214 | How to create Agentic applications, in minutes with no experience with Shubham Saboo

0:00Hello and welcome to another episode of the Leveraging AI podcast. the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host, and I've got a really fun, unique, and interesting episode for you today. Now, we're going to touch on a lot more advanced stuff that we don't usually dive into in this podcast, but the idea is exactly that, is to take stuff that sounds more advanced and really break it down and simplify it in a way that you'll be able to understand it, and more importantly, you'll be able to use it for your benefit in your personal life and in your business.

0:34So we're going to talk about what are agents, how are agents different than large language models, how to dip your toe into the advanced agentic world without knowing how to code, without knowing really advanced things. Even more interesting is how you can get access into hundreds of agents that somebody else already built, and you can literally grab them and make them your own without paying anybody anything. And we're going to take away the fear from terminology such as code and GitHub repositories or repos or Python and a lot of other scary terms that you probably hear thrown around when people talk about advanced agents.

1:08And you're like, I don't know what it is and I'm not sure I want to know. Well, you'll see today that the monster is not a big hairy monster. These are actually very easy and very accessible to anyone. And the most important thing is we're going to really take stuff that sounds like really advanced and that really only developers can have access to and show you how you can do this on your own. And we're going to do this with the help of Shubham Sabu, who is incredible. Shubham, if you don't follow him on LinkedIn, you should. He's the founder of Unwind AI, which is an AI education platform in the areas of AI agents and also large language models and RAG applications and stuff like that.

1:44And he shares incredibly well-defined tutorials on LinkedIn on how to perform different things. And we're just going to dive into one of them today and go through this process in order to explain to you, again, stuff that usually sounds really advanced and scary, but make it very accessible to anyone. I don't write code, I don't understand code, and yet I love his content, and I'm sure you're going to love it too. So I'm really excited, A, about this episode, and B, humbled to have Shabu on the show today. Welcome to Leveraging AI.

2:14In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

2:54Thank you, Isar. Thank you for having me. It's a great pleasure to be on this show. And today we are going to do something really, really interesting. Today we are going to look at the code repository, the GitHub repository with 100 plus code examples of AI agents, multi-agent, agent teams, drive applications. Things that look really, really complex. Things that have real business value. These are the applications that can do some amazing things. And we will take one example and we will try to break it down for you in a step-by-step manner on how you can take something from the GitHub repository that already exists.

3:24So some context, some quick context, what I've done in the last eight to nine months is put together an open source repository of 100 plus AI agents, RAG, and LLM apps that anybody can use as a template, as an MVP, as a prototype, and build on top of that. So whatever use case you have, you might find something useful in that. And the good part with that is that is supplemented by Unwind AI. So Unwind AI is a full-fledged AI education platform where we have a newsletter that goes out every week, which covers everything that you need to know about to stay up to date in the AI agent and rag world.

3:59Along with that, all the tutorials, all the code that might look scary that you would find in the HTML and Apps repo is explained in super simple language in Unwind AI with step-by-step tutorials. So everything that you would find in open source repo has a step-by-step tutorial. If you want to understand, if you're confused about code, if you don't know what this code means, you can roll back to the NYDI.com and find a tutorial that corresponds to that code example. So everything is there. The idea was to create super accessible, practical knowledge that people can use to build something quickly and see it work in action rather than wondering and reading for days to make it happen and going through multiple tutorials.

4:39So if you don't mind, we can quickly dive in and look at what the awesome LLM apps and what they look like. Yeah, so let's do that. I think if we can start by explaining to people what is a GitHub repository, I think most of my audience is not technical and it sounds like Chinese to them. Let's explain what that is and how to get to those and then we can dive into the details of the example. Sounds good. So a GitHub repository, if you have to simply, very simply understand in the simplest language is a place where you will store code. think of it as a database where the entire code of the world is stored like it is a standard all the companies google microsoft all the startups they use this to store their code so as the name goes it is a repository of code and that's why when i said you'll find 100 plus code examples that's where i store and what it does is it indexes everything and make sure you have a standard where you have a link where anybody in the world can go and access those assets or those content so all you need to do is go to github.com or Shibam Shabu, awesome LLM app sharepo.

5:39Of course, you can share the links and everything in the description. All you have to do is just go and click there and as I share my screen, I can really clearly explain the steps, the simple steps and everything that once you go to that link, you will see a detailed read me with the instructions on how to clone that repo, how to download the code, how to use that code and how to interrupt that. Everything is broken down in two to three simple steps that's all you have to follow. Okay, so let's dive in. Awesome. Now those of you, as Shubham is sharing his screen, those of you who can watch this on YouTube, I recommend you do that.

6:16And we will share the link to the YouTube channel in the show notes so you can do it from wherever it is that you're listening to the podcast. But if you're driving or walking your dog or running on a treadmill or whatever it is that you're doing and you cannot watch YouTube, we will explain everything that's on the screen as we're looking at it. Awesome. So this is the awesome LLM apps repo. I launched this eight months ago, and this ended up becoming the top trending number one GitHub repo. Multiple times it's trended on GitHub. I'm lucky to have around 57 ,000 stars and 7 ,000 press forms. So what does that mean is it has been cloned and used by people almost 7 ,000 times.

6:53And here, if you scroll down, you'll see all these examples. So what we have done is categorize these examples into simple and easy to understand categories. So if you're starting out with AI agents and wondering what AI agents can do, I would recommend you just go with the starter agents. You can pick any of these examples. Let's see, if you click on this AI travel agent example, every example comes with a detailed readme that you would see here. So it explains what the application is, what are its features, how to get started. Every example will have detailed step-by-step instruction on how to get started all the way to running this example.

7:27Then if you go deep dive into details, you'll see how it works and other features and stuff like that. So let's go back for a second. If you see this repo, it has all these 100 best examples. And now if you're wondering, how do I actually use this stuff? What should I do about it? Okay, it looks good. So three simple steps. Clone this repo. Use your terminal to go to any of the example demo that you would want. So these are all the different folders that you have. just go to that folder, install the requirements, and just run it. And the instructions to run it are, again, in each of these folders, each of these demos.

8:05And the way you clone this repo is just click here. You can download it. At the same time, you can just copy this. So let's try to quickly do it in cursor and really see it in action. But even before that, I just want to show you Unwind AI and what I meant by that. So this is the Unwind AI education platform that we have. You'll find a lot of content here. All the newsletters that go out, paste, they archive here in the panel of blog posts. You can just open any newsletter. If you're not subscribed, if you missed, you can just feed through it here on this platform. It is completely free. All the resources that I've created that I share is 100 % open source, completely free.

8:44So you don't have to pay a single penny for that. You can read this here. And then if you go in this AI tutorials category, you'll find all these tutorial blog posts. Let's quickly open this. it in TickRag with OpenAI GP5, which just launched recently, you'll find all the details, like what are the features, what the application is about, what you need for setup, what the application flow would be like, what would be the technical architecture, and step-by-step it explains the entire code. And if you're not interested in knowing the code details, what you do is just go clone the repo, clone it in your favorite IDE, whatever you use, Terser VS Code, it would not matter.

9:19Let's see, I'll use Terser. I really like using Terser. It's my favorite. So let's quickly clone this repo in Cursor and see how this works. So for those of you who are not watching, you go on this page, there's a green code button. You click on that and then there's a little copy icon next to a link to the code. So you grab that and then you can go straight into Cursor or any other code editing tool. Again, if you need access to Cursor, just go and download Cursor. You don't need any fancy setup. You don't need to do anything. Just go to Cursor, download it for Mac or PC, and then when you're inside of it, and then I'll let Shubham continue from here.

9:58Awesome. Let me open my Cursor and share the screen. So we copied this. So this is what you will see once you download Cursor. You can create a new project, and here you'll see, just in the UI itself, you'll see an option, clone repo. and all you do is just place the link here and it will automatically, it will ask you for a destination to clone your repo and just select any destination and it has started cloning as you'll see here. Once it clones, you will see the folder appearing. It's a big repo that's taking some time to clone it. It has a lot of... Again, just to explain to people what's happening.

10:34We are, what it means to clone a repo is basically to take the code that's in that repository and make a copy that you can edit on your own. basically as if somebody will give you a Word document. You can grab the Word document. To open the Word document, you need Microsoft Word, right? Because otherwise you can't open it. So we're using a tool called Cursor in which we're duplicating that document. But what it has, instead of one document, it has usually a bunch of folders that you would see on the left. That's the structure of the code. And again, you don't need to know anything about it. But literally everything that was in the original code comes with it when you open it inside of Cursor.

11:08And again, to do this, it's a monkey can do it because there's a button that says import repo, and you click on that, and you drop in a link that you copied from Shubham's repository, and that's it. And now you have access to the entire code. Yeah, and once you have access to the code, so what cloning does, again, in simplest language, is just downloading the entire code from the GitHub repo to your computer. That's all that you did with the cloning. Once you've cloned, you'll see all these folders. Just go to the folder or demo that you're interested in. Let's take an example of AI travel agent.

11:37You'll see this lead me, and we will give you all the instructions. So we already did get clone. Next thing you do is copy this, paste it in your terminal. The next thing that you will do is install the requirements. What requirements mean is these are all Python libraries or packages. Think of them as dependencies that you would need to run this code. I've already had them installed, so I'm not going to do it this time. But all you have to do for this, all the commands are here. As I mentioned, anybody can do it. Even a monkey can do it. Just copy, paste. Copy, paste in your terminal. That's all you have to do.

12:08And now when we are - Again, again, just to explain to people what terminal means, because some people may be like, I don't know what that is. So inside of all these tools, you have a terminal window. There's different ways to open them. In cursor, it's on the bottom. You can pick different options. One of them is terminal. So you click on the terminal button, and then you just paste the instructions that exist in a very clear way. First do this, then do this. You just copy and paste, and it will install all the requirements for you. You don't need to know anything about what's actually happening behind the scenes.

12:37Yeah. And when you see these terminologies like Python, Streamlit, and Agno, and all these terms coming to you, don't get scared. These are super simple. Just go do a simple Google search. If you are curious, do a Google search. If you don't care, it's fine. It does not matter. You'll still be able to run the application. If you are curious, just do a Google search. It's super simple. Would be good for your conceptual understanding. What we do is Streamlit, again, in the simplest term, is a library, is a tool that lets you create UI in Python itself. So what we're going to do, we're going to do this in two steps.

13:06First, very simple code that you will find in my record. It already is in the form of Streamlit app. The entire code is in Python. So we are going to run the Streamlit app as is. That's the first step that we're going to do. And the next thing, the more advanced part, I'm going to walk you through how to take this code, put it in a white coding tool, and build a full-fledged, really nice-looking, sleek UI out of it. Again, you don't have to do anything. Just prompt it, prompt, just go to a tool called Lovable, prompt it, create an ICUI for this code. So you already got the recipe to build your next million-dollar startup in days, in matter of days.

13:40Just go to the repo, run the example that you like, run it locally, go to Levable, create a UI, have it all together, spend like a few hours, and then you have it ready. And then the third thing that you're going to look at is, okay, these are the example templates that I have created, and maybe you want something different, and you want to create something custom that you would like. And you're going to see how you can really create a custom agent, and today you're going to end with that. but let's quickly get back to what we are doing here. So streamlit run cloud agent following all these commands that you would see in readme.

14:12Let's run this. Once we run this, so what it is doing is opening an app in my terminal, sorry, in my browser. Okay. Give some errors because maybe I did not had a few packets installed. Let me quickly get back. So, Google search not installed. We get back to that. Sorry, it happens with the live demo sometimes. Oh, good. That's the fun of a live demo. Yeah. So, again, now we get to follow all the steps that we talked about. Pip install requirements.pxt. These are the commands. Are we going to follow? Yeah, all we're doing, by the way, in the background is copying and pasting step by step what was in the readme file.

14:59So you click on the readme file, and then there's literally one, do this, two, do this, three, do this. And most of them are just copying and pasting into the terminal. And after each step you hit enter, it's going to run a bunch of stuff. Once it ends, you can copy the next one and just keep going until you have all the different components. So this is the application that it has opened. Now, what it is asking for is two API keys. So this is the open API key. very quickly. We can get this type. Again, anything that you don't know, type on Google, 99 % of the times you'll just get it. Open API and it will take you to the API platform.

15:34You log in, get the API. Similarly, for SAP API, you just type. SAP API, what it is really doing, again, a tool that is giving you access to Google search. So you go here, you get the API key. But even if you don't want an API key, you can use some of the forms of search, like that. Just again, Again, to explain to people what an API key is, for those of you who don't never heard the term, it basically gives you access to the backend information of a platform through the API. And to make sure it's really you, you give it a unique ID. So if you want the OpenAI key, you're paying for tokens, right?

16:07You're going to consume data from OpenAI. So you're going to open an account, you're going to give it your credit card, and you're going to then create an API key that you can use for this particular tool. You can create as many keys as you want, so you can track how many tokens you're consuming. and don't be afraid you can limit the amount of tokens you're going to consume and the amount of money. So you can say, okay, I don't want to consume more than$5 a month on this thing and then it's going to get stuck. Usually if you pick the right AI version, then your consumption is going to be negligible.

16:35Like it's portions of sense for every time you do something. So you got to run huge volumes in order to actually consume significant amount of money. Okay. And this is your AI travel banner agent. And we asked if we told him, like, where do we want to go? how many days where you can change it generated by doing the google search in the background so you'll see day one day two day three blah blah blah things that you can do in austin where you can stay what are the activities what are the highlights stuff like that this is a very small and simplest example of what you can do the limits are endless like it's just what you can imagine like you can actually build whatever you can imagine this also comes up with this like cool handy feature where you can just download this itinerary as calendar invite save it to your iCal and stuff like that.

17:19And it just like works out of the box. So we just saw how you can take again to recap what we saw. And again, it's really significant what you just saw, because now you can do this for 100 plus AI agents and RAG apps. We went to the GitHub repo, we simply cloned it, you can use the ID that you would want, we used cursor for this. And then we just followed the command, which included installing the dependencies, Python dependencies, the different tools that is required, streamlit, ignore all these frameworks. and then just running a simple command streamlet run travel agent.py and then we get this app out of the box.

17:55It did not take us what is it? We spent five minutes, seven minutes? That's with me slowing you down in the middle. So, yeah. So, that's what I wanted to convey, like the significance of this. I mean, going back, like here you see all these applications. You can do it for all these applications. The next thing that I promised you that I'm going to tell you is just take the code that we have. So let's say if I go to this AI travel agent, or you can do it in your ID version as well. You go here, you just copy the code here. Once you copy the code, you go to white coding app. Again, you can choose the white coding app that you like, but here we use Lovable.

18:32We just copy the code here and ask it to create a nice UI for this app. That's all you did. Like you just prompted, it took two minutes. I already did it before, it took two minutes and created this nice UI. So again, to explain to those of you who are not watching, What we had, the original application works. Like it had a line to put in a field to put in the destination and two other fields. And then you got all the information about the travel and the planning and whatever. But the user interface is really ugly and very basic. It's just square, you know, gray squares with very plain text. You drop the same exact code, not knowing anything about what the code does.

19:08You literally copy and paste it into a tool like Lovable or like Replit or like Base44 or like any one of those. And the prompt was as simple as it can be, create a nice user interface for this app, like nothing. That's all we wrote. And now it looks like Expedia, right? It has a nice background. It has beautiful buttons. The input fields are a lot nicer. Like all of that was created out of thin air, still using the rest of the code in the background, right? So we still don't need to understand what the repository does or how the code works. It just adds a much nicer user interface to the application that Chubham already shared in his repository.

19:45Awesome. So now you have this working code, you have the backend code, you have the UI, and now you can replicate this for WordPress applications. And you can do it for really complex ones. So let's get back to the awesome LLM apps and look at some realistic examples, right, which would really help. So here you would see an AI legal agent team. There's a team of multiple agents acting as an automated AI legal team. And what you can do is you can take this code, follow the same process, run it, get a feel of how this works. Once it is working, go to Lovable, create a UI, merge UI with the backend, and you have production-ready product.

20:20And again, I want to go back to Lovable and show another cool feature that you can do, which is an extension of this. Let's say you like this UI, everything works as this. If you don't like it, you can continue prompting here in this text box. But if you like it, you can just go and click publish. And it will just publish on this website that anybody in the world could access. So you went from idea to deployment in mere minutes. So here is the link. And now I can just open this app that I created. And it's a perfectly deployed app on this website that you see here. It's deployed live on internet.

20:55So this is the second thing that I wanted to talk about, tell you guys about. We looked at the backend code, Streamlit app, and now the frontend. Now the last thing that I want to touch upon, which is even more interesting, is how you can build your own custom agents without knowing anything about the code itself, right? So you know about all these LLM platforms like OpenAI. You all know about ChatGPT. You know about Gemini Cloud. Just use the one that you like. I always go with Gemini when it comes to deep. If there's a reason for that, you'll come to that. So this is the awesome LLM app structure.

21:27We'll keep coming back to this. We'll do a source of everything. You just copy this link, the URL, and then you go to this tool called gitingest.com. So again, you take a step back and explain what it is really doing. It is ingesting your GitHub repository and creating an output or creating an LLM friendly output. Think of this as converting your repository into a prompt that LLM can understand. So now what I'm doing is packaging the entire content or all the content from my awesome LLM apps repo and converting it into an LLM prompt and giving it to the model. I'm going to give it to the model.

22:03All you have to do for that is just paste the app link here. click on ingest and it will create the summary, directory, files, content, everything that you had in that report. You just click on copy all and then you go to your favorite model. So here I've used Gemini and again this is the bigger form that creates like thousands and thousands of lines of code but again we did not do anything here. We just copy pasted the one that we have got from our tool called git ingest. So on some level map URL, paste it git ingest, click on ingest, copy from here so all we are doing is copy pasting there and there we are not relating into any code right again a monkey can do it so you copy this from here pasted the entire thing in gemini and the prompt i prompt that i wrote is super simple it is really basic what i wanted to do and i don't have this kind of app anywhere in my report this is like really super custom i prompted it based on recent events create a new agent app and stream it to analyze the u.s data situation with different countries.

23:02And it will go on, it will think through it, and create the three files. Now, the interesting thing about this that you will see, is following the same structure that we have in our repo. A readme file, a requirement, and the app.twek. A readme file, a requirement, and the app.twek. That's what you see in all these examples. Any examples that you open, open any examples. A readme file, application file, and requirement. So it automatically understood the structure of the repo and created the files in those structures. Let's read the reading. Look at the reading. You have a set of analysis agent.

23:37These are the features with all the instructions. So you also even get the instructions to run it. These are the instructions to run it when you take the code and put it in cursor or whatever. Requirements. These are the requirements that you need or Python packages that you need to install. And this is the application code. Super handy, super straightforward, created this set of analyst agent, provided it with the tool ducted codes, again, a search endpoint. And then what you could do. Now, from here, you have two options. You can just copy this code, paste it back into your cursor. But kudos to Google.