WEBVTT 00:00.120 --> 00:02.000 Welcome back everybody. 00:02.000 --> 00:09.120 The Mac, PC, Linux people reunite for our video where we now start to work with our project. 00:09.120 --> 00:10.240 Your environment is set up. 00:10.240 --> 00:15.240 You should be looking at cursor looking nice and clean like this X out of any windows that are open, 00:15.280 --> 00:17.200 have it look a bit like like me. 00:17.240 --> 00:20.240 And then on the left here you've got the directories. 00:20.520 --> 00:26.520 The first one up here one foundations represents the directory for week one where we are right now. 00:26.560 --> 00:33.240 We open that up and I'd like you to click on lab one that comes up right here at the top. 00:33.240 --> 00:38.680 Here is a little section that tells you about the setup folder, which has all of the setup guide that 00:38.880 --> 00:40.880 by this point you should have already covered it all. 00:40.880 --> 00:41.920 So it should be fine. 00:42.040 --> 00:46.160 And then the guides folder has a bunch of useful guides. 00:46.160 --> 00:52.560 It has some stuff to take you through intermediate level Python, and it also has beginner Python. 00:52.560 --> 00:56.840 If you're new to Python coding, then it will take you through some self-study guides there. 00:56.840 --> 00:59.160 If you're willing to to to do that, I think it will be. 00:59.200 --> 00:59.730 It will. 00:59.730 --> 01:00.530 It will need some time. 01:00.530 --> 01:02.170 Investment will be well worth it. 01:02.170 --> 01:05.930 And then for intermediate level Python there's important stuff there too. 01:06.090 --> 01:07.610 There's a bunch of stuff. 01:07.650 --> 01:12.650 One of the things that comes up a lot in agents is working with async code. 01:12.690 --> 01:14.530 There's async stuff in there. 01:14.570 --> 01:20.370 There's also going to be things about generators which are very important, and a lot of good intermediate 01:20.370 --> 01:22.330 level Python material in there. 01:22.650 --> 01:28.330 If you're also not as experienced with coding, there's there's going to be stuff about debugging how 01:28.330 --> 01:29.610 to go about debugging. 01:29.730 --> 01:34.690 If you get something called a name error, that's something which has always got a way to work at a 01:34.690 --> 01:35.170 name error. 01:35.170 --> 01:36.650 So there's good details on that too. 01:36.690 --> 01:37.970 So please do look at the guides. 01:37.970 --> 01:39.330 I've put tons of stuff in there. 01:39.370 --> 01:41.770 There's also a troubleshooting guide if things go wrong. 01:41.810 --> 01:46.370 Troubleshooting takes you through every step, including some great diagnostics at the end. 01:47.170 --> 01:52.610 And then this next little comment here is saying that these labs think of it a bit like it's like an 01:52.610 --> 01:53.210 e book. 01:53.250 --> 01:57.090 It's got content and information as well as things to try. 01:57.370 --> 02:01.670 And what you'll find is that what you'll see as we go through this may be different. 02:01.710 --> 02:05.150 What you have may look different to what I go through right now. 02:05.190 --> 02:06.430 And you may say, well, that's annoying. 02:06.430 --> 02:07.590 I wish it was all the same. 02:07.590 --> 02:08.550 But here's the thing. 02:08.550 --> 02:13.430 The reason is because I like to keep adding more stuff, make it current. 02:13.470 --> 02:16.630 I try and make it very clear when it's something new that I've added. 02:16.710 --> 02:22.310 I bring in extra information to keep it up to date, and when people struggle and they hit problems 02:22.510 --> 02:25.910 or they need something explained better, I can put that in there. 02:25.950 --> 02:30.790 And that means that everyone benefits from where some people have struggled or I haven't explained something. 02:30.790 --> 02:32.110 Well, I can add it in there. 02:32.110 --> 02:33.990 So for me it's super important to do that. 02:34.270 --> 02:39.310 And the reason I don't come back and record all the videos as I do it well, first of all, that would 02:39.310 --> 02:43.350 take a lot of time and I'd rather spend time getting the this up to date. 02:43.590 --> 02:49.750 But also because that would erase some of the progress that people have made with videos they've watched. 02:49.750 --> 02:53.870 And it can be quite jarring if you're in the middle of a series and then the videos are changing under 02:53.870 --> 02:54.150 you. 02:54.350 --> 03:00.280 So it's really worked out better for me to keep these labs up to date as much as I can. 03:00.320 --> 03:02.720 So treat the labs like a resource. 03:02.720 --> 03:08.560 And the way that I suggest that you work with me is watch the video of me going through a lab. 03:08.600 --> 03:10.280 I'll go through it, I'll explain it. 03:10.280 --> 03:11.520 I'll execute things. 03:11.640 --> 03:15.840 I wouldn't necessarily suggest that you do it at the same time as me, because that's not really the 03:15.840 --> 03:16.400 point. 03:16.440 --> 03:22.200 The point is, as soon as I've done a lab, you come back and you go through it step by step yourself, 03:22.360 --> 03:27.320 and you use that as a way to build the learning, add in print statements and so on. 03:28.040 --> 03:29.680 Okay, a couple more things to say on this. 03:29.880 --> 03:36.520 I'm using here things called notebooks, which are a particular type of Python code that's organized 03:36.520 --> 03:38.280 in this way into cells. 03:38.400 --> 03:42.600 And there's going to be there's a guide on on how to use notebooks like this if you're not familiar 03:42.600 --> 03:43.240 with them. 03:43.240 --> 03:48.120 And I know that a lot of people from an engineering background initially get a bit frustrated like this 03:48.120 --> 03:50.560 because they'd rather work with just Python code. 03:50.800 --> 03:52.760 We will work with Python code as well. 03:52.760 --> 03:57.490 We're going to use both, but this kind of notebook has its place. 03:57.490 --> 04:03.330 It's very useful for research and development, which is an important part of being an AI engineer. 04:03.610 --> 04:09.210 And it's also great for learning because you can run things step by step, experiment, tweak, put 04:09.210 --> 04:10.810 in print statements, and learn. 04:10.850 --> 04:13.370 And so there is a benefit to this too. 04:13.490 --> 04:19.010 So even if you're not used to these kinds of notebooks, there are I promise you there's good stuff 04:19.010 --> 04:19.490 here too. 04:19.530 --> 04:23.770 And we will also do some Python coding using normal modules. 04:24.250 --> 04:24.970 All right. 04:25.010 --> 04:27.730 That's the end of of all of the details. 04:27.930 --> 04:30.330 Let's get to running this first lab. 04:31.450 --> 04:31.970 Okay. 04:32.010 --> 04:36.650 The first step is to go to the top right just here where you see select kernel. 04:36.690 --> 04:41.090 You click there and you have to choose your Python environment. 04:41.090 --> 04:44.890 And the way you do that is you type Python environments right here. 04:44.890 --> 04:49.770 And this first one up here with the star that says venv. 04:49.810 --> 04:50.890 It says recommended. 04:50.890 --> 04:52.810 And it's right in this directory. 04:52.930 --> 04:54.870 That is the one you want to pick. 04:55.110 --> 05:00.990 And hopefully that's going to come up really clear for you as Python 3.12 at the top. 05:01.030 --> 05:03.550 And once you've done that, it will be set up here. 05:03.550 --> 05:06.990 And you'll always when you come to a new lab you'll need to do that. 05:07.030 --> 05:10.270 And so so do bear that in mind for it to work. 05:10.550 --> 05:11.110 Okay. 05:11.510 --> 05:18.190 And then as like a final, final little thing here, it's, uh, got one more link to my LinkedIn in 05:18.190 --> 05:23.350 case in case you haven't succumbed by this point, remember that you can always link them with me. 05:23.510 --> 05:24.230 All right. 05:24.590 --> 05:30.310 So now what we do is we come here to the first of the cells. 05:30.350 --> 05:32.990 Each these things are called cells, these code cells. 05:32.990 --> 05:34.830 And we're going to do an import. 05:34.830 --> 05:36.110 And what import are we going to do. 05:36.110 --> 05:39.270 Well we're going to do this import here from dot env. 05:40.150 --> 05:47.030 And you can see that cursor is already suggesting what we could do right now from dot env import load 05:47.030 --> 05:48.030 dot env. 05:49.030 --> 05:53.480 And I hope you're familiar with what it means to import a function from a module, but if not, then 05:53.480 --> 05:57.000 please do check out the guides for a refresher on that. 05:57.000 --> 06:02.320 But I'm now going to run this code cell by simply pressing shift followed by return or enter. 06:02.320 --> 06:04.600 And that executes the cell and it's done. 06:05.080 --> 06:07.160 So the next thing we're going to do is use that function. 06:07.160 --> 06:08.200 And this is how we do it. 06:08.240 --> 06:11.800 We say load dot env and I'm getting prompted already. 06:11.800 --> 06:12.240 There it is. 06:12.240 --> 06:14.480 Load dot env override equals true. 06:14.680 --> 06:15.680 So what does this do. 06:15.720 --> 06:19.560 Load dot env is just a utility function which does something very specific. 06:19.760 --> 06:26.040 It looks for a file with a very particular name in your project root directory. 06:26.240 --> 06:31.440 And that file name happens to be, as you probably guessed, the name dot e. 06:32.440 --> 06:37.880 And it's assuming that that is where you're setting some of the environment variables that you want 06:37.880 --> 06:39.520 to use for this project. 06:39.520 --> 06:44.600 And it's just a utility that looks at that file, interprets it, and sets environment variables as 06:44.600 --> 06:45.040 a result. 06:45.040 --> 06:46.080 That's all it does. 06:46.200 --> 06:51.130 And override is true is just a nice little trick to know about by default. 06:51.170 --> 06:51.730 Load. 06:51.970 --> 06:52.290 Env. 06:52.330 --> 06:58.410 If you've already got an environment variable set, perhaps with your profile, then that takes priority. 06:58.410 --> 06:59.050 And it won't. 06:59.090 --> 07:00.290 It won't overwrite it. 07:00.450 --> 07:02.730 And that can be a nasty little gotcha. 07:02.810 --> 07:07.890 Because if in the past you've set up an open AI API key and you've and you've stored it somewhere global 07:07.890 --> 07:12.850 on your computer, that would take priority and you wouldn't be able to figure out why on earth OpenAI 07:12.850 --> 07:14.050 is rejecting your key. 07:14.610 --> 07:18.330 And so it turns out that this, as you can imagine, was a common problem in the past. 07:18.330 --> 07:23.250 And and I have learned to tell people to always put override equals true, which means that what's in 07:23.250 --> 07:25.490 the env file takes priority. 07:25.690 --> 07:27.730 And we run this by pressing shift enter. 07:27.890 --> 07:30.810 And now our environment variables will be set. 07:31.130 --> 07:32.610 So let's check that. 07:33.370 --> 07:35.610 This next cell does an import OS. 07:35.890 --> 07:42.810 And then we look for an environment variable that's called OpenAI API key which is exactly what we named 07:42.810 --> 07:44.410 it in the env file. 07:44.650 --> 07:50.870 And if it exists we're going to print it exists and it begins and give the first few letters. 07:50.910 --> 07:53.070 Otherwise it's going to say it's not set. 07:53.110 --> 07:54.950 Please go to the troubleshooting guide. 07:54.990 --> 07:57.350 I press shift and enter to run this. 07:57.350 --> 07:58.230 And there we go. 07:58.270 --> 08:00.110 The OpenAI key exists. 08:00.230 --> 08:02.070 It begins scmproj. 08:02.670 --> 08:05.750 All is good and I hope all is good for you too. 08:05.910 --> 08:07.910 If not, go to the troubleshooting guide. 08:08.790 --> 08:09.310 Okay. 08:09.350 --> 08:12.110 And now we have a super important import statement. 08:12.110 --> 08:15.990 And it looks like this from OpenAI import OpenAI. 08:16.710 --> 08:21.270 So if that seems weird to you, why is one in lowercase and one not anything like that? 08:21.270 --> 08:22.670 Again, go to the guides. 08:22.670 --> 08:26.750 I'll explain the difference between classes and modules and all the rest. 08:26.750 --> 08:32.390 But we are now going to execute this, and if this gives you any problems at all, then that might suggest 08:32.390 --> 08:34.550 a problem with your UV environment. 08:34.590 --> 08:38.950 Troubleshooting guide and always contact me if you get stuck in any way. 08:38.990 --> 08:40.590 I'm here to unstick you. 08:40.990 --> 08:41.750 All right. 08:41.910 --> 08:46.030 So what we're now going to do is create an instance of this class. 08:46.070 --> 08:48.040 Now I tend to always do it this way. 08:48.080 --> 08:49.880 OpenAI equals OpenAI. 08:50.120 --> 08:55.000 And that can be a bit confusing to people because again, when we're using the same word OpenAI, the 08:55.000 --> 09:01.920 proper way to call this would be something like OpenAI Python client, because what we're doing is creating 09:01.920 --> 09:05.000 an instance of the Python client library. 09:05.000 --> 09:08.280 But I just call that OpenAI because it's just simpler and shorter. 09:08.280 --> 09:10.120 But but that's what it stands for. 09:10.160 --> 09:12.320 We're creating an instance of that class. 09:12.320 --> 09:13.640 And let's run that right now. 09:14.000 --> 09:18.240 And it's worth bearing in mind that some people think that when you do this, you're creating something 09:18.240 --> 09:23.280 to do with the actual, like GPT large language model locally, but you're not. 09:23.480 --> 09:32.320 OpenAI is a lightweight library for connecting to OpenAI's endpoints on the cloud, so this is just 09:32.320 --> 09:34.960 what people sometimes call a client library. 09:34.960 --> 09:42.640 It's just a simple wrapper library around endpoints that calls HTTP endpoints to call the API that's 09:42.640 --> 09:43.640 running in the cloud. 09:43.640 --> 09:44.920 And that's all it is. 09:44.960 --> 09:49.530 It's a it's a simple lightweight class for that purpose only. 09:49.970 --> 09:52.170 So with that, let's keep going. 09:52.930 --> 09:57.370 What we're now going to do is create a list called messages. 09:57.370 --> 10:00.570 And this is a format that I hope most of you are very familiar with. 10:00.690 --> 10:03.570 On my LM engineering course, we do this one to death. 10:03.930 --> 10:06.130 And if you're not, I'll trumpet some extra stuff on the guys. 10:06.130 --> 10:10.610 I'm not going to explain too much of it right now, but conversations with large language models are 10:10.610 --> 10:16.090 often represented in this format that OpenAI invented and has become ubiquitous everywhere. 10:16.370 --> 10:20.970 Lists of dictionaries where each dictionary has a role and a content. 10:20.970 --> 10:23.410 And in this case we say the role is user. 10:23.450 --> 10:27.410 The content is this simple question what is two plus two? 10:27.730 --> 10:34.330 So we run that and now we're going to actually make a call to OpenAI and check everything is working 10:35.170 --> 10:35.770 okay. 10:35.810 --> 10:39.650 And so now we're going to come here and type the code to call OpenAI. 10:39.770 --> 10:42.250 And cursor is going to make this very easy for us. 10:42.250 --> 10:52.550 We're just going to type response equals open I dot and it's a chat dot completions dot create. 10:52.550 --> 10:55.310 And you can see that cursor is already filling that in for us. 10:55.310 --> 10:58.510 I can just press tab and it's going to complete that right away. 10:58.710 --> 11:01.670 And it's going to say the model is GPT four mini. 11:01.670 --> 11:03.110 And the messages are messages. 11:03.110 --> 11:04.110 That's perfect. 11:04.310 --> 11:10.070 And it's also going to take the response, take the first choice dot message content and print it. 11:10.070 --> 11:11.990 And that's exactly what we want to do. 11:12.310 --> 11:15.950 This structure of calling OpenAI is something that is so common these days. 11:15.950 --> 11:20.070 If you're new to it, then take a look and I'll try and cover some in the guides. 11:20.070 --> 11:25.710 But people who are familiar with this will see this as a very simple way to call OpenAI. 11:25.710 --> 11:29.990 And if we run this, we get the response to the question what is two plus two? 11:30.030 --> 11:32.110 We get two plus two equals four. 11:32.350 --> 11:37.350 So we know we are connected to OpenAI and so far so good. 11:38.150 --> 11:41.230 So hopefully you're now comfortable that we can both call OpenAI. 11:41.270 --> 11:42.270 We can get responses. 11:42.270 --> 11:48.560 We can use the very standard chat completions API structure and cursor will fill it in for us if we 11:48.560 --> 11:49.680 ever forget what it is. 11:50.000 --> 11:52.600 But now let's ask a harder question. 11:52.600 --> 11:55.200 It's time for us to be a little bit agentic. 11:55.200 --> 11:58.240 We're going to put one foot in the agent camp. 11:58.280 --> 12:02.760 It's more a workflow than an agentic I but but still it's getting there. 12:02.960 --> 12:04.600 We're going to ask a question. 12:04.640 --> 12:09.360 Please propose a hard, challenging question to assess someone's IQ. 12:09.520 --> 12:12.240 Respond only with the question. 12:12.240 --> 12:13.760 So that's going to be our question. 12:13.760 --> 12:17.800 And we're going to put that into the standard messages structure. 12:18.280 --> 12:19.920 And now we're going to ask it. 12:19.920 --> 12:23.800 So again we say response and it's already filling it in for us. 12:23.800 --> 12:25.280 We press tab to complete that. 12:25.280 --> 12:26.280 We press tab. 12:26.320 --> 12:27.960 But we're just going to change this slightly. 12:27.960 --> 12:34.600 We're going to make this reply equals maybe I should make this like like question equals. 12:34.600 --> 12:37.600 So we're going to take back the what comes back and we're going to make it question. 12:37.600 --> 12:39.800 And then we'll just print the question again. 12:39.800 --> 12:41.080 Cursor fills that in. 12:41.120 --> 12:42.450 Let's run that. 12:42.450 --> 12:43.250 And here we go. 12:43.250 --> 12:45.730 Here we have a question from OpenAI. 12:46.290 --> 12:50.570 If a train leaves a station traveling at 60 miles an hour and another train is a different station, 12:50.570 --> 12:53.490 120 miles away at the same time traveling 90. 12:53.610 --> 12:56.290 How far from the first station will the two trains meet? 12:56.330 --> 12:58.810 A nice brain teaser right there. 12:59.570 --> 13:03.450 So what we're going to do now is we're going to take this question, and we're going to make a second 13:03.450 --> 13:06.410 call to an LLM to answer the question. 13:06.410 --> 13:11.490 And that's why I describe this as an agentic pattern, because it is making multiple calls. 13:11.490 --> 13:16.090 We are orchestrating multiple calls to llms to solve a bigger problem. 13:16.090 --> 13:18.410 And so it's going to be relatively easy. 13:18.450 --> 13:22.210 We're simply going to continue this by again making another messages. 13:22.250 --> 13:28.370 Look at how cursor fills it in for us messages with role as a user and the content is that question. 13:28.490 --> 13:29.130 Great. 13:29.130 --> 13:30.170 Let's run that. 13:30.330 --> 13:31.450 That is now messages. 13:31.450 --> 13:34.330 And now we're going to ask OpenAI again. 13:34.490 --> 13:39.570 So we're going to say response is OpenAI create. 13:39.770 --> 13:43.670 And again, cursor fills it all in and it fills in the answer. 13:43.670 --> 13:48.430 And it's even clever enough to recommend that we call it answer instead of question. 13:48.750 --> 13:50.150 Honestly it's incredible. 13:50.390 --> 13:52.430 So this is exactly what we want to do. 13:52.470 --> 13:58.110 We're making the same kind of call to OpenAI's chat completions API. 13:58.150 --> 14:01.150 The simplest way to call OpenAI in the cloud. 14:01.430 --> 14:06.110 And we are now getting back the answer to the question, and we will print that answer. 14:06.110 --> 14:09.350 And I believe the answer we're looking for is 48, I think. 14:09.550 --> 14:10.310 Let's see. 14:10.510 --> 14:12.150 So we'll get that to print. 14:12.710 --> 14:14.070 And here we go. 14:14.310 --> 14:20.630 A bunch of information right here written with 48 at the end there. 14:21.710 --> 14:22.870 And you may know this already. 14:22.910 --> 14:26.710 You recognize that this has come back in a form of markdown. 14:26.710 --> 14:31.430 It's got some markdown stuff in it which llms love to talk in markdown. 14:31.430 --> 14:37.070 We can just add in a little code in here and we can do from there. 14:37.990 --> 14:38.510 Ha ha. 14:38.510 --> 14:38.640 Ah. 14:38.920 --> 14:40.200 Is it spooky? 14:40.240 --> 14:46.280 It is spooky the way cursor does this sometimes from IPython display, import markdown and display. 14:46.280 --> 14:49.040 And then the next line that cursor is telling us to write. 14:49.200 --> 14:50.760 I don't really have a job here, do I? 14:50.920 --> 14:53.720 Display markdown, but the only thing it's got wrong. 14:53.720 --> 14:55.000 Thank goodness it got something wrong. 14:55.000 --> 14:58.400 Is that what we actually want to display is the answer, not the question. 14:58.400 --> 15:04.160 So if we do that and we run this, then we see the answer formatted nicely. 15:04.160 --> 15:11.360 I'll let you read through its work and it comes up with the conclusion of 48 miles. 15:11.640 --> 15:12.840 Uh, so there we go. 15:13.120 --> 15:20.120 That is our first semi agentic pattern, built just in a few minutes. 15:20.200 --> 15:24.480 If any of that was unfamiliar to you, check out the guides, play with it, look at it. 15:24.520 --> 15:26.920 And as I say, congratulations on getting here. 15:26.920 --> 15:30.560 But at the end of this lab there is an exercise. 15:30.560 --> 15:34.040 And I do encourage you to do the exercises. 15:34.040 --> 15:40.130 This is where you really get to to show that you've done this, you may be thinking, okay, there wasn't 15:40.130 --> 15:45.530 much commercial about this project, and I do like to try and weave commercial angles in wherever I 15:45.530 --> 15:48.250 can to make this applicable to business. 15:48.370 --> 15:53.130 And so I've got an exercise for you to now make this applicable slightly to business. 15:53.330 --> 16:02.010 So we are going to try and ask an LLM, first of all, to pick a business area that might be worth exploring 16:02.010 --> 16:04.170 for an AI opportunity. 16:04.690 --> 16:12.450 Secondly, to identify a pain point in that business sector that it's decided on, and then that that's 16:12.450 --> 16:15.130 one that it needs to explicitly explain. 16:15.130 --> 16:19.290 What is that challenge that might be ripe for an agentic solution. 16:19.290 --> 16:25.810 And then the third step will be to actually propose an agentic AI solution to that problem. 16:26.170 --> 16:34.050 So I would like you in this empty cell here, probably assisted by cursor, uh, to, to to go through 16:34.050 --> 16:37.230 and make this happen with three calls to LMS. 16:37.430 --> 16:43.430 First of all, you put something in this message to explain that you want it to pick a business sector. 16:43.910 --> 16:47.350 Then you get back the response based on that response. 16:47.470 --> 16:54.550 You read in the business idea, and then you will repeat these three steps in order to get the the pain 16:54.550 --> 16:59.830 point, and then to get the Agentic AI solution and print them all out. 16:59.830 --> 17:07.390 Print them in markdown if you wish, uh, in nicely formatted, and use that as a way both to to experiment 17:07.390 --> 17:13.830 with some agentic workflows and also actually to get some commercial benefit, to get perhaps an interesting 17:13.830 --> 17:19.350 idea about a way that you can apply Agentic AI to a real business pain point. 17:19.830 --> 17:20.510 All right. 17:20.510 --> 17:22.230 I hope you enjoy that challenge. 17:22.230 --> 17:28.190 If you have any problems with it, of course, then you can either get help from from ChatGPT or cursor, 17:28.190 --> 17:34.990 or just drop me a note and I will see you for the final lecture for this first day.