeah, sure.
17 00:02:01.750 ⇒ 00:02:02.550 Miguel de Veyra: Oh, sure!
18 00:02:02.680 ⇒ 00:02:08.009 Amber Lin: It’s it’s okay. I think we can get started. And when she comes, Phil, we’ll talk.
19 00:02:09.289 ⇒ 00:02:10.020 Amber Lin: So.
20 00:02:10.580 ⇒ 00:02:16.570 Amber Lin: 1st of all, thank you guys so much, so much for working so late yesterday. I am so sorry
21 00:02:17.280 ⇒ 00:02:22.449 Miguel de Veyra: We all got scolded by Utam, so we had to work really late
22 00:02:22.930 ⇒ 00:02:27.369 Amber Lin: So I really thank you guys for doing all of that. That was so good.
23 00:02:28.280 ⇒ 00:02:39.920 Amber Lin: And let’s just start by my updates. So yesterday I was I was doing the
24 00:02:40.380 ⇒ 00:02:44.140 Amber Lin: rollout plans and all the action plans. So
25 00:02:44.830 ⇒ 00:02:48.299 Amber Lin: we might, we might have to think about okay, what kind of
26 00:02:49.082 ⇒ 00:02:52.620 Amber Lin: maybe having Quickstore document and all of that
27 00:02:53.409 ⇒ 00:02:59.899 Amber Lin: I will share that. I’ll share that in our channel, and we can take a if you guys have time can take a look at it.
28 00:03:00.910 ⇒ 00:03:10.739 Amber Lin: and yes, Miguel and I yesterday were mostly. We’re mostly doing all the tickets to
29 00:03:13.340 ⇒ 00:03:19.610 Amber Lin: add requirements and due dates to everything and acceptance. Criteria.
30 00:03:20.237 ⇒ 00:03:23.499 Amber Lin: Miguel, anything else that you were doing yesterday
31 00:03:24.580 ⇒ 00:03:33.190 Miguel de Veyra: It’s probably just fixing this the tickets. And then Casey introduced me to one of cause I lost contact with that guy. But yeah, I’m working on the recruitment stuff, too.
32 00:03:34.370 ⇒ 00:03:35.519 Amber Lin: Hmm, okay.
33 00:03:36.700 ⇒ 00:03:43.059 Miguel de Veyra: And then I believe I I started the bit, but only a bit on the internal team stuff
34 00:03:43.260 ⇒ 00:03:44.260 Amber Lin: Oh, okay.
35 00:03:44.260 ⇒ 00:03:47.459 Miguel de Veyra: So I think Amber in terms. Speaking of that.
36 00:03:47.590 ⇒ 00:03:56.339 Miguel de Veyra: I think that the 1st thing we need to do like the task wise is just me, and you set meetings with the internal teams
37 00:03:56.980 ⇒ 00:04:02.300 Amber Lin: Okay? Yeah. Send meetings for the internal
38 00:04:02.960 ⇒ 00:04:06.079 Miguel de Veyra: Or even identify. Then the different internal teams
39 00:04:08.200 ⇒ 00:04:13.759 Amber Lin: So there’s Ops marketing data teams.
40 00:04:20.640 ⇒ 00:04:21.890 Amber Lin: Interesting.
41 00:04:22.890 ⇒ 00:04:24.140 Amber Lin: That’s it. Right?
42 00:04:24.410 ⇒ 00:04:24.970 Miguel de Veyra: Yep.
43 00:04:25.600 ⇒ 00:04:44.010 Amber Lin: Okay, need to do, assign it to you and due date. Let’s say, Okay, okay, sounds good.
44 00:04:44.970 ⇒ 00:04:46.979 Amber Lin: So that’s something we can do.
45 00:04:47.712 ⇒ 00:04:54.600 Amber Lin: Casey, how about you? I’m so sorry. There was so much stuff that I threw at you. How did
46 00:04:54.600 ⇒ 00:05:02.310 Casie Aviles: Yeah, yeah, sure. Yeah. So I primarily worked on the dashboard. I. This was automatically flagged as done because of
47 00:05:02.860 ⇒ 00:05:06.070 Casie Aviles: the automation. But I put it back in progress, since
48 00:05:06.250 ⇒ 00:05:14.110 Casie Aviles: I think I still need to have this reviewed by the team first, st before we actually have this, you know, for the clients
49 00:05:14.770 ⇒ 00:05:15.169 Amber Lin: Hmm.
50 00:05:16.725 ⇒ 00:05:22.290 Casie Aviles: But yeah, I could. But I did push it live already. And oh.
51 00:05:22.290 ⇒ 00:05:25.820 Casie Aviles: you guys want to see the dashboard right now.
52 00:05:25.820 ⇒ 00:05:26.390 Amber Lin: Short.
53 00:05:27.727 ⇒ 00:05:29.279 Casie Aviles: I’ll have to share.
54 00:05:36.600 ⇒ 00:05:39.409 Casie Aviles: Okay, let me just remove all of these.
55 00:05:41.010 ⇒ 00:05:46.480 Casie Aviles: Okay, so yeah, just I also made the loom video quick loom video. But
56 00:05:46.480 ⇒ 00:05:48.260 Amber Lin: Yeah, I saw, hmm.
57 00:05:48.870 ⇒ 00:05:52.930 Casie Aviles: Yeah, I just removed one of the dashboards since it was just my test.
58 00:05:53.150 ⇒ 00:05:54.929 Casie Aviles: So it’s not as relevant.
59 00:05:57.733 ⇒ 00:06:03.330 Casie Aviles: Yeah, it’s okay. So this is the the latest dashboard. So I tried to get
60 00:06:04.500 ⇒ 00:06:08.739 Casie Aviles: Yeah. I tried to remove all of the the line breaks here the blank.
61 00:06:09.030 ⇒ 00:06:13.990 Casie Aviles: Alright, it’s the nulls. I tried to remove as much as I can
62 00:06:14.340 ⇒ 00:06:15.170 Amber Lin: Oh!
63 00:06:15.420 ⇒ 00:06:20.866 Casie Aviles: And then for here, if there’s no feedback, I just label this as no feedback.
64 00:06:22.800 ⇒ 00:06:25.550 Casie Aviles: Yeah, they should be live since the end.
65 00:06:26.170 ⇒ 00:06:29.120 Casie Aviles: And then here’s like the data that we have
66 00:06:29.330 ⇒ 00:06:30.000 Amber Lin: Hmm.
67 00:06:31.587 ⇒ 00:06:44.910 Casie Aviles: Yeah, I mean, that’s pretty much it. I mean, most of the work that I did was just that’s where it. It’s yeah, I mean, it’s the data part. It’s the data cleaning, because our data was really messy and had a lot of missing values. So
68 00:06:45.020 ⇒ 00:06:47.199 Casie Aviles: that’s where I spent most of the time
69 00:06:48.850 ⇒ 00:06:49.570 Amber Lin: Oh,
70 00:06:51.240 ⇒ 00:06:54.879 Patrik: Oh, where is this data coming from? Just out of curiosity.
71 00:06:55.806 ⇒ 00:07:01.500 Casie Aviles: Yeah, it’s from this new table that I created. It’s on Snowflake. Oh, wait.
72 00:07:04.000 ⇒ 00:07:07.030 Casie Aviles: Yeah, it’s from Snowflake. And this is these are the data that
73 00:07:07.560 ⇒ 00:07:11.830 Casie Aviles: we’re logging from the workflows that we have for the AI agent
74 00:07:16.980 ⇒ 00:07:18.159 Amber Lin: A quick question about that
75 00:07:18.160 ⇒ 00:07:19.340 Casie Aviles: Is it
76 00:07:19.500 ⇒ 00:07:29.780 Amber Lin: Is it? Does it include the thumbs up thumbs down data? It’s like the updated version of that. Or is this still just our testing data
77 00:07:30.570 ⇒ 00:07:40.200 Casie Aviles: Oh, I removed the test data. That’s why I deleted the other dashboard. And this contains the the actual thumbs up thumbs down data from the Csrs
78 00:07:40.980 ⇒ 00:07:47.739 Amber Lin: Oh, okay, so I can move. You know, we have a task of like, Ask added to
79 00:07:49.330 ⇒ 00:07:53.170 Amber Lin: I’ll show you the show you the task in a moment.
80 00:07:55.160 ⇒ 00:07:55.780 Amber Lin: Cool.
81 00:07:57.560 ⇒ 00:08:00.870 Amber Lin: So it’s this one ABC logs backfield
82 00:08:01.660 ⇒ 00:08:03.019 Casie Aviles: Yes, it’s this one
83 00:08:07.550 ⇒ 00:08:11.370 Patrik: How’d you? How’d you generate the table
84 00:08:13.660 ⇒ 00:08:21.130 Casie Aviles: Oh, how did I did? How did I generate? I, yeah, this is the main. This is the the initially. This is the table that we were using.
85 00:08:21.680 ⇒ 00:08:27.690 Casie Aviles: ABC. Bought feedback, and there are some missing columns. So I just made a copy of this
86 00:08:28.963 ⇒ 00:08:35.599 Casie Aviles: and then I also ran this through an AI step that generated the quality score.
87 00:08:38.250 ⇒ 00:08:46.189 Casie Aviles: So yeah, we have all these quality scores. So it’s just checking the input against the output or yeah. And then also the expected.
88 00:08:46.945 ⇒ 00:08:47.700 Casie Aviles: yeah.
89 00:08:48.400 ⇒ 00:08:56.749 Patrik: So we wanted to track something other than that ABC workflow. We have to duplicate the snowflake stuff
90 00:09:00.810 ⇒ 00:09:06.099 Casie Aviles: yeah, I I mean, yeah, I just, I wanted to clean it up. So that’s why I duplicated it.
91 00:09:06.200 ⇒ 00:09:07.440 Casie Aviles: Yeah.
92 00:09:08.120 ⇒ 00:09:10.289 Patrik: Gotcha? Alright, yeah. Just wondering.
93 00:09:11.090 ⇒ 00:09:14.401 Casie Aviles: Okay. But yeah, I guess that’s pretty much it.
94 00:09:15.660 ⇒ 00:09:31.690 Amber Lin: That’s great. Case, can you? If you look at my screen, we have this little task, the ticket of adding feedback and thumbs data to real we have. We have it in the real dashboard. Is it up to date? Can I check this box and then move it
95 00:09:32.080 ⇒ 00:09:32.890 Amber Lin: out?
96 00:09:36.250 ⇒ 00:09:38.540 Amber Lin: This is up to date. Yeah.
97 00:09:38.540 ⇒ 00:09:44.240 Casie Aviles: I’ll have to check if they added a new added more. So I can update it again.
98 00:09:45.070 ⇒ 00:09:47.330 Amber Lin: Oh, so right now it’s manual, right?
99 00:09:47.910 ⇒ 00:09:54.219 Casie Aviles: Yeah, because I I created a new database. Yeah, I mean a new table to clean everything
100 00:09:54.680 ⇒ 00:09:56.269 Amber Lin: Okay, sounds good.
101 00:09:57.467 ⇒ 00:10:00.039 Amber Lin: That is in.
102 00:10:00.820 ⇒ 00:10:02.870 Amber Lin: I’ll move that that’s for me.
103 00:10:03.260 ⇒ 00:10:11.620 Amber Lin: So we’re just checking on the different dashboards.
104 00:10:13.210 ⇒ 00:10:14.490 Amber Lin: Who is?
105 00:10:17.850 ⇒ 00:10:18.770 Amber Lin: We did
106 00:10:18.770 ⇒ 00:10:24.469 Casie Aviles: One. I did not include an indication of the quality score. How it’s calculated
107 00:10:24.920 ⇒ 00:10:29.469 Amber Lin: I mean, we could either have it on the dashboard or have it on the spreadsheet.
108 00:10:31.300 ⇒ 00:10:36.629 Amber Lin: So is this still in progress? Should I just leave it there to
109 00:10:44.974 ⇒ 00:10:52.899 Casie Aviles: So there! There are those I mean this Rainforge one I don’t know who that is right now, so that’s why I left it unchecked.
110 00:10:54.149 ⇒ 00:11:06.019 Casie Aviles: For filtering the internal AI team. I didn’t really implement anything for that, but there is a filter feature for the dashboard. I could show as well again
111 00:11:06.370 ⇒ 00:11:07.472 Amber Lin: Okay, sounds good.
112 00:11:10.747 ⇒ 00:11:17.909 Amber Lin: That’s probably just adding a tag to the different people, and then filter by tag. I don’t know
113 00:11:17.910 ⇒ 00:11:18.540 Miguel de Veyra: Yeah.
114 00:11:18.540 ⇒ 00:11:27.779 Casie Aviles: There’s this filter feature here, and you could filter it according to the people. Let’s say, from the people, from ABC. So something like this
115 00:11:29.180 ⇒ 00:11:31.330 Miguel de Veyra: I have a question. Sorry, Casey.
116 00:11:32.290 ⇒ 00:11:35.089 Miguel de Veyra: cause we have to be very careful about the Rs.
117 00:11:35.430 ⇒ 00:11:41.040 Miguel de Veyra: How how long do you think this will still take to finish just on a rough estimate
118 00:11:42.410 ⇒ 00:11:46.572 Casie Aviles: I mean, it’s just, you know, it’s just cleaning up the rest of the
119 00:11:47.790 ⇒ 00:11:50.879 Casie Aviles: the ones that I haven’t done like, for example. I mean.
120 00:11:51.730 ⇒ 00:11:55.660 Casie Aviles: how important is the finding out who the brain forge
121 00:11:55.880 ⇒ 00:12:00.340 Casie Aviles: AI service about this like I I’m not sure how to go about this
122 00:12:00.510 ⇒ 00:12:03.370 Miguel de Veyra: Okay, okay, is okay.
123 00:12:03.610 ⇒ 00:12:06.760 Miguel de Veyra: Yesterday you spent, I assume, the whole 8 h. Right?
124 00:12:07.810 ⇒ 00:12:08.860 Casie Aviles: Yes.
125 00:12:08.860 ⇒ 00:12:10.190 Amber Lin: Okay, okay, okay.
126 00:12:11.033 ⇒ 00:12:18.880 Amber Lin: do you think we should just move forward with this like we can check if it’s good enough? If it’s good enough, we can save
127 00:12:18.880 ⇒ 00:12:23.079 Casie Aviles: Yeah, yeah, I want to check with him also. And
128 00:12:23.080 ⇒ 00:12:23.500 Miguel de Veyra: Yeah.
129 00:12:23.500 ⇒ 00:12:26.766 Casie Aviles: Confident with this, because you know
130 00:12:27.140 ⇒ 00:12:29.620 Miguel de Veyra: We can probably just send them a message, hey? With them. Here’s
131 00:12:29.620 ⇒ 00:12:37.839 Amber Lin: That’s good. So let’s wait from him to do until we do any more work on this, because it looks pretty good to me. I know there’s like nitty pick pick
132 00:12:37.840 ⇒ 00:12:41.470 Miguel de Veyra: Yeah details. But the execs probably won’t know this. So
133 00:12:41.470 ⇒ 00:12:54.380 Miguel de Veyra: yes, because just for reference, because I think amber in case he doesn’t know it yet, or the other member for ABC. We want to get down to around. Was it 25 or so we total spend for all members
134 00:12:56.160 ⇒ 00:13:02.730 Miguel de Veyra: Right. So I ideally, Casey, you’re gonna spend like 4 HA day on this, probably less
135 00:13:05.310 ⇒ 00:13:05.820 Amber Lin: Yeah.
136 00:13:05.820 ⇒ 00:13:11.719 Amber Lin: And we already spent quite a bit yesterday because we had an emergency
137 00:13:11.720 ⇒ 00:13:12.686 Miguel de Veyra: Yeah, yeah.
138 00:13:14.870 ⇒ 00:13:15.870 Amber Lin: Okay,
139 00:13:17.360 ⇒ 00:13:29.000 Amber Lin: yeah. The test list I will get from Denise on Friday. I can work on these and I’ll work on a presentation today. Okay, Patrick, any updates on your side?
140 00:13:31.723 ⇒ 00:13:34.690 Patrik: Yes, yeah. So I am
141 00:13:35.850 ⇒ 00:13:38.039 Patrik: trying to look at ways to make the
142 00:13:39.390 ⇒ 00:13:41.939 Patrik: Like chap up faster, really.
143 00:13:42.220 ⇒ 00:13:42.920 Amber Lin: That’s
144 00:13:43.850 ⇒ 00:13:48.379 Patrik: That’s core. Let me share my screen.
145 00:13:50.205 ⇒ 00:13:57.900 Patrik: So what I did was, I kind of rewrote our workflow yesterday in a couple different architectures.
146 00:13:59.260 ⇒ 00:14:07.579 Patrik: one following this kind of document stuffing pattern. You can see here one using
147 00:14:09.050 ⇒ 00:14:17.949 Patrik: like a similarity search to slim down the context that we’re putting into.
148 00:14:18.525 ⇒ 00:14:22.620 Patrik: What’s it called slim down the context that we’re shoving into the Llm
149 00:14:24.730 ⇒ 00:14:29.612 Patrik: Another one which actually like persists
150 00:14:30.460 ⇒ 00:14:39.589 Patrik: the embeddings locally. And then we just instead of rebuilding the embeddings, we actually just like, pull them from file.
151 00:14:41.820 ⇒ 00:14:43.950 Patrik: One of these is testing
152 00:14:44.845 ⇒ 00:14:48.940 Patrik: essentially having open AI generate the embeddings.
153 00:14:49.510 ⇒ 00:14:59.009 Patrik: Actually sorry. This one tests using like a local sentence transformer to generate the embeddings. This one is Gen is having open AI generate the embeddings.
154 00:14:59.330 ⇒ 00:15:06.020 Patrik: And then this one uses a cache embeddings. So we actually like skip, basically, it’s just like skip that step.
155 00:15:07.870 ⇒ 00:15:14.490 Patrik: and I’ve yet to implement like a full rag solution with, you know, like some sort of vector, database
156 00:15:17.050 ⇒ 00:15:23.400 Patrik: but I ran this through some benchmarking stuff so you can see the output here
157 00:15:25.280 ⇒ 00:15:35.929 Patrik: Basically, what I’ve found is on the Llm. Side. I’m pretty sure Openai is caching
158 00:15:36.320 ⇒ 00:15:39.120 Patrik: all the tokens that it generates.
159 00:15:39.290 ⇒ 00:15:42.409 Patrik: So as long as the context doesn’t change
160 00:15:43.288 ⇒ 00:15:49.912 Patrik: it’s just gonna hit the cash and we actually like don’t incur that much of a cost.
161 00:15:51.110 ⇒ 00:15:56.330 Patrik: But as soon as the what’s it called?
162 00:15:58.380 ⇒ 00:16:03.099 Patrik: As soon as this ABC, central Doc.
163 00:16:05.320 ⇒ 00:16:10.324 Patrik: this file that has all the all the information
164 00:16:11.450 ⇒ 00:16:17.240 Patrik: pest control as soon as that changes. Then basically, we will have a cache, miss, and then
165 00:16:20.720 ⇒ 00:16:22.900 Patrik: what’s it called? And then we’ll
166 00:16:23.300 ⇒ 00:16:23.930 Miguel de Veyra: I’ll have to read
167 00:16:23.930 ⇒ 00:16:25.410 Patrik: Be like super slow again
168 00:16:25.670 ⇒ 00:16:32.370 Miguel de Veyra: Yeah. But yeah, I mean, cause they change the document. Basically every day, like, everything changes
169 00:16:33.320 ⇒ 00:16:39.490 Patrik: Yeah, yeah. So I think there are a couple other like
170 00:16:39.680 ⇒ 00:16:42.329 Patrik: reasons why we would want to go with
171 00:16:42.550 ⇒ 00:16:46.750 Patrik: rag or or like a vector, solution.
172 00:16:47.569 ⇒ 00:16:48.920 Patrik: One is like.
173 00:16:49.200 ⇒ 00:16:55.999 Patrik: the document is going to continue to grow. So where does inevitably hit some sort of token limit on the Llm side?
174 00:16:56.760 ⇒ 00:16:57.750 Patrik: Yeah.
175 00:16:57.750 ⇒ 00:17:04.550 Miguel de Veyra: And then I think it will also reach a certain point where there’s nothing we can do in terms of speed, because the context is just too big
176 00:17:05.140 ⇒ 00:17:06.859 Patrik: Yes. Correct. Yeah.
177 00:17:07.700 ⇒ 00:17:11.329 Miguel de Veyra: Unless we slice it up to very smaller chunks
178 00:17:12.050 ⇒ 00:17:14.610 Patrik: Yeah, yeah, yeah, exactly.
179 00:17:15.730 ⇒ 00:17:23.629 Patrik: So that’s yeah. Basically was using this like this text splitter here to
180 00:17:24.069 ⇒ 00:17:30.707 Patrik: pull out. You know, relative chunks, and we’re shoving those into the context instead.
181 00:17:31.410 ⇒ 00:17:35.690 Patrik: so I think I’m like, I’m getting closer to a solution.
182 00:17:36.580 ⇒ 00:17:43.119 Patrik: like a better understanding of how we can work with. You know, this huge, this huge document?
183 00:17:43.430 ⇒ 00:17:50.180 Patrik: And I think the next step before we like move over to
184 00:17:50.620 ⇒ 00:17:57.190 Patrik: some sort of vector, solution is making sure that we can understand the quality
185 00:17:57.380 ⇒ 00:18:01.009 Patrik: coming out of the response.
186 00:18:01.480 ⇒ 00:18:05.039 Patrik: So that when we push it live, we know that we’re not like.
187 00:18:05.750 ⇒ 00:18:09.329 Patrik: you know, we’re not losing like 2030% on quality
188 00:18:10.440 ⇒ 00:18:14.869 Miguel de Veyra: And then I think I’m not sure if it’s possible via vector search.
189 00:18:15.040 ⇒ 00:18:22.409 Miguel de Veyra: because one of the client requests that I think eventually we’re gonna do is, for example, it gives out the certain, for example, about thermosel.
190 00:18:22.640 ⇒ 00:18:25.469 Miguel de Veyra: They want to basically include the page.
191 00:18:25.590 ⇒ 00:18:28.570 Miguel de Veyra: Like, you know, if you want to read more. Here’s the data.
192 00:18:29.250 ⇒ 00:18:31.210 Miguel de Veyra: Here’s the actual documentation
193 00:18:32.360 ⇒ 00:18:33.800 Patrik: Hmm, yes, yes.
194 00:18:33.800 ⇒ 00:18:36.430 Miguel de Veyra: Because I don’t think it’s possible via rug.
195 00:18:36.810 ⇒ 00:18:38.570 Miguel de Veyra: But context, it should be.
196 00:18:39.300 ⇒ 00:18:41.150 Miguel de Veyra: But yeah, not sure.
197 00:18:42.450 ⇒ 00:18:44.099 Patrik: Also, you can always split.
198 00:18:44.310 ⇒ 00:18:54.690 Patrik: You can always see if they ask for a document, or they want to include a document. You can grab it, split it, and then do a similarity search on that all in all in process
199 00:19:00.450 ⇒ 00:19:01.589 Patrik: that makes sense
200 00:19:02.790 ⇒ 00:19:05.050 Miguel de Veyra: No, I, yeah, a bit. But I think
201 00:19:05.320 ⇒ 00:19:18.639 Miguel de Veyra: what they specifically wanted was, it’s on the reply of the bot. So if they click that it will automatically go to Google Docs, or something like that, like the documentation. So they can read the entire thing because the bot only gives like a 2 to 3. Sentence. Summary right
202 00:19:19.470 ⇒ 00:19:26.069 Patrik: Oh, oh, once you want you’re saying you want to link out to the response should link out to the actual Google Doc.
203 00:19:26.070 ⇒ 00:19:30.390 Miguel de Veyra: Yeah. Or you know, whatever we use to display the actual documentation
204 00:19:31.270 ⇒ 00:19:33.009 Patrik: Gotcha gotcha gotcha
205 00:19:34.350 ⇒ 00:19:36.810 Miguel de Veyra: Which complicates a lot of things.
206 00:19:37.590 ⇒ 00:19:39.139 Patrik: Yeah, yeah, yeah.
207 00:19:39.848 ⇒ 00:19:42.400 Patrik: We need some sort of like, structured output
208 00:19:42.400 ⇒ 00:19:47.140 Miguel de Veyra: Yeah, but I don’t think we’re gonna be working on that anytime soon, though, because we’re
209 00:19:47.580 ⇒ 00:19:49.159 Miguel de Veyra: bringing down the arms a lot
210 00:19:53.580 ⇒ 00:19:58.400 Amber Lin: Yeah, Patrick. So what’s the what’s the response? Time right now?
211 00:20:00.220 ⇒ 00:20:02.270 Patrik: I haven’t touched anything on the
212 00:20:02.470 ⇒ 00:20:04.729 Amber Lin: Actual workflow. I see, I see.
213 00:20:04.730 ⇒ 00:20:06.210 Patrik: Is this kind of just like
214 00:20:06.510 ⇒ 00:20:09.729 Patrik: exploration, or more of like a spike? I guess
215 00:20:09.930 ⇒ 00:20:12.080 Amber Lin: Cool. So do you think
216 00:20:12.490 ⇒ 00:20:22.550 Amber Lin: we can change the response time like anytime soon? Or how long do you estimate this will take because we do want to reduce the hours we spend on this
217 00:20:31.970 ⇒ 00:20:35.410 Patrik: So you’re asking how long, how long it’s gonna take
218 00:20:35.630 ⇒ 00:20:37.450 Amber Lin: Yeah, what do you estimate?
219 00:20:39.755 ⇒ 00:20:40.360 Patrik: Question.
220 00:20:40.700 ⇒ 00:20:41.180 Amber Lin: Yeah.
221 00:20:45.050 ⇒ 00:20:48.881 Patrik: Yeah, that’s a good question. Let me look at my clockify.
222 00:20:50.470 ⇒ 00:20:58.980 Patrik: Let’s see, I spent 2 h on this yesterday. So let’s say
223 00:21:00.800 ⇒ 00:21:07.159 Patrik: another hour or 2 to get the rag running tested, and then
224 00:21:08.180 ⇒ 00:21:09.850 Patrik: Understanding, quality.
225 00:21:11.420 ⇒ 00:21:17.520 Patrik: And then maybe another like 2 to 3 h to
226 00:21:17.760 ⇒ 00:21:21.110 Patrik: actually do the NAN implementation
227 00:21:23.417 ⇒ 00:21:25.919 Miguel de Veyra: Amber. So thanks, Patrick Amber, because
228 00:21:25.920 ⇒ 00:21:26.500 Patrik: Yeah.
229 00:21:26.890 ⇒ 00:21:31.709 Miguel de Veyra: Put them, said the response. Times are like it’s between 5 to 10 seconds. Right? Casey
230 00:21:33.240 ⇒ 00:21:34.500 Casie Aviles: Oh, yeah. Yeah.
231 00:21:35.060 ⇒ 00:21:35.410 Amber Lin: Yeah.
232 00:21:35.410 ⇒ 00:21:40.450 Miguel de Veyra: Last time we were with Utam. He said, that should be fine, for now, because the the Yvette
233 00:21:40.450 ⇒ 00:21:48.930 Miguel de Veyra: originally wanted 30 Steve wanted. Was it Scott Scott wanted like 50. And then we what’s his name?
234 00:21:49.150 ⇒ 00:21:50.830 Miguel de Veyra: I forgot. There’s so many names.
235 00:21:50.990 ⇒ 00:21:56.809 Miguel de Veyra: But our partner basically said 5. So I think we’re on the we’re on where we want. Anyways.
236 00:21:57.490 ⇒ 00:22:21.680 Amber Lin: Okay. Sounds good. I think, Patrick. I know you’re like Super Super busy. And you had. It’s amazing you had time to explore all of this. Probably this would also apply to some more internal chat bots eventually. Right now, I think we see that’s all we wanted to do with this. I put their meeting pot in the wait room. But that’s all we probably wanted to.
237 00:22:22.520 ⇒ 00:22:46.450 Amber Lin: for now, because we’re over allocated. So that means we’re not the money right now. So after our AI team, me and Miguel, and we’ll talk about the internal AI backlogs, and probably there we’ll we’ll need your help on that as well. So I think we’ll tag you, or we’ll we’ll have another meeting when that comes up
238 00:22:47.860 ⇒ 00:22:55.250 Patrik: Alright. That sounds good. Yeah. I mean, I definitely wanna like, write down like, let me
239 00:22:55.370 ⇒ 00:23:10.750 Patrik: put all this information down. So there’s actually like a knowledge share for the rest of you guys. On what I’ve learned. But yeah, I mean, ideally, like, you know, we we can apply this to all the other workflows and and things like that.
240 00:23:10.750 ⇒ 00:23:19.020 Amber Lin: Yeah, sounds good. So I think in in linear, just to keep things clean, I will change your project just to update knowledge base. And that will be it
241 00:23:20.630 ⇒ 00:23:21.460 Patrik: Sounds good
242 00:23:21.800 ⇒ 00:23:26.609 Amber Lin: Thank you guys. And also just for just for the team, I think
243 00:23:27.800 ⇒ 00:23:35.290 Amber Lin: we are probably going to have a separate ABC stand up versus AI standup. So
244 00:23:35.300 ⇒ 00:23:58.530 Amber Lin: I will. I will text in the slack how that goes, because I want, if we have any internal, a discussions it to be on our separate meeting. And also probably moving forward the Friday client meetings. I think this week we still want Casey to present the dashboard, but moving forward. Probably the client meetings that you, the engineer, shouldn’t have to go
245 00:24:01.700 ⇒ 00:24:02.360 Casie Aviles: Okay.
246 00:24:03.080 ⇒ 00:24:08.729 Amber Lin: Yeah. But probably this Friday, we still need you guys to be there. Okay, that’s all for me.
247 00:24:08.870 ⇒ 00:24:16.759 Miguel de Veyra: Amber. I think the other thing is, do you? When do you want to start what we talked about yesterday about the end of day stuff, so we can send it all to Utah.
248 00:24:17.630 ⇒ 00:24:22.910 Amber Lin: Sure! I’ll slack you. We can meet just whenever
249 00:24:22.910 ⇒ 00:24:23.949 Miguel de Veyra: Okay. Okay. Sure.
250 00:24:24.180 ⇒ 00:24:25.810 Amber Lin: Okay, thanks guys.
251 00:24:25.810 ⇒ 00:24:27.250 Miguel de Veyra: Thanks. Everyone have a good day.
252 00:24:27.500 ⇒ 00:24:28.470 Amber Lin: Alright!