WEBVTT 00:00.440 --> 00:04.520 And here we are back in cursor and I've collapsed all the folders. 00:04.520 --> 00:09.880 But I'll go back into the foundations folder and to lab three. 00:10.360 --> 00:11.960 Lab three for week one. 00:11.960 --> 00:13.800 Day four, which is where we are. 00:14.440 --> 00:17.840 So this is where things get interesting. 00:18.400 --> 00:27.520 So I have a directory called me in in within foundations and within me I've put some files. 00:27.520 --> 00:29.400 One of them is called LinkedIn. 00:30.840 --> 00:37.080 And LinkedIn is uh I see it's not I don't think I can open it in here. 00:37.080 --> 00:39.600 I imagine this doesn't like being a PDF. 00:39.600 --> 00:40.080 Nope. 00:40.400 --> 00:42.960 So you you will have to take my word for it. 00:43.000 --> 00:48.640 What this is, is a PDF version of my LinkedIn profile. 00:48.640 --> 00:50.240 And you can just get that from within LinkedIn. 00:50.240 --> 00:56.320 You can go into LinkedIn and and click on on one of the dot dot dot ellipsis menus and download your 00:56.320 --> 00:57.400 profile as a PDF. 00:57.400 --> 00:58.480 And that's what I did. 00:58.560 --> 01:01.050 And I've put it right here as LinkedIn PDF. 01:01.570 --> 01:06.370 And I've also got a summary text which is just called summary txt. 01:06.610 --> 01:09.250 And it's just two sentences about me. 01:09.410 --> 01:11.690 And those two things are in me. 01:11.850 --> 01:17.010 And the to do for you as you do this lab is to replace these documents with something about you. 01:17.050 --> 01:22.690 The LinkedIn should be your LinkedIn PDF profile or a PDF of your resume or anything like that. 01:22.770 --> 01:27.690 The summary put some stuff about you and include some extra some fun stuff. 01:27.730 --> 01:35.170 A fun fact about you, something that you would like to people to know about you and all will become 01:35.170 --> 01:35.530 clear. 01:35.570 --> 01:38.050 Don't worry, it's not going to go anywhere without your control. 01:38.050 --> 01:40.890 But this will hopefully become clear very soon. 01:41.450 --> 01:42.210 All right. 01:42.370 --> 01:44.170 So if we go back to the lab. 01:44.530 --> 01:47.610 So as I said, we're not going to actually use tools just yet. 01:47.650 --> 01:48.890 Even though we've been talking about them. 01:48.930 --> 01:50.170 Tools is going to come tomorrow. 01:50.210 --> 01:52.290 We're going to lay the groundwork. 01:52.810 --> 01:53.330 Okay. 01:53.610 --> 01:59.340 So I'm now going to import a few packages load of you know well but an open eye, you know. 01:59.380 --> 02:06.580 Well, but PDF pypdf2 and gradio you might not know gradio prior students of mine will know well because 02:06.620 --> 02:08.540 you know that I am super. 02:08.540 --> 02:12.020 I adore Gradio, I am passionate about this, this this platform. 02:12.020 --> 02:17.700 I am a horrible front end engineer and gradio is something that can make beautiful front ends. 02:17.740 --> 02:19.660 Even for someone terrible like me. 02:19.700 --> 02:20.660 It's really great. 02:20.660 --> 02:24.900 It's a way to build data science user interfaces very easily indeed. 02:24.940 --> 02:27.620 We will do that, but I won't cover it in any detail. 02:28.180 --> 02:34.420 Pypdf2 is an example of a popular library for parsing PDF files, and if you want to know more about 02:34.420 --> 02:40.140 these, you can simply ask ChatGPT to write you a quick guide on them if you ever want to. 02:40.180 --> 02:45.940 To get the right kind of popular resource for doing something, you can just ask Claude or ChatGPT. 02:46.260 --> 02:46.940 If you want. 02:46.940 --> 02:47.620 You can, you can. 02:47.660 --> 02:50.500 The actual place they come from is called PyPI. 02:50.700 --> 02:52.980 And this is the the the package index. 02:52.980 --> 02:56.300 This is where open source Python packages live. 02:56.300 --> 03:01.590 And you can go there and search for different packages, look at their GitHub repos and find out more 03:01.590 --> 03:03.270 about them that way as well. 03:03.590 --> 03:03.950 All right. 03:03.950 --> 03:06.470 But I'm going to import these. 03:06.790 --> 03:09.670 And by the way I already set up here on the top right. 03:09.670 --> 03:14.230 I had uh, made sure that my virtual environment was set. 03:14.870 --> 03:16.430 It's taking a bit of time there. 03:16.430 --> 03:22.950 I guess it's a gradio that needs to load in a bunch of things, and it's done 14 seconds to to do all 03:22.990 --> 03:23.870 of those imports. 03:23.910 --> 03:25.190 I guess a lot was happening. 03:25.470 --> 03:25.910 All right. 03:25.910 --> 03:31.790 So now we load our env and we initialize the OpenAI client library. 03:32.190 --> 03:32.950 So look at this. 03:32.950 --> 03:40.230 I now simply create a PDF reader and I point it at the my LinkedIn PDF. 03:40.590 --> 03:44.310 And I read in the pages calling page extract text. 03:44.310 --> 03:46.710 And I just bring it all together. 03:46.990 --> 03:48.870 And that's as simple as that. 03:48.870 --> 03:55.590 So so now this LinkedIn variable, let's just print that LinkedIn variable so we can we can see it print 03:55.720 --> 04:01.840 LinkedIn and you'll see a whole bunch of stuff about me, and the output is truncated. 04:01.840 --> 04:03.200 So you're only seeing a little bit about me. 04:03.200 --> 04:06.240 But but there's there's more than you would need to know right there. 04:06.240 --> 04:09.680 But since you're surely already connected with me on LinkedIn, you know all this already. 04:10.000 --> 04:16.400 Um, then, uh, here I am going to load in the summary text about me. 04:16.600 --> 04:20.720 This is needed for PC users sometimes, and if you're not from. 04:20.720 --> 04:23.240 Yeah, I'm sure you are familiar with this, but it's covered in the guide. 04:23.240 --> 04:25.480 And then I'll set a variable called name to my name. 04:25.480 --> 04:27.120 And you should set that to be your name. 04:27.680 --> 04:28.480 All right. 04:28.520 --> 04:32.760 And now so one thing we didn't talk about before is system prompts and user prompts. 04:32.760 --> 04:34.280 And they are fairly commonplace now. 04:34.280 --> 04:40.120 So you probably know that we've always been just prompting with a single user prompt up to now. 04:40.120 --> 04:43.240 But you can actually specify two different prompts. 04:43.240 --> 04:50.600 The system prompt is intended to be more the overall instructions that sets the context for the task 04:50.600 --> 04:56.330 at hand, and the format and the way it should be responded to, and then the user prompt is the actual 04:56.330 --> 04:58.130 question coming from the user. 04:58.130 --> 05:02.530 And in this case, it will be handy for us to separate out these these two concerns. 05:02.690 --> 05:07.650 So for the system prompt I'm saying you are acting as name. 05:07.650 --> 05:10.130 So this will be me and it will be you. 05:10.410 --> 05:15.810 You're answering questions on that person's website, particularly questions related to their career, 05:15.850 --> 05:17.810 background, skills and experience. 05:18.050 --> 05:23.890 Your responsibility is to represent for interactions on the website as faithfully as possible. 05:24.330 --> 05:27.610 You're given a summary of the background in LinkedIn profile, blah blah blah blah blah. 05:28.090 --> 05:29.570 It be professional and engaging. 05:29.570 --> 05:33.010 So this is about setting the tone, which is a good thing to do in a system prompt. 05:33.010 --> 05:36.810 And then if you don't know the answer, say so is very good. 05:36.850 --> 05:39.250 Prompting context to give. 05:39.770 --> 05:40.410 Okay. 05:40.690 --> 05:45.050 So then I'm adding in the summary using that summary variable. 05:45.050 --> 05:47.490 And I'm also putting in the LinkedIn profile. 05:47.490 --> 05:51.370 And I'm using markdown tags to kind of show this as sort of headings. 05:51.930 --> 05:56.410 And then I end with with this context, please chat with the user. 05:56.450 --> 06:00.970 Always staying in character as as you. 06:01.570 --> 06:02.050 All right. 06:02.090 --> 06:02.970 Now the plot thickens. 06:02.970 --> 06:03.450 Now you know what? 06:03.450 --> 06:04.890 I'm the game I'm playing here. 06:04.890 --> 06:05.890 You know what we're doing. 06:06.010 --> 06:08.890 Let's run that and let's just print that. 06:08.890 --> 06:12.090 So we see what system prompt actually looks like. 06:13.010 --> 06:13.810 And here we go. 06:13.850 --> 06:17.330 You can see it's just true to form exactly as we were expecting. 06:17.810 --> 06:19.050 And it will have within it. 06:19.090 --> 06:28.090 Of course the stuff about me and the LinkedIn URL, LinkedIn PDF contents and so on and hopefully yours 06:28.090 --> 06:29.170 will have yours. 06:29.210 --> 06:31.130 And you should put in some print statements to check it. 06:31.170 --> 06:34.090 Make sure you're very comfortable with everything happening so far. 06:34.850 --> 06:41.370 So we're about to use Gradio to bring up a user interface that will allow us to chat based on an LLM 06:41.370 --> 06:43.570 that's armed with this system prompt. 06:43.770 --> 06:49.370 And the way you work with Gradio is you need to write a function which will be like a callback function 06:49.370 --> 06:54.460 that Gradio will call back to when it needs to to do some processing, when when a user has typed something 06:54.460 --> 06:54.900 in. 06:54.900 --> 07:00.500 And the style of this particular callback function is a style where you write a function called chat, 07:00.540 --> 07:07.100 which takes a message that the user is typing in, a message being sent, and the history of all prior 07:07.100 --> 07:10.340 messages, which comes in OpenAI's format. 07:10.620 --> 07:18.860 And what we need to do is call an LLM and return the the next the response from the LLM to go in the 07:18.860 --> 07:19.380 chat. 07:19.660 --> 07:23.620 So I'll let you look through this to convince yourself that what I'm doing is the right thing. 07:23.620 --> 07:28.580 But basically I build an OpenAI style list of dictionaries. 07:28.580 --> 07:33.660 I begin with the system prompt role system content and the system prompt. 07:33.660 --> 07:40.660 I add in the history from Gradio, and then I bung in this muses message that's just come in right now. 07:41.100 --> 07:51.310 I then call OpenAI Dot create for GPT four mini with exactly these messages, and I return the response 07:51.350 --> 07:53.950 choices zero message content. 07:54.150 --> 07:57.830 So this should all be stuff that's fairly familiar to you. 07:57.830 --> 08:00.590 And then using Gradio is beautifully simple. 08:00.750 --> 08:01.550 You just call. 08:01.550 --> 08:02.790 In this case we want a chat. 08:02.790 --> 08:04.750 And so you can call chat interface. 08:04.750 --> 08:08.910 We tell it the function that we've written so it knows it can call this as the callback. 08:09.190 --> 08:11.070 And we ask it to launch. 08:11.510 --> 08:14.990 And when I run that I get an interface like this. 08:15.350 --> 08:17.750 And I can say hi there. 08:18.710 --> 08:19.830 Hello, welcome to my website. 08:19.830 --> 08:21.070 How can I assist you? 08:21.070 --> 08:27.030 And so I can say something like, what is your greatest accomplishment? 08:27.870 --> 08:29.150 And we'll see what it says. 08:29.750 --> 08:36.310 I consider my greatest accomplishment to be the co-founding of Nebula, where we're leveraging generative 08:36.310 --> 08:40.470 AI and proprietary Llms to transform how people source and engage talent. 08:40.590 --> 08:41.670 Fantastic. 08:41.950 --> 08:45.070 And and it's also talking about the company I founded as well. 08:45.350 --> 08:49.120 So it's a very legit, very good answer. 08:49.120 --> 08:51.400 I think it's probably true, which is always a good thing. 08:51.600 --> 08:54.880 And so very nicely handled, let's say. 08:55.040 --> 09:03.160 What is a challenge that you encountered and needed to overcome? 09:03.440 --> 09:07.840 It's a hard kind of professional question that you might be asked in an interview. 09:08.480 --> 09:09.200 Okay. 09:09.400 --> 09:13.200 One of the significant challenges I faced during my transition from a successful career at JP Morgan 09:13.200 --> 09:16.440 to starting my own company, until this is all very true. 09:16.480 --> 09:19.360 I don't know how he got that from my from my LinkedIn profile. 09:19.360 --> 09:21.280 I guess it must be somewhere or something. 09:21.280 --> 09:23.120 That's that's kind of implied within it. 09:23.120 --> 09:25.640 But I mean, it happens to be spot on. 09:25.840 --> 09:28.160 So this is a very good answer. 09:28.160 --> 09:35.880 And so wonderfully, in the space of just a few minutes, we've built a chat interface and we've used 09:35.880 --> 09:42.920 resources to arm it with information about me so that it's able to act as me and be like a professional 09:42.960 --> 09:48.080 avatar, an alter ego for me in answering questions about my career.