WEBVTT 00:00.120 --> 00:04.160 And so this is what I have in store for you over the next three weeks. 00:04.160 --> 00:05.960 This is the curriculum ahead. 00:06.240 --> 00:08.280 There's going to be three different types of session. 00:08.280 --> 00:15.600 There's going to be some core sessions that focus on Agentic AI, on pn10, on the core capabilities. 00:15.840 --> 00:22.200 There's going to be some sessions that are purely dedicated to integrations, because integrations is 00:22.200 --> 00:25.920 such a huge part of working with an A10 and delivering business impact. 00:25.920 --> 00:28.160 So that's a whole topic of its own. 00:28.440 --> 00:30.960 And then there's real world projects. 00:30.960 --> 00:34.560 That is the third type of focus area for us. 00:34.720 --> 00:35.000 Okay. 00:35.040 --> 00:37.040 So how does this fit into the three weeks. 00:37.240 --> 00:37.720 All right. 00:37.720 --> 00:43.040 So each of the three weeks I have given a punchy title to describe what we're going to try and achieve. 00:43.200 --> 00:45.680 The first week is called automate. 00:45.800 --> 00:51.080 It's about using Anytime Cloud to make workflows for your business. 00:51.080 --> 00:52.280 Automate is week one. 00:52.280 --> 00:54.080 Week two is accelerate. 00:54.080 --> 00:55.520 This is next level stuff. 00:55.520 --> 00:59.200 This is saying we want to add more juice to our business or our clients. 00:59.200 --> 01:00.640 We want to have voice agents. 01:00.640 --> 01:02.040 We want to use rag. 01:02.240 --> 01:06.130 You may not know what Rag is or you might you certainly probably you've probably heard of it because 01:06.130 --> 01:07.970 it's one of these hyped ideas. 01:08.010 --> 01:11.690 We're going to dig into it in week two as we accelerate. 01:11.930 --> 01:15.690 And that will take us to week three, which I'm calling amplify. 01:16.010 --> 01:22.170 This is something which really allows you to dial up your business or your client's businesses or your 01:22.170 --> 01:24.610 product using multi-agent systems. 01:24.610 --> 01:26.970 We'll use MCP, another hype word. 01:27.090 --> 01:32.850 Uh, we're going to see the reality, the truth, where it's super impactful and where it's probably 01:32.850 --> 01:36.050 a red herring so that we can focus on business value. 01:36.090 --> 01:42.090 And by the end of week three, you will have automated, accelerated and amplified your business. 01:42.090 --> 01:45.490 And each of these three weeks have five days. 01:45.730 --> 01:51.410 Uh, and these are the five days in week one, shown as the day one, two, three, four, five, five 01:51.450 --> 01:52.090 boxes. 01:52.410 --> 01:56.650 And the first two are purple, which means that there are about core skills. 01:56.890 --> 01:59.130 And then we've got some integrations days. 01:59.130 --> 02:03.690 And then we end with a project, a business project that you're gonna love. 02:04.050 --> 02:06.690 The second week here come the five days. 02:06.750 --> 02:07.470 There they are. 02:07.510 --> 02:08.390 It's a purple day. 02:08.430 --> 02:09.910 A yellow day for integrations. 02:09.910 --> 02:16.230 Another purple day when we go into rag, a yellow day with super bass, and then the voice agent that 02:16.230 --> 02:20.190 we deliver at the end of week two and week three. 02:20.430 --> 02:25.910 Week three is starts with a purple as we self-host n810 and we use local models. 02:26.070 --> 02:28.150 Then we have some advanced integrations. 02:28.470 --> 02:35.350 Then we go back to purple for MCP and then Context engineering Subagents hot topics right now in Atlantic 02:35.350 --> 02:35.670 AI. 02:35.870 --> 02:42.630 And then we conclude with the capstone project to amplify your business at the end of week three. 02:42.830 --> 02:49.150 And after all of that, you will be able to to take this celebration moment, get your certificate and 02:49.150 --> 02:54.910 your your, your cup, your chalice and be able to say, I am now an agentic AI builder. 02:54.950 --> 02:58.870 Okay, that's it for our objectives and our course description. 02:58.870 --> 03:01.190 It's time for us to actually get going. 03:01.350 --> 03:05.590 I'm going to start with a little bit of theory stuff, and then we're going to go back to action. 03:05.590 --> 03:06.630 Back to N810. 03:06.790 --> 03:08.550 Back to building some workflows. 03:08.550 --> 03:14.280 So first up just so we're on the same page, I have some sort of basic definitions, some foundational 03:14.280 --> 03:14.720 stuff. 03:14.920 --> 03:19.720 I imagine most of you are very comfortable with this already, but I think it's good to make sure that 03:19.720 --> 03:25.680 we we agree on this, and maybe I'll try and add in some some juicy extras to keep it interesting. 03:25.720 --> 03:31.760 We should start with a definition of an LM, because otherwise I'm just going to be using this assuming 03:31.760 --> 03:33.000 that you know all about this. 03:33.360 --> 03:37.440 So, you know, there's lots of ways to to define what an LM is. 03:37.480 --> 03:42.720 It's basically it's the sort of core model, the data science model that's at the heart of what we call 03:42.720 --> 03:43.840 generative AI. 03:44.240 --> 03:48.080 It's an AI program that it's able to generate text. 03:48.120 --> 03:50.600 What it does is spits out text. 03:50.600 --> 03:56.280 And typically that text is something which is either answering a question, or maybe that text represents 03:56.320 --> 03:58.360 an action that should be taken. 03:58.360 --> 04:00.360 If the question is, what action should I do next? 04:00.400 --> 04:02.080 It's the answer to that question. 04:02.080 --> 04:04.480 It's just generating content. 04:04.520 --> 04:08.600 It's given some sort of an input and it's generating output. 04:08.800 --> 04:16.250 Now all of us are guilty, myself included, of sometimes treating llms large language models in an 04:16.250 --> 04:20.330 almost human like way, giving it human like behaviors. 04:20.530 --> 04:23.450 Thinking of it as if we are making some kind of connection. 04:23.490 --> 04:26.770 Putting motivations and intentions behind what's going on. 04:26.970 --> 04:28.690 A chat that we're having with it. 04:28.690 --> 04:32.050 And it's always important to stay grounded in the reality. 04:32.530 --> 04:38.650 Llms are very powerful, very effective statistical programs. 04:38.650 --> 04:41.890 They are pattern matches on a massive scale. 04:42.050 --> 04:47.450 They have been given tons and tons of data taken from the internet, taken from all sorts of sources. 04:47.450 --> 04:49.130 They've analyzed the patterns. 04:49.130 --> 04:50.250 We call it training. 04:50.450 --> 04:56.490 They have been built so that when they are given an input sequence of some text, they are very effective 04:56.490 --> 05:00.690 at predicting the most likely text to come next. 05:00.690 --> 05:05.330 And if that text is some sort of a question, they are likely to predict a good answer. 05:05.330 --> 05:10.210 If it's if it's a request for what should I do next, then it will predict actions. 05:10.210 --> 05:12.490 It can predict tools to be called. 05:12.490 --> 05:14.090 It can have a conversation. 05:14.330 --> 05:21.310 And all of this is built around really effective pattern matching, really effective statistics. 05:21.350 --> 05:25.430 And we always have to keep that in mind as we work with large language models. 05:25.430 --> 05:32.430 And there's a whole bag of tricks and techniques that we use around large language models to give this 05:32.470 --> 05:35.510 impression that you're in a conversation with them. 05:35.510 --> 05:41.150 There are tricks like this, this approach called memory, for making it seem as if the LLM is able 05:41.150 --> 05:45.030 to remember the conversation, keep, keep the the flow of it. 05:45.030 --> 05:48.990 But it's all done with, with, with clever little techniques. 05:48.990 --> 05:56.030 It is simply a pattern matcher, a statistical engine, that's all it is behind the scenes. 05:56.230 --> 06:03.030 And it's somewhat bewildering and astonishing that as a result, purely as a result of the level of 06:03.030 --> 06:11.110 scale that these llms are run at, that they they have this apparent intelligence and it is just a byproduct 06:11.110 --> 06:11.950 of scale. 06:11.990 --> 06:14.310 People like to call it emergent intelligence. 06:14.750 --> 06:19.880 And if this is something that you're still puzzled by, I've got a bunch of videos on YouTube that try 06:19.880 --> 06:25.960 and take you through the journey of coming to grips with how it is that just by virtue of the trillions 06:25.960 --> 06:32.640 of these parameters, these internal weights within a modern LLM, we're able to achieve this illusion 06:32.640 --> 06:34.040 of intelligence. 06:34.240 --> 06:38.640 And as a final point, I'll make here that something that most of you probably are very clear on. 06:38.640 --> 06:46.320 But some people do get, get tripped up by is the distinction between ChatGPT the product and GPT the 06:46.320 --> 06:47.040 model. 06:47.200 --> 06:49.840 GPT is a model, a large language model. 06:49.840 --> 06:51.720 It is the statistical engine. 06:51.720 --> 06:54.960 It is something which takes an input sequence and gives an output sequence. 06:54.960 --> 07:00.160 It's what we would call stateless, which means that every time that you make a call to it, every time 07:00.160 --> 07:03.560 that you invoke GPT, it's completely fresh. 07:03.560 --> 07:09.040 It has no knowledge of what came before, and it's just given a new input and it gives an output that 07:09.040 --> 07:10.000 is the LLM. 07:10.040 --> 07:19.080 The large language model called GPT, uh and ChatGPT is a software product also written by OpenAI. 07:19.240 --> 07:22.340 They built the model The GPT LLM. 07:22.460 --> 07:23.500 And then they thought, you know what? 07:23.500 --> 07:27.380 We're going to make a product that uses this in the most effective way. 07:27.380 --> 07:28.860 That's going to appeal to people. 07:28.860 --> 07:29.860 It's got to have a memory. 07:29.900 --> 07:31.580 It's got to be able to do other things. 07:31.620 --> 07:35.020 In fact, nowadays it can also search the web and do all sorts of stuff. 07:35.020 --> 07:39.620 But this was all built and written with software, with with normal software. 07:39.620 --> 07:45.460 And it uses GPT, the LLM to to add in this apparent intelligence. 07:45.540 --> 07:53.460 So keeping in your mind that separation between software, the ChatGPT the product and the LLM, GPT, 07:53.500 --> 07:59.540 which is the underlying model, the AI which is stateless and which gets an input and gives an output, 07:59.540 --> 08:04.260 that's an important distinction to have in your mind the product versus the model. 08:04.420 --> 08:09.740 And indeed, what we were doing in the instant gratification earlier is we were building a product that 08:09.740 --> 08:15.900 used a model sitting with open router, and we were able to add a little chat functionality on top of 08:15.900 --> 08:16.660 the model. 08:16.660 --> 08:19.940 And that's why it's important to keep keep that difference in your mind. 08:19.980 --> 08:21.940 And if you knew all this already then hang in there. 08:21.980 --> 08:23.540 We'll we'll cover a lot more. 08:23.540 --> 08:25.030 And if you didn't then. 08:25.030 --> 08:27.310 And if you weren't, if you've got questions, you're not entirely sure. 08:27.310 --> 08:30.430 You could also hang in there because we're going to cover so much of this and you're going to this. 08:30.430 --> 08:32.910 This is going to be old hat to you in no time. 08:33.310 --> 08:37.950 The other thing that I just wanted to mention, again, probably people are super familiar with this, 08:37.950 --> 08:40.110 just the concept of an API. 08:40.230 --> 08:44.710 People talk about APIs all the time, and you may be thinking, I've never I've never really been sure 08:44.710 --> 08:46.870 what an API actually is. 08:47.110 --> 08:48.510 And there's good reason for that. 08:48.510 --> 08:50.830 That API is loosely defined. 08:50.950 --> 08:58.110 It's generally you could think of it as simply just a way to connect different applications and services, 08:58.110 --> 09:05.590 different software products, typically using common standards, using using some typical techniques 09:05.590 --> 09:09.950 for connecting together different bits of software in a way that's going to make it relatively easy 09:09.950 --> 09:11.350 to plug them together. 09:11.750 --> 09:14.270 And years ago, it used to be really, really hard. 09:14.310 --> 09:19.830 Even even with APIs, it was still incredibly challenging to connect different software products. 09:19.870 --> 09:25.030 Nowadays, it's much easier because there are such common standards, but it's still always a challenge. 09:25.030 --> 09:29.280 Connecting things is always a challenge, and that's the challenge that N810 has made. 09:29.320 --> 09:30.560 Unbelievably simple. 09:30.840 --> 09:34.480 But that's that's at the heart of APIs connecting different systems. 09:34.520 --> 09:40.680 And while you can use the term API in many contexts, it's most often used to describe connecting different 09:40.680 --> 09:44.360 applications using a web connection, a web interface. 09:44.400 --> 09:49.680 Typically you're using HTTP, the same technology we use to collect web pages. 09:49.680 --> 09:56.360 It's an HTTP connection that you're making the URL, the web page that you're connecting to in order 09:56.360 --> 09:58.760 to make your request to another piece of software. 09:58.800 --> 10:00.640 It's sometimes called the endpoint. 10:00.680 --> 10:05.200 The endpoint is what you hear people calling the URL, which you're not using to collect a web page. 10:05.200 --> 10:09.240 You're using to actually call another piece of software to tell it to do something. 10:09.240 --> 10:15.640 That's that's called the endpoint, an HTTP endpoint, just a web, a web URL that you are calling out 10:15.640 --> 10:22.600 to and the data that you typically exchange using the most common kinds of APIs is this format called 10:22.600 --> 10:26.040 JSON stands for JavaScript Object Notation. 10:26.080 --> 10:31.260 You've probably seen JSON many times before, and if you haven't, then you will get somewhat used to 10:31.300 --> 10:32.620 it on this course. 10:32.620 --> 10:39.180 This is definitely like like whilst it is perhaps code, it is the lowest of the low code that is as 10:39.340 --> 10:40.580 sophisticated as it's going to get. 10:40.580 --> 10:44.660 We are going to get pretty used to looking at JSON if you're not already. 10:45.140 --> 10:52.020 And it's also worth mentioning that typically with APIs you need to have some kind of password to say, 10:52.020 --> 10:57.140 I want to use this API and I am who I say I am, and if you've got my credit card, then you should. 10:57.140 --> 10:58.620 You should build this against me. 10:58.620 --> 11:03.380 And a common way of doing it there are there are different techniques, but a simple and common one 11:03.380 --> 11:09.900 is to use an API key, which is just a password that you have that identifies you to a third party like 11:09.940 --> 11:10.740 open router. 11:10.940 --> 11:16.740 And when you make your call to open router, you pass in the API key as a way of saying, hey, it's 11:16.740 --> 11:17.020 me. 11:17.020 --> 11:19.300 I set up my account on you, you've got my credit card. 11:19.340 --> 11:23.700 Or maybe you don't in the case of Open Router, but you might have my credit card and and this is me 11:23.700 --> 11:25.820 and please do this request for me. 11:25.820 --> 11:28.020 And it responds with the answer. 11:28.020 --> 11:30.060 That's what an API key is all about. 11:30.060 --> 11:33.820 It's a password to use a third party API service.