WEBVTT 1 00:00:06.280 --> 00:00:09.759 Hey, everyone, welcome to another episode of React Ground Up. 2 00:00:09.919 --> 00:00:12.279 I am your host today, Paige ne dream House, and 3 00:00:12.320 --> 00:00:14.279 I am joined by our panelists TJ. 4 00:00:14.439 --> 00:00:16.320 Vantol, Hey everybody. 5 00:00:15.880 --> 00:00:16.839 And Jack Harrington. 6 00:00:17.120 --> 00:00:17.719 Hey there. 7 00:00:18.079 --> 00:00:20.800 And our special guest today is Victory Nwani. 8 00:00:21.079 --> 00:00:22.199 Looking welcome. 9 00:00:22.280 --> 00:00:25.120 Victory. We're really happy to have you here with us, 10 00:00:25.199 --> 00:00:28.079 and if you'd like to introduce yourself to our audience 11 00:00:28.160 --> 00:00:30.399 and tell us a little bit about why you're famous, 12 00:00:30.440 --> 00:00:32.240 I think that would be a great place to start. 13 00:00:32.560 --> 00:00:35.920 Thank you for giving me the opportunity visically, I'm Victor Money. 14 00:00:36.079 --> 00:00:40.439 I'm based in Africa. I work remotely with an IoT 15 00:00:40.640 --> 00:00:44.960 company in Austin, Texas called Ambionik, and we viewed an 16 00:00:45.039 --> 00:00:49.240 I application for old detection. So it's used primarily in 17 00:00:49.320 --> 00:00:52.799 the healthcare line. And aside that, I'm also a technical 18 00:00:52.880 --> 00:00:57.799 author on Smashing Magazine, Love Growth, It, Digital Ocean, and 19 00:00:58.119 --> 00:01:02.439 her so I write organizations as a technical OO and 20 00:01:02.520 --> 00:01:04.920 I'm also interested in cloud in DA so when I 21 00:01:04.959 --> 00:01:08.200 love working on the front and side, working as an 22 00:01:08.280 --> 00:01:11.120 architect and the Google Cloud and Azure and yes, that's 23 00:01:11.159 --> 00:01:12.079 pretty much about it. 24 00:01:12.319 --> 00:01:15.480 Wow, that's that's plenty. I think you've got a lot 25 00:01:15.560 --> 00:01:21.719 going on, so I see that you have. Like you said, 26 00:01:21.760 --> 00:01:25.159 you're an author for a lot of different awesome publications. 27 00:01:25.200 --> 00:01:27.599 But the one that we were interested in or that 28 00:01:27.640 --> 00:01:31.560 we came across most recently was integrating a dialogue flow 29 00:01:31.599 --> 00:01:34.000 agent into a React app. And I was wondering if 30 00:01:34.040 --> 00:01:36.280 you could tell us a little bit more about that, 31 00:01:36.319 --> 00:01:39.879 and I'll link it for anybody in the podcast who 32 00:01:39.920 --> 00:01:41.400 wants to check it out later on. 33 00:01:41.879 --> 00:01:45.159 Yeah, So that was actually a two part article, so 34 00:01:45.239 --> 00:01:48.200 it focused on that log flue. I needed to use 35 00:01:48.239 --> 00:01:50.040 that log flue for a project, so I had to 36 00:01:50.079 --> 00:01:53.200 mean about it, and during the process, I also wrote 37 00:01:53.200 --> 00:01:56.040 that article on that flu. So what it doesn't it 38 00:01:56.040 --> 00:01:59.400 basically explains our dialogue flue is that's the first part 39 00:01:59.400 --> 00:02:02.519 of the au and the second one talks about integrating 40 00:02:02.640 --> 00:02:06.159 that log flu into your React application. So we use 41 00:02:06.280 --> 00:02:10.520 that stand alone dialogue flu agent into a React application. 42 00:02:10.680 --> 00:02:12.360 So it's a two path. That's a cool and are 43 00:02:12.520 --> 00:02:13.199 glad to explain? 44 00:02:13.800 --> 00:02:16.240 Yeah, maybe you can start by explaining just what dialogue 45 00:02:16.240 --> 00:02:18.520 flow is because I definitely am familiar. 46 00:02:18.919 --> 00:02:21.800 Okay, So that look flow is a savage oft on 47 00:02:21.840 --> 00:02:24.599 the Google Cloud, and it's what it does basically is 48 00:02:24.639 --> 00:02:28.639 to it makes building natural language procession applications much much 49 00:02:28.680 --> 00:02:32.080 more easier, so real and having to build the machine 50 00:02:32.159 --> 00:02:35.680 learning models, having to train everything on your own. What 51 00:02:35.759 --> 00:02:38.159 that log Flue does It gives an opportunity to do 52 00:02:38.199 --> 00:02:41.599 this while leveraging the cloud, so you could basically build 53 00:02:41.719 --> 00:02:45.319 the natural language professional application straight up from your iPad. 54 00:02:44.759 --> 00:02:47.400 It all leverages the cloud for you. 55 00:02:47.639 --> 00:02:50.039 Some curious that you come across this for like a 56 00:02:50.080 --> 00:02:52.879 work purpose or was this just I'm curious what your 57 00:02:52.960 --> 00:02:55.800 use use case was for diving into this. 58 00:02:56.319 --> 00:03:00.520 So my my was to build a food application. So 59 00:03:00.800 --> 00:03:03.800 wanted to build a chat assistant for a food application. 60 00:03:03.840 --> 00:03:06.039 So it's one of the things which allow Flue does. 61 00:03:06.120 --> 00:03:08.800 It helps you build chat assistance, so where you have 62 00:03:08.960 --> 00:03:12.960 chat boats like Cey, those series quite complex as we 63 00:03:13.000 --> 00:03:14.840 all know. But what it does is it helps you 64 00:03:14.960 --> 00:03:19.080 build chat assistants, chat boats applications that can classify texts 65 00:03:19.080 --> 00:03:20.960 and things like that. So that was what I needed 66 00:03:21.000 --> 00:03:22.840 to use it for. It was to be used for 67 00:03:23.120 --> 00:03:26.360 the chat assistant for a food software. Whether instead of 68 00:03:26.360 --> 00:03:30.280 having a human respond to chats, the chatboats would do 69 00:03:30.319 --> 00:03:31.520 that on behalf of a human. 70 00:03:31.759 --> 00:03:34.599 That's very very cool. So give me a little bit 71 00:03:34.639 --> 00:03:38.319 more about dialogue flow. Does it use like machine learning 72 00:03:38.479 --> 00:03:41.000 or some kind of artificial intelligence to drive it, or 73 00:03:41.039 --> 00:03:44.439 how does it If you say, I guess I'm looking 74 00:03:44.479 --> 00:03:47.280 for a recipe, or maybe I'm looking for a restaurant 75 00:03:47.360 --> 00:03:50.439 recommendation or something like that, how does it know what 76 00:03:50.599 --> 00:03:53.439 to respond with as a human might? 77 00:03:53.840 --> 00:03:55.759 Okay, so I'd just like to say a little bit 78 00:03:55.759 --> 00:03:58.199 of a disclaimer. One of the things that long flud 79 00:03:58.199 --> 00:04:02.680 does is it's abstract a Lady onderline complexity. So I 80 00:04:02.719 --> 00:04:06.000 cannot fully say I understand how it operates because it 81 00:04:06.039 --> 00:04:08.960 abstracts a way that complexity from you. But it does 82 00:04:09.000 --> 00:04:11.800 it gives you a service to use. But like you said, 83 00:04:12.000 --> 00:04:18.360 dialogue flu performs text classification, text extraction, intense classification, and 84 00:04:18.439 --> 00:04:21.839 special cognition. So, like you said, what it does in 85 00:04:21.879 --> 00:04:24.639 that scenario would be it would be able to classify 86 00:04:24.680 --> 00:04:27.600 your text knowing that page wants to buy something. It 87 00:04:27.639 --> 00:04:30.959 classifies your intents you want to purchase something, and it 88 00:04:31.120 --> 00:04:33.000 responds back using a trained model. 89 00:04:33.240 --> 00:04:33.720 Very cool. 90 00:04:34.120 --> 00:04:37.079 So your article is actually a really nice like end 91 00:04:37.079 --> 00:04:40.000 to end the full stack system. I mean, it's you've 92 00:04:40.040 --> 00:04:44.000 got like the training or the agent definition of the 93 00:04:44.519 --> 00:04:47.079 Google Cloud stuff upfront. Then you've got an express server, 94 00:04:47.600 --> 00:04:50.839 you've got MOBS, and you've got React up on top 95 00:04:50.879 --> 00:04:52.879 of am I summarizing that correctly. 96 00:04:53.360 --> 00:04:56.560 Yeah, so those are actually two parts split the path. 97 00:04:56.879 --> 00:05:00.680 First off is the dialogue Flue application and cond part 98 00:05:00.720 --> 00:05:03.519 is the front end application. So that log flow is 99 00:05:03.600 --> 00:05:06.360 quite standalone or do. It's offered through the Google Cloud. 100 00:05:06.399 --> 00:05:08.120 So to use that log flow you need to have 101 00:05:08.279 --> 00:05:10.720 Google Cloud project because it's one of the offerings of 102 00:05:10.759 --> 00:05:13.319 the Google Cloud. But there are several ways you could 103 00:05:13.399 --> 00:05:15.959 use that log flu Basically, you could also use it 104 00:05:16.000 --> 00:05:19.360 through the Google Assistance, so you could deploy Google that 105 00:05:19.439 --> 00:05:21.959 log for application to the Google assystem. We all know 106 00:05:22.240 --> 00:05:25.279 Series for io s. Why the Google Assistance is for 107 00:05:25.360 --> 00:05:27.720 Android devices, so you do not need to have like 108 00:05:27.879 --> 00:05:30.600 a front end application to use that log flu. You 109 00:05:30.600 --> 00:05:33.560 could deploy to Google Assistant platform and use it Vibe 110 00:05:33.560 --> 00:05:34.360 and a smart works. 111 00:05:34.560 --> 00:05:36.959 Oh nice, so we have like the smart speaker or 112 00:05:37.279 --> 00:05:39.560 an Android phone that's got the assistant. You can actually 113 00:05:39.600 --> 00:05:43.279 integrate with that too, exactly. Yeah, nice, very cool. 114 00:05:43.639 --> 00:05:43.800 Yeah. 115 00:05:43.879 --> 00:05:45.839 No, it's a nice full stack system there. So if 116 00:05:45.839 --> 00:05:49.160 somebody wants to learn essentially what go from fro an 117 00:05:49.240 --> 00:05:51.160 end to full stack. This is actually a really nice 118 00:05:51.160 --> 00:05:52.839 demo application to walk through. 119 00:05:53.120 --> 00:05:55.560 Yeah, exactly, And I wrote that in the second part 120 00:05:55.639 --> 00:05:56.240 of the article. 121 00:05:56.639 --> 00:05:59.480 Yeah, so I'm just curious, why did you choose my backs. 122 00:06:00.040 --> 00:06:02.920 It's like a controversial choice in the React world. 123 00:06:03.800 --> 00:06:09.720 Mobex is just quite like a state management package for reacts. 124 00:06:10.199 --> 00:06:14.399 It's quite simpler than REDUCS. Like that argument is kind 125 00:06:14.399 --> 00:06:20.519 of yes, that's kind of Mobis brings in the object 126 00:06:20.519 --> 00:06:25.720 oriented programming into step managements, while REDUCS is more functional programming. 127 00:06:25.879 --> 00:06:28.360 That's about the best synopsis I think I've ever heard 128 00:06:28.600 --> 00:06:31.240 about those two. I can never hone it down to 129 00:06:31.279 --> 00:06:31.800 that level. 130 00:06:32.319 --> 00:06:38.279 Yeah, I've never actually used Mobex or flux or any 131 00:06:38.360 --> 00:06:40.879 of the other ones. I've only used REDUCTS. Although I've 132 00:06:40.920 --> 00:06:44.240 been hearing pretty good things about the new Redux toolkit, 133 00:06:44.319 --> 00:06:47.040 which I think is kind of REDUCS two point oh, 134 00:06:47.639 --> 00:06:50.759 that it's abstracting away a lot of the boiler plate 135 00:06:50.800 --> 00:06:54.199 and the things that people didn't particularly like about REDUCS. 136 00:06:54.199 --> 00:06:57.639 But I mean that's cool. Mobex is definitely still a 137 00:06:57.680 --> 00:07:00.279 good a good option. And I think there's an other 138 00:07:00.319 --> 00:07:03.560 one that Facebook came out with, and now I'm trying 139 00:07:03.560 --> 00:07:04.680 to think of the name of it. 140 00:07:05.319 --> 00:07:07.120 I think that's the context API. 141 00:07:08.360 --> 00:07:09.000 Context is. 142 00:07:09.079 --> 00:07:12.199 I mean, that's definitely been a game changer. But there 143 00:07:12.319 --> 00:07:14.759 was a new a newer one, and now I can't 144 00:07:14.759 --> 00:07:16.279 think the name of it. I'll have to look it up. 145 00:07:16.319 --> 00:07:18.519 That it seems like was maybe it was Recoil. 146 00:07:18.879 --> 00:07:22.680 Oh, yeah that was yeah, yeah, I think recoil is 147 00:07:22.759 --> 00:07:25.399 quite new. Or do I read the ad crew the 148 00:07:25.439 --> 00:07:27.920 while back? So I didn't think there was a context 149 00:07:27.959 --> 00:07:31.600 API when I read that, I've used the context API. 150 00:07:32.279 --> 00:07:34.839 Well, context API has been undocumented for a long time, 151 00:07:34.879 --> 00:07:37.600 and then it got formal. Finally, I think I can 152 00:07:37.600 --> 00:07:39.920 read it like sixteen, I think it got formal, and 153 00:07:39.959 --> 00:07:42.279 then then reactors came in and got. 154 00:07:42.240 --> 00:07:46.879 Yeah, that's I think really revolutionized using the context APIs 155 00:07:46.959 --> 00:07:52.000 as a real solution. So Nwani our victory, what what 156 00:07:52.680 --> 00:07:56.040 kind of spurred you to learn about dialogue flow Because 157 00:07:56.040 --> 00:07:58.800 this is a it's a very cool API. But with 158 00:07:58.839 --> 00:08:02.199 all these cloud pro with Amazon, with Google, with Azure, 159 00:08:02.360 --> 00:08:07.639 even there's so much that's hidden in all the services 160 00:08:07.639 --> 00:08:10.360 that they offer. So how did you, I guess, find 161 00:08:10.480 --> 00:08:13.160 this and decide this is what I wanted to test 162 00:08:13.199 --> 00:08:15.079 out and see how it works. 163 00:08:15.360 --> 00:08:19.600 Basically, when I would when I needed this service. Dialogue 164 00:08:19.600 --> 00:08:24.040 Flu was kind of the most popular service. The other 165 00:08:24.079 --> 00:08:26.920 services apart from Dialogue Flue like the watson as system 166 00:08:27.040 --> 00:08:31.279 from IDM. What didn't Google Dialogue Flu was the simplicity 167 00:08:31.560 --> 00:08:34.039 and how easy it was to use, so I didn't 168 00:08:34.120 --> 00:08:37.120 have to lend the basic things because I'm not like 169 00:08:37.240 --> 00:08:40.360 a machine learning expot I'm more of a front end engineer, 170 00:08:40.480 --> 00:08:42.679 so it wasn't so difficult for me to pick up 171 00:08:42.799 --> 00:08:45.000 and that was what made me use that look flasically. 172 00:08:45.559 --> 00:08:48.679 Was the documentation for it pretty decent as well, because 173 00:08:48.720 --> 00:08:52.080 that's always my thing is I need to see working 174 00:08:52.159 --> 00:08:54.519 code examples and then I can figure out how to 175 00:08:54.519 --> 00:08:57.799 put it into my own applications for my own code basis. 176 00:08:58.039 --> 00:09:00.960 Yeah, pretty much pretty much you know of back Google. 177 00:09:01.080 --> 00:09:05.080 Everything from Google is quite comprehensive in terms of the documentation. 178 00:09:05.600 --> 00:09:08.480 It's also important to note that I think Google both 179 00:09:08.559 --> 00:09:12.080 dialog Flu. They're not the original creators. I'm trying such 180 00:09:12.120 --> 00:09:14.679 of the name, but it was originally called AI. That's 181 00:09:14.679 --> 00:09:16.879 something I can't remember the name now, which was both 182 00:09:16.919 --> 00:09:17.759 about by Google. 183 00:09:18.399 --> 00:09:21.240 So I'm curious and this is a little bit of 184 00:09:21.279 --> 00:09:24.120 a broader question, and I'm not sure a nice way 185 00:09:24.159 --> 00:09:26.480 to phrase this, but I'm curious if dialogue helps you 186 00:09:26.480 --> 00:09:30.360 build chatbots that are actually good. And I think what 187 00:09:30.399 --> 00:09:32.600 I'm the broader thing to me is like I feel 188 00:09:32.600 --> 00:09:37.159 like chatbots are one of the more underutilize things in 189 00:09:37.279 --> 00:09:40.759 software development because I feel like there's an absolute enormous 190 00:09:40.759 --> 00:09:43.559 amount of potential there, but every time I personally go 191 00:09:43.679 --> 00:09:46.919 to use one, I hate it. I've yet to use one, 192 00:09:47.240 --> 00:09:49.879 and I'm seeing on the call here page and Jack 193 00:09:49.919 --> 00:09:50.360 shaking their. 194 00:09:50.320 --> 00:09:56.440 Heads as well, like it's I avoid them every Yeah. 195 00:09:55.919 --> 00:09:58.039 But it's so frustrating because they should be good, right 196 00:09:58.120 --> 00:10:00.320 like this is twenty one, we should like you should 197 00:10:00.440 --> 00:10:04.279 have like the technology behind it. So I'm curious victory, like, 198 00:10:04.679 --> 00:10:07.360 have you had like positive experiences? Do you get like 199 00:10:07.559 --> 00:10:11.039 analytics from dialogue flow that tells you like how well 200 00:10:11.080 --> 00:10:13.279 this is performing? And I'm just curious your thoughts on 201 00:10:13.799 --> 00:10:14.399 that topic. 202 00:10:15.399 --> 00:10:18.159 I would pretty much say yes, dialogue flow is quite 203 00:10:18.200 --> 00:10:21.840 good because, like you mentioned monitoring, we have the Dialogue 204 00:10:21.840 --> 00:10:25.080 Fluw console which you can use to monitor statistics about 205 00:10:25.159 --> 00:10:29.240 your chatbots. So delog Flow is basically offered in two editions. 206 00:10:29.720 --> 00:10:33.639 We have the ES edition and the CX edition. The 207 00:10:33.840 --> 00:10:37.679 EES is for quite small use cases like small startups, 208 00:10:37.960 --> 00:10:40.840 while the ES is for larger use cases that are 209 00:10:40.840 --> 00:10:44.720 being used by larger startups and companies. So one of 210 00:10:44.759 --> 00:10:47.399 the things I really love about dialog FLUW is the 211 00:10:47.480 --> 00:10:50.639 console itself. You can do basically a lot of things 212 00:10:50.639 --> 00:10:54.000 with the console. You can export a dialogue flu application. 213 00:10:54.480 --> 00:10:58.080 You can also import a dialogue application. So the pretty 214 00:10:58.120 --> 00:11:00.000 much a lot of things you can do with dialogue 215 00:11:00.039 --> 00:11:03.720 flu and it's quite difficult to And so when you 216 00:11:03.799 --> 00:11:06.799 said there are love teams that are frustrated about using 217 00:11:06.879 --> 00:11:09.960 childboats because I don't really know those pin points which 218 00:11:09.960 --> 00:11:12.039 you have using childboat providers. 219 00:11:12.240 --> 00:11:14.840 Yeah, well I'm more curious, like how do you know? 220 00:11:15.039 --> 00:11:18.840 Like so, like, hypothetically I'm building a food application, right 221 00:11:18.879 --> 00:11:21.879 and I build a bunch of different chat conversations and 222 00:11:21.960 --> 00:11:25.679 dialogue flow. What would give me confidence that my users 223 00:11:25.679 --> 00:11:29.600 are actually using this thing and succeeding with it. Because 224 00:11:29.600 --> 00:11:32.639 I think a lot of people debating this, right, Like, 225 00:11:32.639 --> 00:11:36.039 like I said, there are tons of potential applications for this. 226 00:11:36.919 --> 00:11:39.679 Do you want to know that you aren't frustrating your 227 00:11:39.759 --> 00:11:42.919 users or your your customers that they're actually using this 228 00:11:43.000 --> 00:11:45.519 thing successfully? So I'm just curious like what sort of 229 00:11:45.639 --> 00:11:49.480 signals dialogue flow gives back that people are like successfully 230 00:11:49.480 --> 00:11:51.200 going through these things that you're creating. 231 00:11:51.480 --> 00:11:54.240 So dialogue flu at the basic level, it gives you 232 00:11:54.600 --> 00:11:58.639 some level of observability and monitoring for you to know 233 00:11:58.720 --> 00:12:00.960 that these are the parts of your caation being used 234 00:12:01.000 --> 00:12:04.919 all through the console. So after deploying application, you can 235 00:12:05.240 --> 00:12:09.080 run test yourself or you could give an interval to 236 00:12:09.240 --> 00:12:11.799 know to see the statistics coming from an application. 237 00:12:12.200 --> 00:12:14.960 So okay, walk me through this flow a little bit 238 00:12:15.000 --> 00:12:15.759 with this food app. 239 00:12:15.799 --> 00:12:20.200 So you're you're trying to sell wine essentially, what how 240 00:12:20.240 --> 00:12:22.759 does how does the agent know. 241 00:12:22.639 --> 00:12:25.799 What wines are available? Like or is that part of 242 00:12:25.840 --> 00:12:26.360 the service. 243 00:12:26.360 --> 00:12:28.120 Can I say, like give me a good pino or 244 00:12:28.159 --> 00:12:30.639 something like that, Like how does that actually? Like walk 245 00:12:30.679 --> 00:12:33.279 me through the whole data flow here just at a 246 00:12:33.320 --> 00:12:33.799 high level. 247 00:12:34.039 --> 00:12:37.120 So at a high level, that loog flow is broken 248 00:12:37.159 --> 00:12:41.159 down into entities, and like you mentioned, the first thing 249 00:12:41.200 --> 00:12:44.080 you would have to do would be to create intents. 250 00:12:44.240 --> 00:12:47.039 So your first intents would be I want to buy wine. 251 00:12:47.480 --> 00:12:50.120 The block flow agents is going to be able to 252 00:12:50.360 --> 00:12:54.720 classify do some extraction from that text to know that 253 00:12:54.879 --> 00:12:58.120 this individual wants to opportunity whine, and there's going to 254 00:12:58.200 --> 00:13:02.039 be a key word. There's that keyword falls under a context, 255 00:13:02.120 --> 00:13:04.360 so it knows that this is the name of the wine. 256 00:13:04.480 --> 00:13:07.559 Funny enough, I did demo applications within the article is 257 00:13:07.559 --> 00:13:10.399 on a wine application. So if you put in the 258 00:13:10.440 --> 00:13:14.279 name of the wine, like for example, the Malotte Red Wine, 259 00:13:14.399 --> 00:13:17.639 it has the ability to extract that name. So while 260 00:13:17.679 --> 00:13:22.159 training the chatbots, you're providing a very long list of names. 261 00:13:22.360 --> 00:13:25.799 Then whenever the user and has that name, it has 262 00:13:25.840 --> 00:13:28.039 the ability to extract the name. I know that this 263 00:13:28.120 --> 00:13:30.840 is the wine exactly the user wants to buy. Then 264 00:13:30.879 --> 00:13:33.960 it can respond back using a restaurant that you have provided. 265 00:13:34.240 --> 00:13:34.600 I see. 266 00:13:34.639 --> 00:13:37.240 So as part of the message that you get about 267 00:13:37.279 --> 00:13:40.200 the intent like, you get essentially a broken down thing 268 00:13:40.200 --> 00:13:42.679 that says, here's an event is a customer wants to 269 00:13:42.679 --> 00:13:45.399 buy a wine event and here's the type of wine 270 00:13:45.399 --> 00:13:47.240 that they want to buy, and then you can go 271 00:13:47.320 --> 00:13:48.000 and respond to. 272 00:13:48.000 --> 00:13:50.240 That exactly cool, And they're also. 273 00:13:50.080 --> 00:13:52.879 Quite sounded good there. I think we lost. 274 00:13:52.799 --> 00:13:55.679 Okay, okay, I think yeah, type broke up. So those 275 00:13:55.720 --> 00:13:58.080 things are on a basic level, we can know them 276 00:13:58.080 --> 00:14:01.159 as keywords, but on the dialogue flow they're known as entities, 277 00:14:01.279 --> 00:14:05.200 so that lock Flue attracted those entities from a single 278 00:14:05.279 --> 00:14:08.960 user statement. And these entities could include price, So you 279 00:14:09.000 --> 00:14:10.879 could say I want to buy a red wine for 280 00:14:11.240 --> 00:14:14.000 five dollars. It's going to extract a five dollars from 281 00:14:14.320 --> 00:14:18.440 from that text and respond back its own response. The 282 00:14:18.639 --> 00:14:20.840 entity could also be a link. Well, most of the 283 00:14:20.919 --> 00:14:24.399 times you provide those entities while you're training the agent, 284 00:14:24.519 --> 00:14:27.799 So there's a whole training process that goes on before 285 00:14:27.879 --> 00:14:30.960 deploying your agents for use, and that's where you define 286 00:14:31.000 --> 00:14:31.759 most of these things. 287 00:14:31.879 --> 00:14:35.440 To keep going with this example, so you create intents 288 00:14:35.559 --> 00:14:38.799 it says, okay, the person wants to buy wine for 289 00:14:39.159 --> 00:14:42.519 five dollars or whatever the case may be. Presumably then 290 00:14:42.559 --> 00:14:44.320 you have a chance to wire that up to some 291 00:14:44.360 --> 00:14:47.559 sort of endpoint because like you might have like a 292 00:14:47.679 --> 00:14:50.639 service that returns here are the wines that we have 293 00:14:51.080 --> 00:14:53.679 for five dollars, So is there like some way to 294 00:14:53.720 --> 00:14:55.279 wire it up to a back end for that sort 295 00:14:55.279 --> 00:14:56.519 of thing exactly? 296 00:14:56.559 --> 00:14:58.679 And that falls on that for three minutes. So like 297 00:14:58.759 --> 00:15:01.279 I said, that luck Flu is broken down into several 298 00:15:01.440 --> 00:15:04.399 entities the same way within the objects in the programming 299 00:15:04.440 --> 00:15:07.039 could know them as objects. So that's on that the 300 00:15:07.080 --> 00:15:10.840 fulfillment entity. So the fulfillment allows you to call the webook, 301 00:15:10.960 --> 00:15:14.039 so you could have like a custom database of your 302 00:15:14.039 --> 00:15:17.240 own weby. You have like a cloud function, so when 303 00:15:17.279 --> 00:15:20.679 that intent is being recognized, it caused your cloud function, 304 00:15:21.000 --> 00:15:23.879 which achieves it from a database. So you say I 305 00:15:23.919 --> 00:15:26.559 want a wine of five dollars, it sends five dollars 306 00:15:26.639 --> 00:15:29.039 to the database and it retrieves a wine for that 307 00:15:29.159 --> 00:15:32.559 exact amount. So fulfillment allows you to give a better 308 00:15:32.679 --> 00:15:35.960 dynamic response to an intent rather than just some have 309 00:15:36.120 --> 00:15:36.919 cood add value. 310 00:15:37.159 --> 00:15:40.799 Interesting, So you were saying that you need to train 311 00:15:40.879 --> 00:15:43.720 the models before you can really kind of let the 312 00:15:43.799 --> 00:15:47.120 chat loosen to the world. How long How long did 313 00:15:47.120 --> 00:15:49.759 it take you to kind of train them up to 314 00:15:49.840 --> 00:15:53.120 get them good enough that they could interact with customers. 315 00:15:53.240 --> 00:15:57.039 It took about two weeks. But this varies basically based 316 00:15:57.039 --> 00:15:59.559 on your use case. Because you could have like a 317 00:15:59.639 --> 00:16:03.480 module larger application, it would definitely take you mucho long. 318 00:16:03.480 --> 00:16:05.840 Got time to trink Well, I'd like to clarify that 319 00:16:06.080 --> 00:16:09.519 on like where you're building a traditional machine learning application, 320 00:16:09.799 --> 00:16:13.480 it takes a few seconds to trim chatboats. So basically 321 00:16:13.600 --> 00:16:16.480 it's at it's almost instant at the click of a button. 322 00:16:16.559 --> 00:16:19.039 It takes a few seconds when you click the trim 323 00:16:19.080 --> 00:16:21.720 button and all goes up to what I'm trying to 324 00:16:21.759 --> 00:16:24.320 say is it doesn't take so much where you have 325 00:16:24.440 --> 00:16:27.120 you probably have to sit and list for the entire 326 00:16:27.120 --> 00:16:29.399 application to get tramed, then you have to come back. 327 00:16:29.679 --> 00:16:32.200 It doesn't take that process with that log flu, it 328 00:16:32.320 --> 00:16:33.480 all leverages the cloud. 329 00:16:33.799 --> 00:16:37.039 So then I'm curious, then, how what's the step that 330 00:16:37.200 --> 00:16:41.120 gets you this into a React application. So presumably there's 331 00:16:41.440 --> 00:16:45.399 you in your own app have to capture inputs text 332 00:16:45.519 --> 00:16:49.320 or maybe even like audio voice audio. I'm not sit 333 00:16:49.480 --> 00:16:52.000 so I'm curious about that. But then also like this 334 00:16:52.080 --> 00:16:55.600 dialogue flow give you like some API, some like javascripting 335 00:16:55.679 --> 00:16:58.480 you have to like MPM install that to send requests to. 336 00:16:58.879 --> 00:17:03.120 Yeah, so two major means to use dialogue Flu. There's 337 00:17:03.159 --> 00:17:06.640 a rest API which you could make yourself for There's 338 00:17:06.680 --> 00:17:09.599 a rest API with an end point for dialogue Flu, 339 00:17:10.119 --> 00:17:12.960 but you have to authenticate through the API. There's also 340 00:17:13.000 --> 00:17:15.880 an MPM package you can which you can install. But 341 00:17:16.079 --> 00:17:19.920 there was like a drawback to the MPM package, which 342 00:17:20.000 --> 00:17:24.079 is it uses gr PC within jrpcy is not yet 343 00:17:24.119 --> 00:17:26.680 fully compliance with the browser, so I had to build 344 00:17:26.839 --> 00:17:29.799 a back end for the entire for the dialog Flul application, 345 00:17:30.079 --> 00:17:33.000 so you can only you can use gepcy within the 346 00:17:33.039 --> 00:17:34.839 new the environments, but if you want to use it 347 00:17:34.880 --> 00:17:37.759 within the browser, you'd have to play around your web pack, 348 00:17:37.920 --> 00:17:39.799 which I wasn't willing to do, so I had to 349 00:17:39.880 --> 00:17:41.720 use it from an experts application. 350 00:17:42.000 --> 00:17:45.680 So, building on TJ's question from before about how to 351 00:17:46.200 --> 00:17:48.839 check if this is actually doing the right things the customers, 352 00:17:48.839 --> 00:17:51.359 I mean, there's obviously sales data in this case, but 353 00:17:51.960 --> 00:17:54.279 how would you imagine testing this from like a unit 354 00:17:54.319 --> 00:17:57.400 test perspective or just a is there a way where oh, 355 00:17:57.440 --> 00:18:00.359 I've uploaded I've uploaded a new agent, I mean, run 356 00:18:00.359 --> 00:18:01.960 a set of tests where it's like, let me just 357 00:18:02.000 --> 00:18:04.200 make sure that it's recommending the right one to me. 358 00:18:04.279 --> 00:18:06.319 Again, how do you manage that? 359 00:18:06.640 --> 00:18:10.160 I didn't actually make more such on that, because I 360 00:18:10.200 --> 00:18:12.359 don't really think there's a way you could unit test this, 361 00:18:12.880 --> 00:18:15.880 because it's it's more like I said, there's a lot 362 00:18:15.880 --> 00:18:19.720 of abstraction from the underlying implementation that looks through. What 363 00:18:19.759 --> 00:18:22.279 it gives you is basically the comes well. One of 364 00:18:22.319 --> 00:18:24.720 the things which you could write tests for are your 365 00:18:24.759 --> 00:18:26.519 integrations basically, Oh. 366 00:18:26.480 --> 00:18:27.319 Yeah, that makes sense. 367 00:18:27.440 --> 00:18:29.880 Yeah, if you're doing like a full integration test, I 368 00:18:29.880 --> 00:18:32.960 suppose you could use their endpoints. Let's say if I 369 00:18:33.000 --> 00:18:37.039 send in this, then I expect this to come back, 370 00:18:37.200 --> 00:18:38.119 which I think would work. 371 00:18:38.599 --> 00:18:42.480 Yeah, but that would mean you'd running live a FI tests. 372 00:18:42.640 --> 00:18:46.319 But wanting to notice the API requires some form of authentication, 373 00:18:46.680 --> 00:18:48.319 which is a Google account. 374 00:18:48.240 --> 00:18:52.319 Oh for for users, or well for like one like 375 00:18:52.880 --> 00:18:55.160 your like business or your personal cab for running it. 376 00:18:55.680 --> 00:18:58.400 Yeah, you would still need to use an authentication method. 377 00:18:58.599 --> 00:19:00.880 Yeah, I mean that could be that could be doable. 378 00:19:01.440 --> 00:19:03.400 So one thing that I was curious about, and this 379 00:19:03.480 --> 00:19:05.960 is this is kind of a good segue into it, 380 00:19:06.000 --> 00:19:10.519 is how is the pricing for something like this using 381 00:19:10.559 --> 00:19:11.160 this service? 382 00:19:11.240 --> 00:19:14.119 I guess yeah, so the present for dialog flu is 383 00:19:14.200 --> 00:19:17.200 quite low, and you can also use the pricing calculator 384 00:19:17.240 --> 00:19:20.200 on the Google Cloud. But I think for the for 385 00:19:20.319 --> 00:19:23.960 the ES version, which is majorly for small start ups 386 00:19:24.000 --> 00:19:27.880 and small use cases, there's a high amount of CAUs 387 00:19:28.160 --> 00:19:30.880 so each call you make is kind of methid. So 388 00:19:30.960 --> 00:19:33.880 there's an allocation of calls which you are given for 389 00:19:33.920 --> 00:19:36.640 free to use. I'm trying to pop up the present calculator. 390 00:19:37.000 --> 00:19:39.599 The present calculator is basically used for all services and 391 00:19:39.759 --> 00:19:42.279 Google Cloud, so you could just check up the price, 392 00:19:42.440 --> 00:19:44.440 how long it's going to take you to run dialogue 393 00:19:44.440 --> 00:19:45.599 flu for probably. 394 00:19:45.440 --> 00:19:48.279 More, but that's good to know. So if you were 395 00:19:48.640 --> 00:19:51.079 like kind of like you or like one of us 396 00:19:51.119 --> 00:19:53.079 getting started with it and you just wanted to try 397 00:19:53.079 --> 00:19:55.680 it out, see how it works, tinker with it, there 398 00:19:55.839 --> 00:20:00.319 is either practically no cost or a free out that 399 00:20:00.359 --> 00:20:03.920 you can use just to get started and figure out 400 00:20:03.960 --> 00:20:04.759 if it's right for you. 401 00:20:05.240 --> 00:20:08.240