WEBVTT 00:00.240 --> 00:00.640 Okay. 00:00.640 --> 00:04.200 This is the single most important moment of the entire lab. 00:04.680 --> 00:09.320 So it's the chat function, and I've spaced it out a bit to really explain what's going on. 00:09.320 --> 00:14.080 It's a more powerful chat function that includes the calling of a tool. 00:14.560 --> 00:17.120 It starts exactly the same way as it did before. 00:17.240 --> 00:19.120 We create some messages. 00:19.120 --> 00:20.640 Role of system. 00:20.640 --> 00:22.440 The content is the system prompt. 00:22.560 --> 00:24.280 We add in the history from before. 00:24.680 --> 00:25.680 Role of user. 00:25.680 --> 00:27.680 The content is the user's message. 00:28.200 --> 00:35.120 And now do a pretty standard trick of looping again and again and again until a variable called done 00:35.160 --> 00:36.520 gets set to true. 00:36.680 --> 00:39.480 So it's going to keep going until done is true. 00:39.880 --> 00:41.160 What is it going to keep doing? 00:41.600 --> 00:43.360 It's going to call open AI. 00:43.880 --> 00:46.640 And this is the usual thing OpenAI. 00:49.280 --> 00:51.360 Create we're familiar with that. 00:51.400 --> 00:52.680 We pass in the model. 00:52.920 --> 00:56.320 We pass in the messages this set of of messages. 00:56.560 --> 00:59.520 But look we also pass in Tools. 00:59.720 --> 01:06.080 Tools is that JSON blob, that structure of JSON, which is going to be sent to the LM saying, these 01:06.080 --> 01:07.360 are all the things you can do. 01:07.360 --> 01:09.760 This is the description of everything you can do. 01:10.280 --> 01:10.680 All right. 01:10.720 --> 01:12.800 And back it comes into response. 01:12.800 --> 01:18.880 And what comes in response might just be the output to go to the user, or it might be a response that 01:18.880 --> 01:21.600 indicates that tools need to be called. 01:21.600 --> 01:23.520 So that is what we do next. 01:23.560 --> 01:25.880 We look at response choices. 01:28.800 --> 01:35.200 And this tells us whether the LM is finished and it's done or whether it wants to call a tool. 01:35.640 --> 01:40.080 So if the finish reason is tool calls, it wants to call a tool. 01:40.480 --> 01:46.960 Then we we look at the message itself, we pluck out the tool calls from it. 01:47.560 --> 01:53.880 And this this is where we call that function that we just wrote, that clever function, that sneaky 01:53.880 --> 01:54.600 function. 01:54.840 --> 01:55.760 Right here it is. 01:55.760 --> 01:57.560 Remember it handle tool calls. 01:57.650 --> 02:02.810 This is the one that is the glorified if statement, the one that has the fancy if statement. 02:02.810 --> 02:07.970 But there's also this more silly version here that just has an actual if statement. 02:08.210 --> 02:09.850 But it's going to call this. 02:10.210 --> 02:11.490 So back we go. 02:12.170 --> 02:16.850 So again if it wants to call tools then we do that. 02:16.850 --> 02:17.930 We handle the tool calls. 02:17.930 --> 02:19.010 We call the tools. 02:19.170 --> 02:21.570 And then into the set of messages. 02:21.570 --> 02:25.130 We put what it wanted to do and the results of doing it. 02:25.490 --> 02:29.890 And then we will loop back and call a second time. 02:29.890 --> 02:32.810 And we will keep doing this until the finish. 02:32.810 --> 02:34.770 Reason is not tool calls. 02:34.770 --> 02:40.850 It does not want to call tools with whatever, it's just responded, in which case we return the content 02:40.850 --> 02:45.250 because that is going to be the actual response to the user. 02:45.610 --> 02:46.930 I hope that made sense. 02:46.970 --> 02:50.450 If not, go through it slowly, carefully until it does. 02:50.650 --> 02:52.330 This is this is pivotal. 02:52.330 --> 02:54.690 This is how the whole thing hangs together. 02:55.170 --> 02:56.450 Now, I know what you're thinking. 02:56.450 --> 02:58.650 You're thinking it's all very well showing us all this code. 02:58.650 --> 03:00.170 But does it actually work? 03:00.410 --> 03:01.770 Well, let's find out. 03:01.770 --> 03:03.170 Let's run the code. 03:03.210 --> 03:04.370 Here we have it. 03:04.370 --> 03:05.450 Let's say hi there. 03:05.810 --> 03:06.170 Hello. 03:06.170 --> 03:07.370 How can I assist you today? 03:07.370 --> 03:08.250 What's your job? 03:09.490 --> 03:11.650 I'm currently the co-founder and CTO of Nebula. 03:11.690 --> 03:12.570 Very nice. 03:12.730 --> 03:14.050 Uh, do you have a patent? 03:15.490 --> 03:16.850 Yes I do. 03:16.890 --> 03:17.650 Very nice. 03:17.730 --> 03:21.410 Uh, who is your favorite musician? 03:22.210 --> 03:23.210 That's what I want to know. 03:25.090 --> 03:29.370 I hope you, uh, heard that success. 03:29.490 --> 03:30.050 So it says. 03:30.050 --> 03:32.010 I'm not sure about my favorite musician. 03:32.010 --> 03:33.130 I appreciate a variety. 03:33.130 --> 03:35.010 If you have a favorite, I'd love to hear about it. 03:35.050 --> 03:36.370 It called the tool. 03:36.610 --> 03:37.570 You heard my phone. 03:37.570 --> 03:38.010 Go. 03:38.210 --> 03:39.890 We have success. 03:39.970 --> 03:42.810 The agent was able to take that JSON. 03:42.850 --> 03:45.570 It knew that this was a tool that it could call. 03:45.610 --> 03:48.410 It was able to respond with the finish. 03:48.410 --> 03:50.250 Reason was was tool calls. 03:50.250 --> 03:56.420 And that caused us to to interpret that to make the call to R our function handle tool calls, which 03:56.420 --> 04:03.500 had our fancy code that ended up calling this function that ended up calling pushover, that then sent 04:03.540 --> 04:07.820 the push notification to my phone and it all hung together. 04:08.180 --> 04:09.660 All right, let's keep going. 04:09.980 --> 04:11.220 I'd like to get in touch. 04:12.220 --> 04:13.020 That's great to hear. 04:13.020 --> 04:14.260 Please share your email address. 04:14.300 --> 04:19.340 I'm at Edward donna.com. 04:23.340 --> 04:24.740 I know you're expecting it. 04:24.740 --> 04:25.100 Now. 04:25.100 --> 04:27.380 The, uh, the magic's worn off. 04:27.420 --> 04:28.780 You're not impressed anymore? 04:28.900 --> 04:30.500 Well, I think that's really cool. 04:30.620 --> 04:36.220 So it has indeed, of course, notified me about my own email address, which is great. 04:36.220 --> 04:39.260 And so it shows you that it does hold together. 04:39.420 --> 04:44.460 Uh, and you are indeed able to have these interactions and get it to text you. 04:45.060 --> 04:45.780 Okay. 04:46.100 --> 04:46.860 That's great. 04:46.860 --> 04:48.420 I hope you enjoyed it. 04:48.420 --> 04:54.900 But next it's all about doing this for you and actually deploying this live to production so that you 04:54.900 --> 04:59.180 can serve your own avatar on your personal website. 05:00.060 --> 05:02.940 So if you've followed everything so far, then many congratulations. 05:02.940 --> 05:07.660 If you haven't, then also congratulations because it gives you this great opportunity to go through 05:07.820 --> 05:09.500 work through this until you do. 05:09.540 --> 05:10.700 And if I can help. 05:10.740 --> 05:13.260 Email me or contact me anytime. 05:13.260 --> 05:13.860 LinkedIn with me. 05:13.860 --> 05:16.980 Message me so that I can help you out. 05:17.300 --> 05:24.900 So what I want to do now is show you how you can deploy this application in production for yourself, 05:24.900 --> 05:28.100 so that you can have this as your virtual resume. 05:28.380 --> 05:30.460 Surely this is the future of resumes. 05:30.500 --> 05:36.900 No longer will we have profiles or CVS resumes where you list out your skills and experience, but rather 05:36.900 --> 05:41.220 you'll have a chat bot that people can interact with to learn about your career. 05:41.220 --> 05:48.020 And what better way to highlight your your AI abilities and your abilities to work with Agentic AI than 05:48.020 --> 05:55.870 to have an agentic solution up on your website that will allow you to interact with people and talk 05:55.870 --> 05:56.910 about your career. 05:57.510 --> 06:01.550 So let me show you some steps to deploying this. 06:01.950 --> 06:06.590 So the first thing to do is to put the code into a Python module. 06:06.590 --> 06:09.910 And I have done that right here for you in App.py. 06:10.230 --> 06:13.710 And you'll see that this has all of the stuff you would expect. 06:13.710 --> 06:20.550 It's got the same information in here, but organized into nice little functions at the top of the functions 06:20.550 --> 06:22.430 that cover the tools that we call. 06:22.750 --> 06:26.710 And then there is a class me as in me or or you. 06:27.070 --> 06:33.070 And that's something which has the handle tool call that we know so well that's right here. 06:33.070 --> 06:40.110 And it has the system prompt and this loop that we just went through, this chat loop that has the while 06:40.150 --> 06:46.590 not done it calls open I create, it passes in the tools. 06:46.710 --> 06:48.670 I don't think I mentioned this actually, so I'm pleased. 06:48.670 --> 06:51.430 I pause for a minute here because I do want to show this one more time. 06:51.630 --> 06:53.590 We call chat completions create. 06:53.630 --> 07:00.270 We select the model, we pass in the messages, and we pass in the tools JSON, the description of the 07:00.270 --> 07:01.550 tools that it can call. 07:01.910 --> 07:04.790 And with what comes back, we take choices zero. 07:04.910 --> 07:08.030 And we look at the finish reason back in the response. 07:08.030 --> 07:11.230 And we see whether or not that is tool calls. 07:11.430 --> 07:19.510 And if so, then we loop through to make sure that we run all of the tools before calling it again to 07:19.550 --> 07:20.950 get the final response. 07:21.350 --> 07:26.910 So this is so critical, this this chat with its loop that I do encourage you to take a look at it. 07:27.390 --> 07:30.950 And now at the bottom here we have the Gradio code. 07:30.950 --> 07:38.270 We've got it set up here so that if this is just run as it is then it will create a gradio chat interface 07:38.470 --> 07:44.550 with showing me chat and with the messages and it will launch it. 07:44.950 --> 07:49.840 So what I can do now, in fact, is I can bring up a terminal which you may remember. 07:49.840 --> 07:53.200 You press control and the backwards tick mark. 07:53.360 --> 07:56.200 I can go into the week one folder. 07:56.200 --> 07:59.200 Here we are and I can now type. 07:59.240 --> 08:01.120 Remember you don't type Python. 08:01.160 --> 08:04.040 You type, you run to run something. 08:04.040 --> 08:06.960 In this way you've run app.py. 08:07.680 --> 08:11.720 And if I run that then it should come up. 08:11.960 --> 08:13.520 So this link. 08:13.520 --> 08:17.920 Now on this link I can press control and click it to bring it up. 08:18.400 --> 08:19.160 Up it comes. 08:19.200 --> 08:20.840 Let me make that a bit bigger for you. 08:21.120 --> 08:26.240 This is now an app running on my box and I can say hi there. 08:27.040 --> 08:27.400 Hello. 08:27.400 --> 08:34.560 How can I assist you and, and uh, do you have a patent and so on and there we go. 08:34.760 --> 08:35.360 Okay. 08:35.680 --> 08:43.240 So so that works going back here, what we now want to do is look at actually deploying this. 08:43.560 --> 08:48.920 So there are various ways you can deploy an application like this, but one of them that's particularly 08:48.920 --> 08:55.800 simple and elegant is brought to you by the lovely AI company Hugging Face, which also happens to be 08:55.800 --> 08:57.960 the company that owns Gradio. 08:58.000 --> 09:02.960 That beautiful, framework hugging face has something called hugging face spaces, and hugging face 09:02.960 --> 09:06.880 spaces is a way that you can deploy things really simply. 09:06.880 --> 09:10.280 And then if you wish, you can also embed them in your own home page. 09:10.280 --> 09:13.680 So it gives you a really nice, simple way to do it. 09:13.960 --> 09:15.880 So this is how you do it. 09:15.880 --> 09:21.400 You first have to go to Hugging Faces website and set up your own account if you don't already have 09:21.400 --> 09:21.880 one. 09:21.880 --> 09:25.360 So you may already have a hugging face account, but if not, you need to set one up. 09:25.760 --> 09:31.640 Once you've done with that, you go into the foundations folder in a terminal and you run this command 09:31.680 --> 09:33.400 Gradio deploy. 09:34.000 --> 09:39.240 It's then going to ask you a series of questions, a bunch of simple questions to which I've got the 09:39.240 --> 09:40.920 answers you have to give right here. 09:41.360 --> 09:42.760 And I'll go through and do it right now. 09:42.760 --> 09:46.370 But I'm not going to do it properly, because one of the questions it's going to ask you about some 09:46.370 --> 09:52.890 secrets that you'll say yes, and you're going to provide your OpenAI API key so that it can use that 09:52.890 --> 09:54.050 in the deployment. 09:54.050 --> 09:57.250 And you can also give your pushover tokens as well. 09:57.530 --> 10:01.330 And then I have got the rest of the instructions right there for you. 10:01.810 --> 10:06.810 And once that's done, it will be deployed into a hugging face space. 10:06.810 --> 10:08.850 And you can choose whether that is public or private. 10:08.850 --> 10:12.810 You can have it be private if you don't want other people using your key. 10:12.850 --> 10:16.170 Although as I say, it's super cheap, but people would have to. 10:16.210 --> 10:20.690 It would have to go viral for you to have any kind of significant spend. 10:20.690 --> 10:25.450 And of course, make sure that your OpenAI account, you never want to have that auto refilling so that 10:25.450 --> 10:30.210 if that ever were to happen, if your app did go viral, then you would be protected. 10:30.210 --> 10:33.490 And anyway, OpenAI has limits, as you probably know. 10:34.170 --> 10:36.890 So, uh, let's go and actually do this. 10:36.890 --> 10:38.570 Let's run Gradio deploy. 10:38.770 --> 10:43.970 So I'm going to bring up the terminal now, and then we will go and do this ourselves together.