RAG Ai Agents made simple (no code)

In this video, we're going to be making a RAG AI agent. And if you don't know what that means, RAG stands for Retrieval Augmented Generation. And then a RAG agent is basically just an agent that's able to look through a bunch of vectorized data and answer questions based on that data. So a lot of the time, companies will use this where they can have all of their company docs, you know, all the information about their company, the procedures, the type of font they use, onboarding processes, everything like that. And that can all be added to a vector store, and then we can link an AI agent to that vector store it can answer questions based on that data. You haven't seen me before, my name is Rory Ridgers and I'm an AI enthusiast and I'm passionate about helping business owners integrate AI into their business. So that's something you might be interested in, then I've left a link to my website in the description and you could go ahead and book in a call with me and we could talk about some ways to integrate AI into your business. Also, I just want to mention that the templates that I use in this video will be available for completely free in the description. So if you go into the description, you should see a link that says something like Free Templates, and you can click on that and just get all the templates for free. So let's not waste any more time, let's get into it. So in this video, we're going to be using Relevance AI, make.com, and then also the Assistance API by OpenAI. And the reason we're using Assistance API is because we're actually going to be using this as our vector store for this video. A lot of the time people use things like Pinecone for their vector stores, but I found that Assistance API just make it way easier to have these vector stores and then also way easier to retrieve data that are in these vector stores. So I thought I'd do a little bit of a simpler tutorial and use Assistants APIs. I think if you're just a beginner doing this, it is a lot easier to do it this way. So first of all, I want to show you this scenario on make.com, and what this does basically is it takes your files from your Google Drive folder and it puts that into the vector store on Assistants API, because you don't want to have to put all your files into Assistants API every time you have a new file. You want to just be able to put it into your Google Drive folder and then that gets transferred over automatically into your VectorSource. So it's a really simple scenario to be able to do that. All we have is a Google Drive Watch Files in Folder trigger node. So this is just going to trigger when the selected folder has a new file in it. So for this one, it's Rack Demo. That's just a demo folder I put together for this video. So when a new file was put in there, it will trigger this scenario and then we're simply just going to download the file. So we use a Google Drive download file, put in the file ID that we got from the node before here, then we use an OpenAI upload a file node and literally just click the Google Drive download a file. So the step before, what this is going to do is literally just going to download that file from the Google Drive onto our Assistants API account, set the purpose to assistance, and then add files to vector store. I've put together a vector store as you can see here, just called RAG demo again. So all we do is that you just put that file ID that we got from the step before. We save that to the RAG demo vector store. And the difference between just having files and having vector stores on Assistants API is a nice way to think about it is just the vector stores are kind of just like folders and then you can attach these folders to AI agents. It just makes it a little bit easier. So that's why we upload it as a file and then put it into a vector file. So I'm going to give you a quick demo of how this might look. We can just add a file quickly. So I've just added a file and now I've just made up some example PDFs. I did actually use these in the last video, so apologies for being lazy, but it's basically just sort of 5 PDFs that a digital marketing agency might have. It's sort of a design style guide with their fonts and their color palette, things like that, onboarding processes, that kind of thing. So we've added the file. Let's look at the Make.com scenario. Obviously this would be on automatic, but for this we're going to run it manually. So I said run, it's recognized this file. So yeah, you can see, yeah, Design Style Guide Template, and then it's downloaded the file, uploaded that to our Assistants API account, and then it's uploaded that to the Vector Store. So perfect. Now we should see it inside of our Vector Store. Might have to refresh it. Okay, yeah, you can see we've got two. Obviously I already had this in here, so We've got two of them now, so just get rid of that. But yeah, that's how that works. Really, really simple. So every single time that you add something to a specific Google Drive folder, it's instantly going to get put into this vector store on your Assistants API account. So really cool. Now I'm going to show you inside of Relevance AI this really simple RAG tool. And literally all this is, is you have one input here, which is the— I've just called it knowledge base query. So it's literally just going to be what the user wants to find out from the data. So as I've said before, things like what font do we use, or what is our process for this, what's the procedure for that, that kind of thing. That's just going to go straight into this input here. And then all we have is just one step, which is an API step where we link to another Make.com scenario. So I'll show you that quickly. So this is a scenario on Make.com. So really, really simple. We just have the webhook here, and then we have an OpenAI message and assistant node. And the reason we've done this is because if we come over to Assistance API, go to Assistance, all I did was I just made an assistant called RAG. So I'll show you it. So assistant's called RAG. The only instructions are simply search the vector store data you've been given to accurately answer the user's request. Really, really simple prompt. Then we've basically just given it the file search function and attached that vector store to it. So now this agent here, or this assistant, basically has the vector store uploaded to it so it can answer any questions on the files in that vector store. And what we want to do is we want to be able to communicate with this agent outside of Assistants API platform. So the way we do that is we just set up a webhook on Make and then message an assistant node. And you can see the assistant we've selected, RAG, roles user. The message is literally just what comes through on the webhook. Leave all of these blank. The tool choice is going to be file search. This just means that the assistant or the agent, it has to search for files before answers. So we're basically just making sure that it's actually looking through the database to find our answers and not just making stuff up. And then we just use a webhook response, basically just send that back to Relevance AI. So you can see in here we just have the result from this step, we send that back over to Relevance AI, and yeah, that is literally all the tool is. So I'll quickly show you If you don't know how to set up these webhooks, how it works, what you do, you set up an API step. Always put the method as POST because we're going to be sending data. And then the URL is going to be this URL. When you put a webhook down, you'll get this URL here. You literally just copy that, put it in there. In this case, I've just set the body to raw. And then in the— what we're actually sending over, I've literally just put the query input. So that's what's going to be sent over. So they're really, really simple to set up, these webhooks. Just a heads up, if it's your first time doing this, you might need to run the step and then hit redetermine data structure a couple times for it to work. You might have to play around with it, but eventually it will connect, and then from there you're good to go. So really, really simple tools here. You can hit save, and now I'll show you how we put all of this into an agent on Relevance AI. And you might be wondering, why don't we just talk to this agent on Assistance API? And the reason for that is simply just because I don't know about you guys, but for me, all of my agents that I use are on Relevance AI, and I don't want to deviate and go to different platforms to talk to different agents. I want one space where I can talk to all of them. So that's why I'm connecting it up to a Relevance AI agent. And then also I can connect this agent up to my personal assistant agent that I made in my last video. So a multi-agent system. So I can just add this agent on to my personal assistant agent, and I can have them all on one platform, being able to talk to all of them in one place. It just makes a life a lot easier. So that's why I've done it that way. So I'm going to show you the prompt and the tools that we have for this agent. If we come on in, the agent name, just RAG Agent. The agent description, agent that can retrieve info on company docs. The tools is obviously going to be the RAG tool. I'm not sure why it's on here 3 times. We can get rid of 2 of them. Set this to auto run. So it's obviously going to be that RAG tool that we just used. And then the core instructions, keeping it quite simple for this one because there's not actually much it has to do. It literally just has to take the user's query and then send it off to the tool, and then the tool does the work. So all I've said, you're helpful assistant that can retrieve data from company documents using Make.com. The task is you'll receive a query about something that could be answered from company docs. This could be, what font do we use? Or, what's our client onboarding process? You'll then simply pass this query straight to the RAG tool that will search through company docs to find the answer and then relay it back to the user as straightforward and as simply as possible. I'll give you a breakdown of the tool now. And then I've given it a little breakdown of the tool here. All I've said is this tool allows you to search through company docs to find the answer to their question. The only input is query, and in here should go the query from the user. And then I've just sort of given it a little bit of context on how the tool works. The way this all works is it connects to a Make.com scenario, which connects to a fine-tuned agent that can easily access company documents. So Really, really simple. You can hit Save Changes and then we can test this agent out. So if we look at our vector store here, you can look at some of the data that's in this store and sort of decide what we want to ask the agent. So SEO best practices, common client FAQs, website launch checklist, the client onboarding checklist. Let's say we're going to ask something about the client onboarding checklist. So what is our onboarding process. So something like, what is our onboarding process? We should see that it uses the only tool it has, the RAG tool, and then that will set off a Make.com scenario which will trigger a fine-tune assistant on Assistants API, and it's going to send it all back through, through the webhook, back to our Relevance AI agent. So it sounds like a really long process, but as you can see, it only took a few seconds. So pretty quick. And this is what it's given us: initial consultation, contract and agreement, asset collection, project road back, assigned tasks. Obviously, this is all from the PDFs. And we can ask anything else. We can say something like, what is our color palette? So I've said, what is our color palette? And it should come back to us with some hex codes that is found in our company docs. So we go again, that was pretty quick, maybe 5 seconds that one. And this is what's come back to us, these hex codes. So yeah, I mean, perfect. That's really, really cool. And I think it's a much simpler way of doing it rather than using some of these other vector stores that can be a little bit tricky to set up. On Assistants API, they make it really, really easy for you. And they also make it really, really easy on Make where you can literally just use these OpenAI nodes that can download files and automatically add these files to vector stores that you make. It just makes the process so much easier. So if you're a slightly smaller company that doesn't have, you know, thousands and thousands of docs, if you've just got a few hundred docs or maybe even a thousand docs and you don't want to have to search through them all the time in your Google Drive, this method will be perfect for you because you can just ask the AI agent anything about your docs and the Assistants API, the vector store, won't get overloaded or anything. So yeah, it's a perfect way of doing it. So not a very long video today, guys, but I just wanted to show you this really, really simple method to create these RAG agents because they are super useful for businesses. Because I know how stressful it can be when you have all of these documents and you're just trying to find that one thing, you can't find it, and you spend ages searching for all these documents. Something like this, a simple system, just will make your life so much easier. So like I mentioned at the start of the video, if you are looking to implement some AI systems like this into your business, then just check out my website in the description. You can book in a free consultation call with me and we can just go through some ways that we can integrate AI into your business to help you. Also, all the templates that I use in this video, as I said, will be available for completely free. Go grab them in description. You should see free templates or something like that and a link. Click that, you can get all the templates for free. Also, if you guys have any video suggestions you'd like to see, please do just drop them in the comments and I'll try my best to get around to doing them. But that's it for this video, so thanks a lot for watching.