1 00:00:03,540 --> 00:00:13,110 In this lesson, we are going to preview a professional LMS application that we will see in detail later. 2 00:00:14,190 --> 00:00:20,160 So the llama index team has open source the project. 3 00:00:20,160 --> 00:00:21,990 Seek insights. 4 00:00:22,500 --> 00:00:30,990 Right now this is one of the most advanced and sophisticated production ready LM apps available. 5 00:00:31,920 --> 00:00:36,600 It will be worthy to study it in detail. 6 00:00:36,930 --> 00:00:39,930 We will see the code in detail. 7 00:00:39,930 --> 00:00:50,760 We will see how to build it in bit in detail, and we will see that this application is a chat application. 8 00:00:50,760 --> 00:00:53,070 We will see now how it works. 9 00:00:53,100 --> 00:01:03,600 It uses a technique that we are going to master, which is the rack technique and a it answers questions 10 00:01:03,600 --> 00:01:04,560 about. 11 00:01:05,820 --> 00:01:14,790 Sick 10-K and 10-q documents, so it answers questions about a financial documents. 12 00:01:15,730 --> 00:01:17,620 It is production ready. 13 00:01:18,670 --> 00:01:22,240 It is using a full stack a. 14 00:01:23,650 --> 00:01:28,780 Quote a it is ready for you to fork and use. 15 00:01:29,510 --> 00:01:31,790 And all the setup. 16 00:01:31,790 --> 00:01:38,210 All the setup is open source and it is easy to deploy on vercel and render. 17 00:01:38,210 --> 00:01:38,960 Com. 18 00:01:39,910 --> 00:01:51,340 So you will see that a this application is a QA chat grounded in source of truth. 19 00:01:51,340 --> 00:01:53,290 Seek documents. 20 00:01:53,290 --> 00:01:55,930 It has a PDF viewer. 21 00:01:56,820 --> 00:02:01,830 It has a token level streaming of chat responses. 22 00:02:02,250 --> 00:02:09,060 It will stream of reasoning steps like subquestions. 23 00:02:09,180 --> 00:02:21,330 It has citation of source data and it makes use of API based tools in addition to semantic search. 24 00:02:21,840 --> 00:02:27,300 The architecture of this application that we will know in detail. 25 00:02:28,460 --> 00:02:34,070 Uses a render Dccom as backend server. 26 00:02:34,070 --> 00:02:37,790 Fast API as backend framework. 27 00:02:39,190 --> 00:02:42,430 Postgres as vector database. 28 00:02:43,350 --> 00:02:56,700 Next.js as frontend framework, Vercel as frontend server, and then it uses several external APIs and 29 00:02:56,700 --> 00:03:03,660 it stores a private documents in Amazon Amazon S3. 30 00:03:05,970 --> 00:03:12,120 So we will know the details of this application later. 31 00:03:12,300 --> 00:03:21,000 Right now I only want to show you a quick demo of the application you can use. 32 00:03:21,000 --> 00:03:27,780 You can play around with this application in Insights.ai, but. 33 00:03:28,740 --> 00:03:31,770 A very quickly. 34 00:03:31,770 --> 00:03:40,770 This is an application that we can use to make questions about different, uh, financial documents. 35 00:03:40,770 --> 00:03:51,900 For example, we can select documents from Apple and documents from Amazon from for the same year. 36 00:03:52,850 --> 00:03:59,300 And after we, we can add until, uh, eight, uh, documents. 37 00:03:59,420 --> 00:04:01,070 Let's try with these two. 38 00:04:01,100 --> 00:04:09,440 So these two are financial documents for Amazon company and for Apple company in the 2020 a year. 39 00:04:09,440 --> 00:04:13,880 And this is a kind of a financial type of document. 40 00:04:13,880 --> 00:04:22,130 So what we are doing is we are providing the LM application with private documents. 41 00:04:22,130 --> 00:04:23,660 So we try to do this. 42 00:04:23,660 --> 00:04:25,190 For example with ChatGPT. 43 00:04:25,310 --> 00:04:31,700 We cannot do it because ChatGPT ChatGPT doesn't have this knowledge, this private data. 44 00:04:32,180 --> 00:04:40,430 But in our LM application in this and this application is built on top of ChatGPT, we can use private 45 00:04:40,430 --> 00:04:41,750 documents like this one. 46 00:04:41,750 --> 00:04:48,980 So once we have loaded our documents and these documents come come from an external API. 47 00:04:49,850 --> 00:04:53,780 We can start having a conversation about these documents. 48 00:04:53,780 --> 00:05:01,610 And as you see in the following screen, here we have the the PDFs we are using. 49 00:05:01,610 --> 00:05:09,020 And in the left side of the screen we have a chat bot where we can make questions about these documents 50 00:05:09,020 --> 00:05:09,830 we have loaded. 51 00:05:09,830 --> 00:05:12,740 For example, we can say um. 52 00:05:16,670 --> 00:05:17,720 Compare. 53 00:05:19,080 --> 00:05:21,810 Both documents. 54 00:05:23,220 --> 00:05:24,000 Um. 55 00:05:24,950 --> 00:05:25,970 Make. 56 00:05:28,170 --> 00:05:30,540 Make recommend. 57 00:05:34,430 --> 00:05:36,590 Make investment recommendations. 58 00:05:39,890 --> 00:05:41,330 Commendations. 59 00:05:42,140 --> 00:05:43,250 About. 60 00:05:45,520 --> 00:05:46,540 Them. 61 00:05:47,090 --> 00:05:48,500 Citing. 62 00:05:49,070 --> 00:05:50,450 Sources. 63 00:05:51,910 --> 00:06:03,520 Okay, so if I click enter, you will see that this application is going to start preparing the response. 64 00:06:03,520 --> 00:06:08,950 And while it is preparing the response is going to allow to see. 65 00:06:09,010 --> 00:06:12,760 It's going to allow us to see how it is progressing. 66 00:06:13,980 --> 00:06:23,580 So right now, as you see it, is reviewing, uh, documents in order to prepare the response. 67 00:06:23,580 --> 00:06:33,120 And every time it is reviewing documents, it is going to show us the link to the source. 68 00:06:33,120 --> 00:06:36,150 It is used to prepare the response. 69 00:06:36,150 --> 00:06:38,700 So for us right now, it is not important. 70 00:06:38,700 --> 00:06:44,370 The response we have here probably is a good and very interesting response. 71 00:06:44,370 --> 00:06:52,440 But what I wanted to show you at this point is the way a professional LLM application works. 72 00:06:52,440 --> 00:07:02,490 So far from the typical toy demo you can find in many courses with or without a toy user interface, 73 00:07:02,490 --> 00:07:04,710 this is a real application. 74 00:07:04,710 --> 00:07:09,720 This is a professional application ready to ready for you to start using it. 75 00:07:09,720 --> 00:07:14,100 And as you see, we have front end elements. 76 00:07:14,100 --> 00:07:17,040 We have back end elements, we have. 77 00:07:17,040 --> 00:07:26,490 This is a way to be able to confirm that the response that is provided by the application is an accurate 78 00:07:26,490 --> 00:07:27,060 response. 79 00:07:27,060 --> 00:07:34,830 Do you remember that we said that sometimes applications can, uh, make hallucinations, fake responses. 80 00:07:34,830 --> 00:07:44,100 So it is very important for you to have a way to check if the response is based on, uh, true data. 81 00:07:44,100 --> 00:07:44,460 Right. 82 00:07:44,460 --> 00:07:46,920 So this is what this application is doing. 83 00:07:46,920 --> 00:07:54,060 And you can see, you know, the front end elements you have here because they are actionable items 84 00:07:54,060 --> 00:07:55,020 as well here. 85 00:07:55,020 --> 00:08:03,270 But it is very interesting because if you click here, you can go exactly to the uh, paragraph or the 86 00:08:03,270 --> 00:08:06,330 page that it is using in any case. 87 00:08:06,330 --> 00:08:06,840 Right. 88 00:08:06,840 --> 00:08:15,600 So I just wanted to show you very quickly what a professional LMS application, uh, looks like. 89 00:08:15,600 --> 00:08:16,140 Right. 90 00:08:16,140 --> 00:08:22,500 So you will see that this is what we are going to learn to do. 91 00:08:23,220 --> 00:08:25,980 This is very far from the toy demos. 92 00:08:25,980 --> 00:08:30,270 This is very far from the level one application and level two applications. 93 00:08:30,270 --> 00:08:36,419 This is what we call a level three application, which is a professional application ready to use in 94 00:08:36,419 --> 00:08:37,350 the real world. 95 00:08:37,380 --> 00:08:37,890 Okay. 96 00:08:37,890 --> 00:08:44,700 So we are going to study this application further in the next lessons and uh, other applications as 97 00:08:44,700 --> 00:08:45,210 well.