hi welcome to Joe's Diner I'm Jackie how can I help you today hey Jackie how are you today I'm doing great thanks for asking how can I help you today yeah I wanted to see if you could tell me the uh prices for T-bone steak at Jes Diner of course our T-bone steak is 500 gr and costs $40 is there anything else I can help you with today oh yeah I wanted to see if I can make a reservation absolutely I can help you with that what day and time are you looking to make a reservation for and how many people will be in your party I want to see if I can make a reservation for December 20th for six people at 2: p. m. and then my name is z of course okay I have a reservation for you on December 20th for six people at 2 p.
m. your name is zuar is there anything else I can help you with uh no I think that's that's all do I get um a reservation confirmation or do I assume that the uh reservation has been confirmed you got it the reservation is confirmed you'll receive a confirmation email within the next few minutes is there anything else I can help you with nope that's all thank you so much you're very welcome I hope you have a great time at Joe's Diner enjoy your meal all right so that was the live demo of a real voice agent that I created using 11 labs and n8n so in this video I'm going to show you exactly how I made this and how you can create one of these realistic voice agents that you can have natural humanlike conversation and the beauty of this is that you can embed this into your website for whatever business you are and you can actually train this voice agent with a knowledge base you can add additional tools to it so I'm going to show you how to connect this 11 Labs voice agent as a backend via web hook to our NAD and AI agent which is going to have logic and therefore be able to communicate and independently make a decision like making a reservation and storing all that data in a Google sheet or other tools that we can add to it this is part of the new deep dive series I'm doing for my school Community called voice AI agents which is a six-part tutorial on how to build several voice agents step by step including an entire introduction where I go and explain every single aspect of this 11 Labs AI agent I'm going to put the link in the description for my schoool community make sure you join if you want to learn how to build this and a lot more with this tool in NN if you're new to the channel my name is bear my school Community my YouTube channel is all about building amazingly useful AI agents that's going to be for businesses for personal use and for someone who is looking to take advantage of this AI Revolution start their own business sell these AI agents that we create we have a great group of like-minded individuals from all over the world who are ready to connect collaborate on projects and learn together whether you're a beginner or an advanced learner you will have plenty of things to learn in this classroom all of my automations all of the templates live in this classroom section including Deep dive topics like vector databases AI agents voice AI agents and much much more we have daily calls in the calendar where you can jump in ask about tech support we have business strategy class with Alex who's one of my aigc business partners on Wednesdays and Fridays so make sure you jump in and ask any question related to business all right so let's get back to the video so if you don't already have an 11lb account make sure you go ahead and sign up for an account once you come to your account you're going to go to the app section so if I go back let me go to the homepage for 11 Labs so once you sign in you're going to end up here you're going to go to your conversational on left hand side right here it says conversational AI you will click on agents and this is where you can go ahead and create your own agents when you click on create an agent this is going to give you several options there's the blank template which is basically you'll start the agent from scratch there's support agent this will have pre-populated uh prompts and system prompt and it will also have um some already selected voices but you can go ahead and start with a blank template you can name your agent um actually no we'll startop with a sport template you can name your agent so you can just say customer support and then you'll click on create agent agent and this will go ahead and initiate the process of creating your agents so just a quick introduction here there's several different tabs here there's the agent tab The Voice tab analysis security Advanced widgets we're going to focus mostly on the agent and the voice tab predominantly in the agents tab so this is where you're going to go ahead and design your agent so you can select the language it has access to several languages which is great so all of these uh agents could be created in um several different languages this is going to be the first message so by default if you select the support agent it's going to say hi I'm Eric how can I help you so you can change this system prompt it's just by default because it's a support agent it says you're a support agent named Eric you're very friendly enthusiastic and you will answer 3 OR7 sentences in most cases so very very simple system prompt so we're going to go ahead and talk up in details about how uh all these things work the knowledge base the tools because the tools is where the magic lies and that's how we're going to connect our uh voice agent to our nadn via that tools with web hook all right so let's go ahead and actually test this real quickly so that way I show you how you can just get started right away so as soon as you come in by default you can click on the test AI agent here so this will pop up this little circular widget where you can actually click on call AI agent so as soon as you doic how can I help you today this will initiate the microphone or initiate the conversation and then it will keep listening to you so you can actually talk to it back and forth and you can actually stop it while it's talking so it's very intuitive but anyway so let's go ahead and let me show you exactly how I built this reservation restaurant reservation a agent again the user is the same thing but I'm going to walk through step by step on how I did this here um so that way you have an idea on how to do this for yourself uh before actually moving forward so the great thing about this uh voice agent is actually if you head over to the widget section you can see right here you can just very easily copy this code and embed it into your website or if you're building this for a client you can embed it in your client website very very simple it's just a copy paste code and on the bottom as you can see so this is what it's going to look like in real uh life for a website when this is incorporated you can make some changes here as far as the look and feel is concerned so let's say if I want to change uh the color of this Avatar or you wanted to let's say add an image you can certainly do that add your own image you can add a link or you can just stick with this orb you can actually change the colors so let's say if I want to change uh the color here to something different let me see here let's say we want to change this to Blue as you can see right there it changes the color so lots of different customizability as far as the look and feel is concerned but for now we're just going to leave it as it is but feel the your play around here and you can change a bunch of other stuff too you can change the appearance the button color the background color so there's a lot you could do and then also the start call so right now it says start call but you can type something different and this is going to change that as well right so there's a lot of things that you could do and once you do that once you make your changes you just click on Save and it gets Incorporated but I'm going to press on clear because I want to keep it as standard all right so now let's go ahead and stick to our agent tab so right here as I mention the first thing is setting the language we'll leave it as English the first message so for this particular demo I created this customer service rep for um this restaurant called Joe's Diner so I just basically put this first message prompt because this is how it will react or this is how it will greet your customers it says hi welcome to Joe's Diner I'm Jackie how going to help you today right system prompts so as you saw in the first one by default it has this simple s system prompt but for me I just added a few more things uh to make sure that this agent is aware of what kind of Persona we're giving and that's what the system prompt is so if you're familiar with uh the AI agents in nadn then you uh understand how system prompt works so for example if I just go ahead and double click on this agent so this is the system message this is exactly equivalent to the system prompt here you're basically giving it you're giving this agent a Persona as you can see right here and you're telling it how to behave if it has access to any tools it can you can um instruct it to use utilize certain tools for uh certain functionalities whenever that's needed uh but anyway so the system prompt again I just said um you know you're a support name agent named Jackie you're very friendly enthusiastic uh and then you answer 37 sentences in most cases you work at a restaurant called Joe's Diner if a customer is asking for information about the diner make sure to retrieve that information from your knowledge base only so let's talk about the knowledge base here this is what you can do to customize this so there's this um actually before we get down there let's quickly take a look at a few other things as well so after the system prompt there's I'll come back to the system prompt um after the system prompt there's the llm you can choose the large language model here the great thing about this is you have options to select several large language models right now as you can see it says currently the llm cost is covered by us in the future this cost will be passed on to you so right now you can use any of these for free uh but in in the future they're going to start to pass this on uh the users so make sure you play around right now with it depending on what you're building to make sure that it meets your needs so by default it has chosen uh it starts with the Gemini 1. 5 flash by Google and the reason why that is because it's the fasted when it fastest when it comes to interaction the great thing about Gemini 1.
1 flash is that it actually is fast and it also has a tool use so um it actually becomes very handy especially when you're adding additional tool to this or if you're reaching out to a third party app like nadn it becomes very useful to to use a model that has the capability of utilizing or using tools obviously you have your standard gp4 mini the gp4s uh the uh Claud uh so there's lots of options available and you can also add a custom llm again I'm going to be building a lot of these voice agents I have already built like three four as part of that deep dive Series in the classroom where I utilize different models so make sure you check that out but in the future I'm going to actually start to build more voice agents because this tool is incredible and I've been playing with this for the past uh couple of weeks since it came out and it it's blowing my mind how good it is but anyway so you select your um model I'm going to leave it as GPT or as Gemini 1. 5 flash temperature so this is the parameter that controls again the uh creativity the randomness of the responses that's generated by this large model so make sure uh you for standard just when you're starting just leave it as in the middle right there but you can play around with that limit token usage this is the configure to configure the maximum token that you want to allow the llm but um but for now just leave it as negative 1 because that's the standard um and you don't want to change that as you're starting and then down the line when you build more complex voice agents you can keep playing with those different parameters so knowledge based this is where now you can start start to really customize this voice agent if you're familiar with rag it's kind of a similar concept um because you can actually give this um uh AI agent or voice AI agent a knowledge base similar to if you have created a custom GPT you can upload documentation and give that uh custom GPT a knowledge base where you can then instructed to utilize the knowledge base it has been given to answer and interact with the customers or with the users so this is pretty much the same exact concept right so the knowledge base could be given in several different options if I click on add item you can add a file up to 21 megabytes it could be PDF text STS and HTML or you can actually add a URL of a website so let's say You're Building this voice agent for a particular website you can actually add that website URL right here and you'll be able to scrape that and grab that information but to be on the safe site you always want to upload um a document uh so that way it has access to that you can also manually enter things so you can add a text name and then you can add whatever you want in the content here and you can just add that item and you can save it so for me for this particular build I put Joe's Diner FAQ so I created this sample let me pull that up real quickly for you so I created this sample FAQ document about this you know uh Joe's Diner restaurant which is again I just use chat GPT to create these things so uploaded that to his knowledge base so that way when somebody is interacting with this thing you can ask it question and it'll actually go ahead and retrieve these exact questions it's very good actually so test it out and make sure uh you check out how good it is um and then same thing on the menu items so during the demo if you notice I asked it what's the price of a T-bone steak it said 500 G and it's $40 so that's basically coming directly from this menu items and prices and the way I did this is I added um that basically from this text so I just gave it the text name menu and I just copy and pasted a bunch of random like menu items and prices which again it says Jo Diner menu uh you know filet steak filet steak Rabbi steak T-bone steak right here there you go right T-bone steak 500 gr $40 so it's able to retrieve that information directly from its knowledge base because in the system prompt we told it to do exactly that right so the tool section this is where this the magic of this voice agent comes in where now you can attach third party tools to this and in our case we're attaching our um nadn AI agent to give this uh even more logic so that way it reaches out to our third party in our case NN AI agent that can have the ability to make independent decision and our AI agent in naden has tools attached to it and it can make independent decision based on whatever the court wor or whatever the user is asking it to and it can make a decision on which tools to use and therefore retrieve that information send that information back to our voice AI agent via the web hook so that's exactly what we're going to take a look at here so if you see on top here we set on the system prompt that when a customer is ready to make a reservation use the reservation tool so this reservation tool is this right here so you have to make sure that whatever uh name you're given here it matches exactly the tool here so that way uh this voice AI agent knows when to reach out to this tool to retrieve that information so let's go ahead and take a look at what this tool is uh so you can add multiple tools you can add a tool by clicking on ADD tool here and there's several option there's two options there's the web hook and client uh for nadn the easiest way is obviously to use the web hook because you can essentially uh directly add um your web hook tra URL right here and that's exactly what I did so let's go ahead and take a look at um my tool already here so again it's the same thing I just added that tool so here the name I gave it a reservation uh name because again that's what we called it on our system prompt the description I said use this tool to make a reservation for Joe's Diner if you're familiar with um again agents inside NN um you can instruct the agent to use a particular tool that's attached to it and in our case we have this Google sheet attached to it so inside this AI agent on the system prompt right I said Google Sheets use this tool to store reservation details this is very typical way to instruct your AI agent inside naden to use all the tools that it has access to to make a decision so it's pretty much the same exact thing here we're telling uh the llm here to use this tool to make a reservation for you know Joe's Diner in this case or if you're putting a tool or if you're doing building this for something else you want to instruct the L the llm when to use this particular tool that you've created so that's the description of that tool the method because again we're using this web hook here so we're using this tool to reach out to our nadn to grab the information um or store information right we're doing one of these things so either um you're sending information or you retrieving information so with the web hook the URL right here is basically the URL that's going to be your web hook trigger so if I double click on this as you can see right here so this I am uh using a test URL but you can change this to production URL but this is exactly what I copied I named this uh path to the reservation and as you can see right here it's basically the exact same URL um with with test because again I wanted to Showcase um the demo but if you want to build this for production make sure you switch that to live and change and remove this test right here but anyway so yeah you put the URL for your web hook the next few things that are here are the parameters so there's body parameters query parameters path and header if you are defining header that will be sent as a request to your external tool in our case you would go ahead and Define that here but in our case we're just sending this reservation details right so the uh name of the person the time the date and the number of people that's um want to make this reservation we're sending this data as body parameters to our nadn AI agent where our nadn AI agent then has access to this Google sheet where it stores whatever information is being sent from this voice agent right in our case it's being sent the name date time and number of people so when we send that information via the the body parameters here to our nadn instance this web hook grabs that information and as you can see right here let me pull this a little bit on this side right here in the bottom if I scroll down here's the body parameter that's being sent right December 20th 2 p. m.
zuar and the number of people are six and that's exactly during the demo when I started this video I was interacting with that voice agent Jackie and it grabbed that information and sent it to this web hook because of the fact that we had this tool here attached and we identified exactly what information to retrieve from the customer during this interaction and how to send that so as you can see right here it says required required required that's exactly what the voice agent is going to understand that okay during this conversation if I'm reaching out to this tool I need to make sure I grab this required information because I need to send this via this web hook back to a third party right so that's what we're doing so the way to identify or the way to define these parameters is uh it is going to require a body parameter the body parameter is going to be the required for any web hook um but you can send these other optional parameters as well like the query parameters header and stuff like that and again on my the other ones that I did doing this uh Deep dive so I created this um more complicated version this Advan uh advanced voice AI support and in this one I send actually multiple tools and I send um multiple parameters including query and then additional body parameters but for this particular one I'm just basically sending only body ERS so you will Define uh how to uh retrieve this information during the conversation so on the description I'm saying in a friendly way ask the customer to provide their name date of reservation time of reservation and the number of people and now I have to Define these uh different um things that I put here like the name date uh time of reservation number of people on the bottom here as these different properties right so the way you add a property is you just come down like you just click on ADD property and you can go ahead and identify the data type so this one is because it's the number of people for reservation it's going to be a number obviously you put the identifier and this identifier is basically this uh uh right here on the body because you need to send this exactly uh to our external um NN agent via that web hook so that's what's happening here it's grabbing that uh number directly from here as body parameter so this will match right here because this is the data that's coming from here so the next thing is date same thing we're sending the date here and then we're sending time and we're sending the name of the person who's making the reservation and on the description you just simply expl um put that together all right so you'll click on Save changes once you do that um the next thing is Secrets don't worry about secrets for now this is if you're building more complex uh more complex voice agents that's when you would use that uh and then the next stage is the voice itself so you can clone a voice or you can use existing voices that 11 lab has uh that you have access to so you have a variety of choices right so things like this is probably the most popular one this bill as we are liberated from our own fear our presence automatically liberates others so Jessica that's the if you spend your whole life waiting for the storm you'll never enjoy the sunshine so that's what we selected right because um our assistant's named Jackie so we S I selected that voice but that's up to you you can select whatever you want so use Flash so this is for um new recommended models by 11 Labs but for our case because we're using uh uh Google's Gemini 1.
