WEBVTT 00:02.000 --> 00:04.560 So how do we get started building a new app? 00:04.920 --> 00:13.600 We could go directly into cursor and say something like build a flashcards app, and cursor will probably 00:13.600 --> 00:20.040 build something decent, but it is very unlikely it will match our expectations and we will need to 00:20.040 --> 00:21.880 have a long conversation. 00:21.920 --> 00:25.800 Practically a lot of back and forth to get all the details right. 00:26.080 --> 00:28.920 So this process is not really optimal. 00:29.280 --> 00:35.560 Instead, it is best to invest some time in the beginning and create a proper specification, or how 00:35.560 --> 00:40.400 it is sometimes referred to as a product requirements document PRD. 00:41.040 --> 00:48.320 And this is a key step in building a proper software, because the AI cannot know what is inside our 00:48.320 --> 00:48.840 head. 00:48.840 --> 00:55.000 So we may know already how the app should look like, but we're going to say that the AI doesn't know 00:55.000 --> 00:55.760 what we want. 00:55.800 --> 00:58.000 So we need to be really, really specific. 00:58.400 --> 01:01.750 So this is why this is a key step in building any software. 01:01.750 --> 01:04.470 Even if you think, oh, this is a relatively small app. 01:04.630 --> 01:06.870 It should be self-explanatory. 01:06.990 --> 01:08.750 Well, nothing really is. 01:09.390 --> 01:15.790 So for the process of creating this PRD, we can use cursor directory and have a conversation here, 01:15.790 --> 01:18.870 because in itself we have here the agent mode. 01:18.870 --> 01:24.510 But we can also have here the ask mode that pretty much the chat mode and say, hey, I'm trying to 01:24.510 --> 01:25.350 build this app. 01:25.350 --> 01:30.990 Can you come up with product requirements document and something like that? 01:31.350 --> 01:35.190 But ideally what I like to use is a regular LM. 01:35.670 --> 01:37.630 In this case I'm using ChatGPT. 01:37.670 --> 01:42.070 But you can use Gemini from Google or something from anthropic or anything else. 01:42.070 --> 01:43.990 Doesn't really matter so much. 01:44.510 --> 01:48.070 It's more important to have this conversation in the beginning. 01:48.670 --> 01:55.230 And if you don't know exactly what you're trying to build, just use ChatGPT to ask questions. 01:55.230 --> 02:01.460 Trying essentially to consult with a chatbot in terms of what are your options to building something 02:01.460 --> 02:06.460 like that and how this PRD document can definitely look like? 02:06.500 --> 02:10.660 And so if you don't know exactly which technology you should use or how exactly you should build something 02:10.660 --> 02:14.620 like that, do have this exploring conversation in the beginning. 02:14.660 --> 02:17.580 Don't try to do everything in one shot. 02:17.620 --> 02:21.940 That would be in a technical language, like a zero shot prompt. 02:22.460 --> 02:24.980 Just sending something and expecting it to work. 02:25.260 --> 02:27.260 It generally doesn't work like that. 02:27.380 --> 02:33.020 So for this flashcard application, I'm gonna attach an initial prompt as a resource to this lecture 02:33.020 --> 02:33.980 as a text file. 02:34.220 --> 02:41.220 This is exactly what I'm going to use next in order to get ChatGPT to write this PRD document, which 02:41.220 --> 02:44.340 is essentially a specification for our application. 02:44.900 --> 02:48.060 So we're going to go ahead and see exactly what's going on here. 02:48.100 --> 02:49.580 I'm going to submit it first. 02:49.980 --> 02:53.140 And we're going to take a look at exactly what it's saying. 02:53.140 --> 02:55.900 So I need to write a software specification. 02:56.220 --> 02:57.940 And I need your help writing it. 02:57.940 --> 03:03.570 I'm describing here what I'm trying to build, and I already know which technology I want to use. 03:04.050 --> 03:08.770 And also I'm listing here the features that I think the application should have. 03:09.090 --> 03:14.690 And the features can be essentially a brain dump, because we're going to use the LLM to try to organize 03:14.690 --> 03:15.410 everything. 03:15.770 --> 03:19.250 Apart from this, we're going to format this as markdown. 03:19.770 --> 03:27.130 And probably the last prompt here is the most important one, because I don't want ChatGPT to start 03:27.170 --> 03:29.930 outputting a specification instead. 03:30.050 --> 03:31.370 This is the consulting part. 03:31.410 --> 03:36.090 This is the part where is there something that's still unclear about my requirements? 03:36.290 --> 03:43.290 Let's clarify that before we generate anything, because that's again a much more efficient process. 03:43.330 --> 03:47.410 And we're going to see here we're going to get here clarification questions. 03:47.690 --> 03:50.930 It's going to ask us exactly how do you want to have this. 03:51.530 --> 03:53.970 Do we want to have text categories. 03:54.490 --> 03:56.930 How about user accounts and so on. 03:56.930 --> 04:00.000 So there can be a lot of things that go into this. 04:00.400 --> 04:07.280 And once we clarify all these questions, we're going to get back a software specification. 04:07.320 --> 04:12.120 And again, a software specification is something that I'm going to attach as a resource to this lecture. 04:12.480 --> 04:14.280 And you're going to take a look at it. 04:14.280 --> 04:16.720 And we're going to see in a second how it looks like. 04:16.760 --> 04:19.560 I'm not going to bore you with answering all these questions. 04:19.560 --> 04:24.480 It's self-explanatory, but I'm going to show you in the next step how the specification looks like. 04:25.320 --> 04:29.040 So as a next step, I answered all the questions to the best of my knowledge. 04:29.400 --> 04:34.880 And what came out of this is this flashcards web application specification. 04:35.120 --> 04:38.360 And it contains various let's say, categories. 04:38.640 --> 04:44.240 So what I would do next is I'm going to go ahead and copy this entire specification that is here. 04:44.880 --> 04:49.680 And for the next step I'm going to check it and make some changes to it. 04:49.680 --> 04:54.360 And once I'm done I'm also going to share this specification with you as a resource to this lecture. 04:54.640 --> 05:00.550 But generally keep in mind that even if you share the exact same prompt and put it into an LM using 05:00.550 --> 05:01.710 the exact same model. 05:02.470 --> 05:09.430 The output will always be slightly different because generally speaking, llms are non-deterministic, 05:09.430 --> 05:13.830 meaning for the same input you can get different outputs. 05:13.870 --> 05:19.590 But what I'm going to do with this specification and it's super important to review it inside here I 05:19.630 --> 05:22.550 created a new folder which is called flashcards. 05:22.630 --> 05:26.190 And inside this folder what I'm going to do here is create a new file. 05:26.190 --> 05:28.310 And I put it in the docs directory. 05:28.670 --> 05:30.790 And I'm going to call it specification dot. 05:31.870 --> 05:34.950 And this is the specification in the markdown format. 05:35.590 --> 05:42.190 Again it's super important that if you're building a product not just to rely on the LM, but to go 05:42.190 --> 05:45.270 over this specification, read absolutely every line. 05:45.790 --> 05:51.950 It does contain a few technical aspects, but most of the content is actually how the application is 05:51.950 --> 05:56.430 supposed to work, and this is important for building the application itself.