1 00:00:04,730 --> 00:00:05,090 Okay. 2 00:00:05,090 --> 00:00:10,130 So this is a very interesting level two application. 3 00:00:10,610 --> 00:00:12,170 We are going to build. 4 00:00:12,170 --> 00:00:20,150 You are going to learn how to build an application to evaluate a drug application. 5 00:00:20,300 --> 00:00:27,650 If you remember we have talked about we have we have talked about that before a LM applications are 6 00:00:27,650 --> 00:00:32,540 not very easy to evaluate, not very easy to test. 7 00:00:32,540 --> 00:00:38,480 Test is the word we use as software engineers for traditional software applications. 8 00:00:38,480 --> 00:00:47,390 So when you build a traditional software application, testing is is is relatively easy because every 9 00:00:47,390 --> 00:00:55,520 time you ask one question to a traditional software application, you are going to have the same answer. 10 00:00:55,520 --> 00:01:04,430 So testing is easy, but in the LM applications we have a different situation. 11 00:01:04,430 --> 00:01:16,430 We can, uh, tell the same question twice to an LM application and we can get two different answers. 12 00:01:16,430 --> 00:01:25,520 So sometimes it is difficult to evaluate the, the quality or to test the quality of a LM application. 13 00:01:25,670 --> 00:01:31,850 So we have different methods, different ways to approach evaluation. 14 00:01:31,850 --> 00:01:34,040 And this is a very simple one. 15 00:01:34,040 --> 00:01:39,770 What we are going to do here, as you can see in the drop down with explanation. 16 00:01:39,770 --> 00:01:44,750 But what we are going to do here is first we have created a Rag application. 17 00:01:44,750 --> 00:01:52,790 So we are going to load a txt and we are going to make questions about the content in the in the txt 18 00:01:52,790 --> 00:01:53,210 file. 19 00:01:53,210 --> 00:01:53,750 Okay. 20 00:01:53,750 --> 00:01:59,900 But in this application we have included this evaluation part. 21 00:01:59,900 --> 00:02:08,479 And in the evaluation part we are going to enter a question and an answer we already know. 22 00:02:08,720 --> 00:02:17,480 And then this application is going to ask this security question if you want to the application in order 23 00:02:17,480 --> 00:02:28,610 to see if the application is going to, uh, give us a similar or an exactly and the same, uh, response 24 00:02:28,610 --> 00:02:29,300 that we have. 25 00:02:29,300 --> 00:02:38,180 And this is the interesting the interesting part, because this application is going to be able to see 26 00:02:38,660 --> 00:02:48,290 if the answer provided by the application is not exactly the same as the answer we have provided, but 27 00:02:48,290 --> 00:02:54,290 it it means the same or it is mostly the same, you see. 28 00:02:54,290 --> 00:03:02,030 So if instead of saying, I don't know, uh, the United States of America, the application says United 29 00:03:02,030 --> 00:03:08,930 States of America without the article first, this application is going to be wise enough to see, okay, 30 00:03:08,930 --> 00:03:10,790 the answer is good is correct. 31 00:03:10,790 --> 00:03:12,770 So the evaluation is positive. 32 00:03:12,770 --> 00:03:13,220 Okay. 33 00:03:13,220 --> 00:03:22,370 So this is the very interesting thing about evaluating an app an application with an LM application 34 00:03:22,370 --> 00:03:23,030 okay. 35 00:03:23,480 --> 00:03:24,950 So let's let's try this. 36 00:03:24,950 --> 00:03:29,600 So first we are going to load a txt file. 37 00:03:29,600 --> 00:03:34,550 This is a txt file from AI accelerator our company. 38 00:03:34,550 --> 00:03:38,750 And it's just explaining the different services we provide. 39 00:03:38,750 --> 00:03:46,880 So artificial intelligence consulting services for enterprise level companies for small businesses LM 40 00:03:46,880 --> 00:03:48,560 app development etc. etc.. 41 00:03:48,560 --> 00:03:48,830 Right. 42 00:03:48,830 --> 00:03:51,680 So we load this file. 43 00:03:51,680 --> 00:03:57,620 Now we are going to enter this security question or this testing question or whatever. 44 00:03:58,040 --> 00:04:00,380 We know the answer to this question. 45 00:04:00,380 --> 00:04:02,780 So that's the purpose of this security question okay. 46 00:04:02,780 --> 00:04:05,090 So I'm going to ask what. 47 00:04:06,380 --> 00:04:10,940 What is the name of the company? 48 00:04:12,180 --> 00:04:13,350 Company. 49 00:04:13,920 --> 00:04:14,880 Company. 50 00:04:15,510 --> 00:04:15,960 When you. 51 00:04:15,960 --> 00:04:20,160 When you talk and write at the same time, what is the name of the company? 52 00:04:20,160 --> 00:04:20,790 Okay. 53 00:04:20,790 --> 00:04:25,620 We press enter and then we enter the real answer to the question. 54 00:04:25,620 --> 00:04:31,290 We say, okay, the name of our company is a AI accelerator okay. 55 00:04:31,290 --> 00:04:33,180 And then we press enter. 56 00:04:33,420 --> 00:04:38,130 Now we enter the OpenAI API key. 57 00:04:38,160 --> 00:04:40,890 We press enter to submit the form. 58 00:04:40,890 --> 00:04:47,160 And the application now is going to make this question to the application okay. 59 00:04:47,160 --> 00:04:50,370 So question what is the name of the company. 60 00:04:50,370 --> 00:04:55,770 Real answer AI accelerator answer provided by the application. 61 00:04:55,920 --> 00:04:56,700 The same. 62 00:04:56,700 --> 00:05:00,750 Therefore the AI application answer was correct. 63 00:05:00,750 --> 00:05:03,780 So the evaluation has been positive. 64 00:05:03,870 --> 00:05:12,600 Usually when we evaluate an application an LM application we have like a range of questions like ten, 65 00:05:12,600 --> 00:05:15,330 15, 20 depends of the case right. 66 00:05:15,330 --> 00:05:22,980 But this is a very simple prototype in order to demo this kind of functionality okay. 67 00:05:22,980 --> 00:05:25,650 So let's take a look at the code. 68 00:05:27,240 --> 00:05:30,360 As always, two main files requirements. 69 00:05:30,840 --> 00:05:34,890 We are familiar with the packages here and Main.py. 70 00:05:35,400 --> 00:05:41,340 As you see, you are also familiar with this drag functionality we have prepared here. 71 00:05:41,340 --> 00:05:46,380 Okay, you did this in the previous blog with level one applications. 72 00:05:46,380 --> 00:05:53,970 Okay, so the only thing is you can see here are the typical Streamlit tricks like encapsulating functionality 73 00:05:53,970 --> 00:05:58,860 and functions, using streamlit elements, etc. etc.. 74 00:05:58,860 --> 00:06:02,520 Okay, so nothing new for you now. 75 00:06:02,520 --> 00:06:10,800 As always, download this code, execute our code in your computer, then repeat the code by yourself, 76 00:06:10,800 --> 00:06:17,130 line by line and try to understand what you are doing whenever you have a question or a doubt ChatGPT 77 00:06:17,130 --> 00:06:25,050 for extremely documentation, etc. etc. and only once you are comfortable with this application you 78 00:06:25,050 --> 00:06:26,880 go to the next lesson.