WEBVTT 00:00.200 --> 00:01.520 Welcome to this tutorial. 00:01.680 --> 00:07.560 Today I'm going to show you how to build an AI voice agent that can answer questions using information 00:07.560 --> 00:08.600 from documents. 00:09.080 --> 00:15.440 So basically you'll be building a RAC agent, but as a voice agent, so you can use it for things like 00:15.480 --> 00:22.800 customer support or even low ticket sales as a widget on your website that visitors can interact with. 00:23.560 --> 00:29.960 We will be using 11 labs and a relatively new conversational AI feature to create our voice agent, 00:30.160 --> 00:32.240 and this is our setup. 00:32.720 --> 00:35.840 So we'll be using Bitcoin as a vector database. 00:36.040 --> 00:39.960 And of course an attempt to integrate everything as a use case. 00:39.960 --> 00:46.240 We are going to build this a voice agent for customer support for a solar panel installer company. 00:46.760 --> 00:53.720 So just to show you how this works with a real world example, I basically scrape information from this 00:53.720 --> 01:00.380 website, save it into two files in Google Drive, and then vectorize it to pinecone using All right. 01:00.420 --> 01:03.780 I'm going to do a quick demo and talk to a voice agent. 01:03.820 --> 01:05.740 We'll be building together in this lesson. 01:06.020 --> 01:12.100 So I'm going to ask is a question based on information from the company's website, which I scraped 01:12.100 --> 01:18.860 and stored in a Google doc inside a Google Drive folder, and then vectorize to create a knowledge base 01:18.860 --> 01:20.060 for our voice agent. 01:20.660 --> 01:22.980 First, I'm going to hit this workflow. 01:23.260 --> 01:24.340 You can see the flow. 01:29.260 --> 01:31.140 Hey, I'm Rob, your solar energy guide. 01:31.140 --> 01:32.420 How can I assist you today? 01:33.540 --> 01:37.420 What are the benefits of adding solar battery storage to my solar panel system? 01:38.260 --> 01:41.180 Let me check our information about battery storage benefits. 01:48.420 --> 01:53.940 Based on our data, battery storage lets you save money by storing excess daytime energy for nighttime 01:53.940 --> 01:56.540 use, reducing your dependence on the grid. 01:56.580 --> 02:01.480 You'll have backup power during outages, which is really handy for keeping your home running smoothly. 02:01.600 --> 02:06.720 Plus, you'll be helping the environment by using more renewable energy and reducing your carbon footprint. 02:09.840 --> 02:10.480 Perfect. 02:10.920 --> 02:13.960 So now I'm going to show you how to do it step by step. 02:14.520 --> 02:15.640 So let's get started. 02:15.960 --> 02:22.240 First, let's actually prepare the content we want to use as a knowledge base for our RAC agent. 02:22.280 --> 02:26.920 So go to your Google Drive account and create a new folder. 02:29.240 --> 02:31.960 I'm going to name mine demo. 02:33.280 --> 02:34.080 Confirm. 02:35.920 --> 02:39.920 Then go in and create a new file. 02:42.920 --> 02:45.080 I'm going to create a Google doc. 02:47.560 --> 02:51.160 Let's name it Solar panels. 02:54.760 --> 02:55.480 Installer. 02:58.980 --> 02:59.620 Confirm. 02:59.940 --> 03:02.060 Then go to the website you want to scrape. 03:04.660 --> 03:07.780 I'm going to grab information from this website. 03:07.780 --> 03:11.220 Just right click open in edit mode. 03:13.020 --> 03:16.460 And now you can see all the information available on this website. 03:16.500 --> 03:22.580 Now copy the information you want to add to the knowledge base for your rack agent such as services, 03:23.300 --> 03:28.860 opening hours, and so on and paste it to the document you have created in Google Drive. 03:29.100 --> 03:34.820 All right, now I've got all the necessary data I want to fit into my customer support agent. 03:35.460 --> 03:41.260 Actually, I'm going to create a second document with general information about this company to show 03:41.260 --> 03:44.020 you how to use group of items. 03:44.300 --> 03:45.460 Note in a ten. 03:45.980 --> 03:47.660 So let's create a new document. 03:49.660 --> 03:50.740 Let's call it. 03:54.160 --> 03:57.760 General Mason. 04:00.040 --> 04:04.160 And let's paste the general information about this company from their website. 04:04.440 --> 04:04.840 Alright. 04:04.880 --> 04:05.560 Perfect. 04:08.400 --> 04:15.240 Now we've got two files to documents in the demo folder for this project. 04:15.280 --> 04:17.160 General information and solar panels. 04:17.160 --> 04:17.920 Installer. 04:18.360 --> 04:21.200 Actually, let's rename it to services. 04:21.200 --> 04:21.680 Because. 04:22.040 --> 04:27.240 Because most of the information is about services, they offer services. 04:29.640 --> 04:29.920 Alright. 04:29.920 --> 04:30.520 Perfect. 04:32.120 --> 04:39.520 Now let's jump into an A-10 and build this part of the workflow to vectorize these files, this document 04:40.120 --> 04:41.840 into the database. 04:43.400 --> 04:47.440 So then our rack agent will be able to retrieve this information from the vector store. 04:49.440 --> 04:50.960 All right so let's start building. 04:51.040 --> 04:52.440 So let's create a workflow. 04:54.300 --> 04:56.540 Now let's add the first step. 04:58.140 --> 05:00.340 So so on chat message. 05:06.700 --> 05:12.540 Now let's add Google Drive note to search the folder we create in our Google Drive account. 05:15.780 --> 05:16.900 To search for Google Drive. 05:21.900 --> 05:24.820 And what we want to do we want to search files and folders. 05:26.380 --> 05:27.860 So choose this one as the action. 05:31.820 --> 05:33.820 Now connect your Google Drive account. 05:36.780 --> 05:38.020 Mine is already connected. 05:40.340 --> 05:44.540 As a result of file slash folder operation search. 05:45.620 --> 05:46.220 Search method. 05:46.220 --> 05:49.380 We want to search file folder name. 05:50.300 --> 05:50.460 No. 05:50.460 --> 05:51.380 Advanced search. 05:51.560 --> 05:53.600 Now the search query is empty. 05:55.520 --> 05:58.080 Let's limit to 20. 05:59.920 --> 06:05.360 This basically means that this is the max number of results to return as filter. 06:05.640 --> 06:09.920 Let that folder and let's select folder from list. 06:11.920 --> 06:13.040 This is the easiest way. 06:13.560 --> 06:17.480 Then let's search for the folder you just created in Grid Drive. 06:17.720 --> 06:22.840 I called my demo, but as you can see I have to refresh my credentials. 06:23.520 --> 06:24.440 So let's do it now. 06:29.160 --> 06:30.160 Let's say Google. 06:30.560 --> 06:32.840 I'm going to choose my Gmail account. 06:32.880 --> 06:39.040 I'm going to select all the permissions to give cloud access to all the files in my Google Drive. 06:39.240 --> 06:39.480 All right. 06:39.480 --> 06:40.640 Connection successful. 06:43.560 --> 06:45.600 Now let's go back to our node. 06:49.200 --> 06:51.100 Let's search for them all. 06:55.340 --> 06:58.260 And as you can see, the folder we have created is pop up. 06:59.180 --> 07:00.260 So let's choose this one. 07:02.100 --> 07:03.340 And that's it for this note. 07:04.820 --> 07:05.660 Let's move on. 07:05.780 --> 07:07.860 Actually I should use a different trigger. 07:08.060 --> 07:13.220 So instead of when Chatmessage received node I should use when you click on the workflow. 07:13.460 --> 07:14.580 So let's remove this one. 07:16.020 --> 07:19.740 Let's add new trigger and trigger manually. 07:20.060 --> 07:23.500 So it runs the flow on clicking a button in N810. 07:23.660 --> 07:29.740 So every time when I click on Test Workflow, it will vectorize the data to our vector store. 07:30.580 --> 07:32.340 So let's choose trigger manually. 07:34.780 --> 07:36.740 And let's connect it to our Google Drive node. 07:39.420 --> 07:39.980 Perfect. 07:40.300 --> 07:43.100 Now let's pull in the information from our two files. 07:43.100 --> 07:45.540 So let's give it a test. 07:45.580 --> 07:46.540 So test step. 07:48.880 --> 07:51.440 And now, as you can see, not executed successfully. 07:52.240 --> 07:56.400 And you have two items as an output with different IDs and names. 07:56.560 --> 08:02.640 So solar panels, installer services and general information as we just created a Google Drive. 08:04.800 --> 08:05.240 All right. 08:05.240 --> 08:05.720 Awesome. 08:05.880 --> 08:08.480 So now let's actually download these files. 08:09.480 --> 08:10.760 So then we can vectorize it. 08:10.760 --> 08:12.160 So our vector database. 08:12.760 --> 08:16.720 So the next node is going to be we drive. 08:18.640 --> 08:19.800 So again Google Drive. 08:20.120 --> 08:23.240 But this time we are going to download file. 08:24.120 --> 08:24.920 So click on it. 08:26.040 --> 08:27.400 And now we have ID. 08:28.240 --> 08:31.520 So our Google Drive account is already connected as resource. 08:31.520 --> 08:32.920 We want to leave it as file. 08:33.960 --> 08:38.400 We want to download the file for different use cases. 08:38.400 --> 08:41.800 You can switch to delete copy move and so on. 08:43.200 --> 08:48.460 But for this scenario you want to download the file and you want to button file fill it. 08:48.500 --> 08:50.860 We want to change it instead of from list. 08:50.860 --> 08:53.540 We want to download the file by ID. 08:54.420 --> 09:01.820 So now we can simply drag and drop this ID we just got from the previous node and drop it right here. 09:03.540 --> 09:04.500 Alright perfect. 09:04.820 --> 09:06.580 So now this is dynamic variable. 09:06.580 --> 09:11.540 So every new document which will be added to our Google Drive will be easily accessible by ID. 09:11.740 --> 09:16.780 And this node will be able to download every new document from Google Drive for us. 09:16.780 --> 09:17.820 So now let's move on. 09:19.580 --> 09:19.860 Alright. 09:19.860 --> 09:20.580 Perfect. 09:21.980 --> 09:25.300 Now the next step is to loop over each file we downloaded. 09:25.580 --> 09:28.580 So this is important because we are not just working with one file. 09:28.980 --> 09:33.780 We might have several documents, so we need to process them one by one. 09:33.980 --> 09:37.740 Embed the content and send it to pinecone individually. 09:38.220 --> 09:40.940 So let's add a loop over items node. 09:41.860 --> 09:45.100 So click on this plus button for loop. 09:46.920 --> 09:47.760 and click on it. 09:48.760 --> 09:51.000 You don't need to change the batch size here. 09:51.720 --> 09:52.600 Just close it. 09:53.800 --> 09:54.120 Alright. 09:54.120 --> 09:54.680 Perfect. 09:55.040 --> 10:00.920 So yeah, there is no simple way to add these files at once to our vector database. 10:01.080 --> 10:03.200 That's why we need to loop over each item. 10:03.520 --> 10:05.200 So let's remove this replace me node. 10:08.520 --> 10:10.000 And also delete this connection. 10:10.240 --> 10:16.280 Now the next step is to load these files into Pinecone Vector store. 10:16.720 --> 10:18.880 So let's add pinecone vector store node. 10:19.280 --> 10:21.960 So click on this plus next to the loop. 10:24.160 --> 10:25.360 And search for pinecone. 10:28.120 --> 10:28.960 Select this one. 10:30.440 --> 10:33.440 And what we want to do we want to add documents to Vector store. 10:35.920 --> 10:36.320 Alright. 10:36.320 --> 10:36.880 Perfect. 10:37.080 --> 10:42.880 Now I'll show you how to create an account with pinecone and explained how the pinecone vector database 10:42.880 --> 10:47.260 and vector store work in one of the previous tutorials when we're building a RAC agent. 10:47.260 --> 10:50.940 But in this tutorial I'm going to remind you how to create new credentials. 10:50.940 --> 10:55.340 So connect to your Bitcoin account to N810 and also how to create vector store. 10:55.380 --> 10:57.340 So let's create a credential. 10:58.660 --> 11:00.180 Now as you can see we need API key. 11:00.220 --> 11:02.140 So let's go to the Bitcoin website. 11:02.140 --> 11:04.540 If you already have an account please login. 11:04.780 --> 11:06.020 Otherwise sign up. 11:06.020 --> 11:09.340 Then in the left sidebar click on API keys. 11:12.260 --> 11:15.420 Create API key and give it a name. 11:18.220 --> 11:19.660 Let's say solar panels. 11:22.180 --> 11:22.740 Demo. 11:24.940 --> 11:25.780 Create key. 11:28.180 --> 11:31.500 Then copy it close. 11:32.260 --> 11:41.500 Go back to LinkedIn and paste it here hit save and now your credentials should be successfully created. 11:43.040 --> 11:44.040 So let's close it. 11:46.440 --> 11:46.800 All right. 11:46.800 --> 11:47.360 Perfect. 11:48.000 --> 11:53.560 All right, so once you connect it to your Bitcoin account to N810, in this step we are setting the 11:53.560 --> 11:59.920 operation to insert documents because we want to load all our embedded data into pinecone. 11:59.920 --> 12:01.920 So that's why you can see here pinecone Index. 12:02.440 --> 12:05.640 We are uploading this into a specific index inside pinecone. 12:05.920 --> 12:12.320 So now let's go back to pinecone and create an index so you can find all the indexes under database. 12:12.720 --> 12:14.120 So just click on indexes. 12:14.120 --> 12:16.600 And here you can create a new index. 12:16.960 --> 12:18.120 So create index. 12:19.440 --> 12:20.280 Let's scroll down. 12:22.760 --> 12:25.920 Now you want to use text embedding three small model. 12:26.160 --> 12:29.480 It's one of the most cost effective models offered by OpenAI. 12:29.640 --> 12:34.880 And later we'll be using the same model as embedding model in our workflow. 12:34.920 --> 12:35.960 So let's choose this one. 12:39.840 --> 12:40.700 Let's scroll down. 12:43.060 --> 12:45.860 You can leave cloud provider as an AWS. 12:47.820 --> 12:49.540 You can change the region if you want to. 12:52.740 --> 12:53.140 All right. 12:53.140 --> 12:53.620 Great. 12:56.860 --> 12:57.900 Now let's give it a name. 12:58.980 --> 13:00.260 I'm going to name mine. 13:02.020 --> 13:02.420 Um. 13:02.900 --> 13:03.820 Solar panels. 13:06.860 --> 13:07.380 Demo. 13:08.060 --> 13:08.500 Create. 13:08.500 --> 13:09.060 Index. 13:14.820 --> 13:15.260 All right. 13:15.300 --> 13:17.340 Now as you can see, our index is active. 13:17.540 --> 13:20.580 Now there is also this option called namespaces. 13:20.580 --> 13:24.300 So think of it like a sub folder inside your index. 13:24.420 --> 13:27.620 So it helps you keep different sets of documents separate. 13:27.620 --> 13:29.060 So let's click on it. 13:29.260 --> 13:36.980 And now when I go back to our workflow and and try to retrieve index from the list. 13:38.580 --> 13:43.920 As you can see, the solar panels, uh, demo index pop up such as this one. 13:45.360 --> 13:47.920 And also I want to add a namespace. 13:49.360 --> 13:50.720 So Python namespace. 13:51.360 --> 13:56.080 So basically this partition the records in an index into namespaces. 13:56.080 --> 14:00.120 So queries and other operations are then limited to one namespace. 14:00.360 --> 14:03.520 So different requests can search different subsets of your index. 14:04.120 --> 14:11.640 So let's name it solar panels docs. 14:13.400 --> 14:14.000 Perfect. 14:14.400 --> 14:21.360 Now before we test the step to vectorize our files into the Python vector store, we need to add an 14:21.400 --> 14:22.200 embedding model. 14:22.840 --> 14:23.960 So let's go back here. 14:26.080 --> 14:27.480 Now let's add embedding model. 14:28.960 --> 14:31.120 We want to use embeddings from OpenAI. 14:34.560 --> 14:36.680 Make sure the OpenAI account is connected. 14:36.680 --> 14:36.860 It. 14:37.260 --> 14:41.220 And as I mentioned earlier, we need to use the same embedding model. 14:41.700 --> 14:43.500 So text embedding is model. 14:44.140 --> 14:48.100 There is also text embedding and text bedding 002. 14:48.140 --> 14:49.540 But we want to stick with this one. 14:50.780 --> 14:51.500 Close it. 14:52.820 --> 14:53.660 Alright great. 14:55.180 --> 14:56.340 Now it's document loader. 14:56.340 --> 14:59.100 We want to use default data loader. 14:59.100 --> 15:01.540 So load data from previous step in the workflow. 15:04.300 --> 15:07.420 Now one thing you need to pay attention to is we. 15:07.420 --> 15:13.380 We don't want to load data in JSON format because we want to download the files from Google Drive and 15:13.380 --> 15:17.980 then load it into Python vector store in binary format. 15:18.420 --> 15:21.100 So to process binary data from previous step in the workflow. 15:21.500 --> 15:22.620 So switch to binary. 15:23.340 --> 15:29.420 Now in mode we want to load our input data and we want to automatically detect by Mime type. 15:29.780 --> 15:31.740 And that's all we need to change in this node. 15:32.620 --> 15:33.660 So that's good to go. 15:35.380 --> 15:38.200 Now the last step, we want to add the text splitter. 15:39.320 --> 15:44.600 And we are going to use recursive character text splitter, which splits text into chunks by characters 15:44.600 --> 15:45.320 recursively. 15:45.320 --> 15:47.800 And it's more recommended for most use cases, as you can see. 15:49.440 --> 15:54.240 And we are going to leave chunk size to 1000 and don't change the chunk overlap. 15:55.440 --> 15:56.400 Alright, perfect. 15:58.320 --> 16:00.160 Now you can actually save the workflow. 16:03.160 --> 16:09.080 And now when I hit Test workflow, the two files I've created in my Google Drive should be downloaded 16:09.080 --> 16:11.160 and then vectorized to my pickle vector store. 16:11.200 --> 16:12.360 So let's see it in action. 16:20.200 --> 16:21.000 Alright, perfect. 16:21.000 --> 16:22.480 So workflow executed successfully. 16:22.480 --> 16:24.720 Now let's go back to my pickle vector store. 16:25.320 --> 16:29.200 And now as you can see the new namespace has been created solar panels docks. 16:29.200 --> 16:30.480 And I have nine records. 16:30.520 --> 16:31.840 Now when I go to browser. 16:34.700 --> 16:36.940 I can see all my records right here. 16:40.140 --> 16:42.500 So basically these two files. 16:43.260 --> 16:49.660 So solar panels, installer services and general information have been vectorized into nine separate 16:50.060 --> 16:50.780 records. 16:51.140 --> 16:51.500 Alright. 16:51.500 --> 16:51.660 Great. 16:51.700 --> 16:53.740 Now let's go back to our workflow. 16:55.340 --> 16:55.700 All right. 16:55.740 --> 16:56.180 Great. 16:56.220 --> 16:58.780 You almost finished this part of the workflow for now. 16:58.980 --> 17:01.540 It's going to vectorize just one document. 17:01.540 --> 17:03.620 So we need to put it into the loop. 17:04.460 --> 17:10.300 So let's grab this plus and connect it to the loop node. 17:10.300 --> 17:15.420 So this way the vectorizing process will be repeated for every document in our Google Drive folder. 17:16.380 --> 17:16.700 Alright. 17:16.740 --> 17:17.380 Perfect. 17:17.700 --> 17:18.620 Let's hit save. 17:19.100 --> 17:24.220 Now imagine you've got hundreds of documents, or maybe you are working on a new project. 17:24.220 --> 17:25.300 Maybe for a client. 17:25.700 --> 17:32.800 They've got tons of documents like PDFs, technical manuals, service brochures, installation guides. 17:33.240 --> 17:40.120 So now you know how to set it up, and you can easily add new documents to your Google folder and just 17:40.120 --> 17:47.040 click on Test Workflow, and all the files will be automatically vectorized for you into the vector 17:47.040 --> 17:47.440 store. 17:48.160 --> 17:49.520 So it saves you a lot of time. 17:50.520 --> 17:55.760 And to be honest, from here building the voice agent on top of that is actually the easy part. 17:56.880 --> 17:57.720 So let's move on. 17:57.960 --> 17:58.720 All right. 17:58.760 --> 18:00.720 Now we are switching over to 11 labs. 18:01.160 --> 18:07.280 If you don't have an account yet, you can use the link we provide in the course setups lesson. 18:07.760 --> 18:09.440 You might get some special offers. 18:09.840 --> 18:10.960 So just simply sign up. 18:11.480 --> 18:18.000 Once you are in head over to the conversational AI in the products section. 18:20.120 --> 18:21.560 Then go to agents. 18:25.040 --> 18:28.400 And here we are going to create our voice agent. 18:29.040 --> 18:30.440 So click on New agent. 18:33.140 --> 18:38.900 Now you can use pre-built template if you wish to, but I'm going to show you how to build this from 18:38.900 --> 18:39.340 scratch. 18:39.540 --> 18:41.940 So we are going to use a blank template. 18:41.940 --> 18:45.460 So click on that create agent. 18:46.900 --> 18:48.580 Actually you have to provide a name. 18:49.780 --> 18:55.260 So let's call it solar panels advisor. 18:57.500 --> 18:58.500 Create agent. 19:01.020 --> 19:01.380 All right. 19:01.380 --> 19:03.260 Here we have to configure our agent. 19:03.980 --> 19:06.700 So first of all here you can change the language. 19:07.780 --> 19:08.860 Now let's scroll down. 19:10.540 --> 19:15.580 The first message the agent will say if empty agent will wait for the user to start the conversation. 19:16.100 --> 19:21.820 So the first message I wanted to say something like, hey, I'm Rob. 19:21.860 --> 19:23.180 You saw an edge guide. 19:23.180 --> 19:24.420 How can I assist you today? 19:27.700 --> 19:28.740 Now this is very important. 19:28.740 --> 19:30.880 We need to add a system prompt. 19:31.240 --> 19:36.240 So this is used to determine the persona of the agent and the context of the conversation. 19:37.240 --> 19:41.040 And here you can go to learn more and read the documentation. 19:44.000 --> 19:45.200 So let's actually have a look. 19:45.840 --> 19:52.400 So by reading this article you can learn about best practices of prompting and learn how to engineer 19:52.440 --> 19:55.000 a lifelike, engaging conversational agents. 19:55.760 --> 19:56.760 We have an overview. 19:59.680 --> 20:00.920 We have some framework. 20:06.560 --> 20:07.720 I think is worth reading. 20:11.720 --> 20:13.920 But I have prepared the prompt for this setup. 20:13.920 --> 20:16.840 So I'm going back to my agent. 20:18.880 --> 20:22.120 Right now I'm going to paste the prompt. 20:22.120 --> 20:27.940 So I want it to act as a customer support assistant for solar panels. 20:28.260 --> 20:32.900 And if you charge the installations, provide clear and accurate information about energy. 20:32.900 --> 20:34.940 Concerns about energy concerns. 20:34.980 --> 20:40.420 Services including options pricing and installation, use the solar installers. 20:40.620 --> 20:45.820 So this is the tool we are going to create later on this configuration page. 20:46.380 --> 20:51.940 So this is basically the way how N810 communicates with 11 laps via webhook. 20:53.180 --> 20:58.340 So use the N810 solar installers to query the database for details on solar panels, battery storage 20:59.500 --> 21:01.540 chargers, and installation procedures. 21:01.580 --> 21:05.100 Respond with concise, friendly answer based on the retrieved data. 21:05.140 --> 21:11.540 Maintain a friendly and approachable tone using natural filters like to keep the conversation engaging. 21:11.820 --> 21:13.700 Avoid repeating the user's question. 21:13.860 --> 21:17.380 Provide a clear and direct response with the necessary information. 21:17.700 --> 21:19.180 Max for sentences. 21:19.660 --> 21:20.620 Alright, perfect. 21:20.660 --> 21:23.340 Of course, feel free to change this system prompt. 21:23.500 --> 21:26.020 You can use it as a framework, as a template. 21:26.320 --> 21:31.880 you can always expand it and modify it to suit your needs and use case. 21:33.760 --> 21:34.120 All right. 21:34.120 --> 21:34.400 Great. 21:34.440 --> 21:35.400 Now let's move on. 21:35.600 --> 21:37.120 Here you can change the LM. 21:39.120 --> 21:42.240 So by default we have Gemini 2.0 flash. 21:42.840 --> 21:50.840 But you can switch to for example Gemini two Gemini 1.5 flash which is a little bit more cost effective, 21:51.040 --> 21:52.400 or to GPT four mini. 21:52.680 --> 21:59.200 Now I tested this setup multiple times with different large language models, and I strongly recommend 21:59.200 --> 22:00.200 you to use. 22:04.360 --> 22:06.920 Cloud 3.5 v2. 22:09.400 --> 22:13.720 Sometimes when I, for example, was using the default model. 22:13.720 --> 22:15.880 So Gemini 2.0 flash. 22:16.520 --> 22:20.680 Even though I had everything set up correctly, the webhook connection didn't work. 22:20.680 --> 22:24.860 I couldn't get any data from 11 labs into my editing workflow. 22:25.620 --> 22:28.060 So I strongly recommend you to use this model. 22:29.260 --> 22:37.020 Of course, you can test other models as well if you wish to, but I really like this model from anthropic. 22:37.060 --> 22:43.660 And as you can see, currently the LM cost is covered by 11 labs and in the future this cost will be 22:43.660 --> 22:45.100 passed on to us. 22:47.660 --> 22:49.500 Now here you can control the temperature. 22:49.500 --> 22:56.660 So this is the parameter that controls the creativity or randomness of the responses generated by the 22:56.660 --> 22:57.180 LM. 22:57.940 --> 22:59.500 I'm going to leave it as it is. 23:00.940 --> 23:04.780 Here you can configure the maximum number of tokens that the LM can predict. 23:05.660 --> 23:06.860 Let's leave it as it is. 23:07.860 --> 23:11.020 Here you can provide the LM with a knowledge base. 23:11.020 --> 23:12.740 So you can simply add document. 23:14.820 --> 23:18.500 First you need to go to the knowledge base section. 23:22.660 --> 23:26.320 And here you can add URL, you can add files, you can create a text. 23:27.440 --> 23:33.800 And then when you go back to and then you can give access to the knowledge base to your agent. 23:42.240 --> 23:42.880 Right here. 23:43.080 --> 23:47.280 So you can add document and the knowledge base will pop up. 23:50.200 --> 23:56.840 So if you are not using any ten and Python 11 laps also lets you upload documents directly into the 23:57.080 --> 23:58.880 agents knowledge base like this. 23:59.640 --> 24:04.160 Um, it's kind of similar to what we are doing with N810 and PyCon. 24:04.440 --> 24:10.000 The big difference is that with our approach, we have more control, so you can connect it to live 24:10.000 --> 24:14.120 data and add custom logic with multiple data sources. 24:14.720 --> 24:19.480 And and we can track or store conversations if you want. 24:20.120 --> 24:27.100 But yeah, if you just need something quick and simple, you can simply upload the document right here. 24:27.620 --> 24:32.060 So this agent will be able to retrieve information from the documents. 24:33.820 --> 24:34.140 All right. 24:34.180 --> 24:34.580 Great. 24:36.780 --> 24:42.220 Now, if you want your agent to sound more creative or conversational, you can move the temperature 24:42.220 --> 24:44.140 slider to the right. 24:44.260 --> 24:49.140 That tells the AI to take a more risk with how it phrases things. 24:49.140 --> 24:49.700 So it might. 24:49.740 --> 24:54.220 So it might use more casual language or vary how it answers similar questions. 24:54.740 --> 25:00.660 But if you are building something like a support agent or a sales assistant, but you want it to be 25:00.660 --> 25:07.660 consistent and clear, it's better to keep the temperature lower so it sticks more closely to the facts 25:07.660 --> 25:10.580 and gives a less randomness in the responses. 25:10.620 --> 25:12.940 So I want it to be a little bit more creative. 25:12.940 --> 25:17.340 So I'm going to increase the temperature right here like that. 25:18.900 --> 25:20.120 All right let's hit save. 25:21.680 --> 25:21.880 Here. 25:21.880 --> 25:26.760 You can also enable a rack to increase the agents maximum knowledge by size. 25:26.760 --> 25:33.080 So as you can see, the agent will have access to relevant pieces of knowledge base during answer generation. 25:33.080 --> 25:35.760 But we are going to build our agent in ten. 25:35.800 --> 25:40.360 Now, this is the part where we connect our agent to the backend. 25:40.680 --> 25:42.200 So to the workflow. 25:42.200 --> 25:43.800 So right here you have elapsed. 25:43.800 --> 25:46.200 This is where we want to add a custom tool. 25:46.240 --> 25:51.680 So this tool will handle the actual communication between the agent and our vector database. 25:51.720 --> 25:54.440 So do you remember earlier in our system prompt. 25:57.840 --> 26:00.800 We mentioned use the Na ten solar installers. 26:01.120 --> 26:07.280 Actually you can add here to make it clear to query the database for details and solar panels. 26:08.120 --> 26:10.240 So in other words to fetch the information. 26:11.000 --> 26:12.520 So this is where. 26:15.040 --> 26:18.500 We actually define what that tool is and how it works? 26:18.980 --> 26:22.020 So to do that I'm going to add tool. 26:24.460 --> 26:28.820 And click Custom tool because we want to add our webhook URL from N810. 26:29.380 --> 26:30.580 So click on the custom tool. 26:32.060 --> 26:32.740 Perfect. 26:33.380 --> 26:35.420 Now the tool type you want to use webhook. 26:36.060 --> 26:41.380 The webhook allows the agent to extract specific information from the call and send it to your server. 26:41.820 --> 26:43.700 Now what's happening behind the scenes? 26:43.700 --> 26:49.620 When someone talks to the agent, the lamp decides it needs extra information. 26:49.620 --> 26:55.940 It will send a request to a URL that's going to be a webhook will generate an n810. 26:56.100 --> 26:58.580 So it's basically just an API call under the hood. 26:58.620 --> 27:05.420 So once N810 gets that request, which includes the user's question, it runs a workflow, looks through 27:05.420 --> 27:11.860 the pinecone vector database, grabs the most relevant answer, and then sends that response back to 27:11.900 --> 27:12.780 the relevant labs. 27:12.820 --> 27:18.440 Then the 11 labs turns that response into speech And the agent speaks it to the user. 27:18.440 --> 27:19.480 So that's the full loop. 27:19.720 --> 27:24.000 Voice 2010 data from pinecone, back to speech and so on. 27:24.040 --> 27:29.800 So yeah, whenever a user says something to the voice agent, even labs, we try to respond. 27:29.960 --> 27:36.640 But if it sees the prompt, tells it to call an external tool like an 810 solar panel tool, it will 27:36.640 --> 27:43.080 send that question as a Post request to a specific URL, and that URL will be webhook from n810. 27:43.120 --> 27:44.960 So now I'm going to show you how to configure it. 27:45.000 --> 27:46.440 Again we want to use webhook. 27:46.480 --> 27:49.440 Now we need to describe to the LLM how and when to use the tool. 27:49.440 --> 27:54.480 So let's name it the same way we are using it in our system prompt 2010. 27:56.120 --> 27:56.920 You forgot. 27:59.240 --> 28:00.280 Solar installers. 28:01.960 --> 28:02.640 All right. 28:05.600 --> 28:13.200 A tool custom tool webhook name and description. 28:13.200 --> 28:14.160 Enter the description. 28:14.160 --> 28:16.380 Let's say use this tool. 28:16.380 --> 28:20.660 When the user asks something about solar panels or installation that you are not sure about. 28:20.980 --> 28:24.260 Send the questions and wait for a reply before responding. 28:24.300 --> 28:24.620 All right. 28:24.660 --> 28:25.020 Great. 28:25.060 --> 28:29.380 Now, the most important part you have to switch the method from get to post. 28:30.220 --> 28:34.980 So you are switching the method from get to post because you want to send the data. 28:35.540 --> 28:43.220 So the user's question to our webhook and post request lets us include a request body, which is how 28:43.220 --> 28:47.300 the agent sends the actual question over for processing. 28:47.500 --> 28:52.620 So now we need to create URL in our workflow and paste it right here. 28:53.180 --> 28:54.380 So let's go to 2010. 28:54.860 --> 28:58.340 And now let's actually start building the second part of the workflow. 29:04.180 --> 29:05.460 So let's search for webhook. 29:09.980 --> 29:12.620 So we want to start the workflow whenever it is called. 29:15.160 --> 29:16.720 Now let's use the test URL. 29:16.920 --> 29:19.880 But for production you want to switch over to production URL. 29:20.280 --> 29:22.280 Now you want to change the method to post. 29:25.040 --> 29:30.480 So we are setting this to post because in the end we are receiving data from the voice agent. 29:30.480 --> 29:36.040 So from 11 labs like the user's question and the post method, that's the webhook. 29:36.080 --> 29:40.040 Accept and handle that information in the request body. 29:41.720 --> 29:42.520 Alright, great. 29:42.560 --> 29:43.440 Now the path is fine. 29:43.440 --> 29:44.800 You don't need to change it. 29:44.840 --> 29:50.720 You can leave authentication as none but here but here in the response. 29:50.760 --> 29:57.880 So when and how to respond to the webhook instead of immediately you want to change it to using response 29:57.880 --> 29:59.000 to webhook node. 30:00.200 --> 30:01.920 So it will be defined in that node. 30:04.320 --> 30:05.200 Alright great. 30:05.680 --> 30:06.720 Now you can close it. 30:11.440 --> 30:13.540 Now make sure you grab the URL. 30:15.340 --> 30:16.420 So test URL. 30:18.740 --> 30:19.700 Here I'll copy it. 30:21.700 --> 30:23.500 Go to your service agent. 30:27.220 --> 30:28.300 And paste it right here. 30:30.780 --> 30:31.780 Then hit that tool. 30:32.220 --> 30:33.260 Now scroll down. 30:36.580 --> 30:38.100 You don't need to define headers. 30:39.380 --> 30:42.620 The only thing we need to do here is to add body parameters. 30:44.460 --> 30:45.780 And this is the next step. 30:45.780 --> 30:47.820 So we need to define the body parameters. 30:48.180 --> 30:52.700 This is what the voice agent will actually send to N810 when it calls the tool. 30:52.940 --> 30:58.740 So as you can read here, define parameters that will be collected by the LM and sent as the body of 30:58.740 --> 30:59.460 the request. 30:59.740 --> 31:01.020 So I'm going to enable it. 31:04.740 --> 31:05.900 As a description. 31:06.220 --> 31:12.560 I can say something like extract the user's query from the conversation and send it as a request. 31:12.960 --> 31:18.960 So this will be passed to LM and should describe in detail how to extract the data from the transcript. 31:19.000 --> 31:19.880 That should be fine. 31:19.920 --> 31:22.200 Now the next thing we need to do we need to add property. 31:22.840 --> 31:24.120 So click add property. 31:25.080 --> 31:30.440 The only thing we really need here is a field called you can call it questions. 31:32.480 --> 31:34.240 You can call it however you want. 31:36.360 --> 31:41.240 And that's where the user's spoken question will be passed along in the request. 31:41.560 --> 31:48.560 You can also add a short description something like user question. 31:50.720 --> 31:51.800 And that's actually it. 31:51.800 --> 31:55.480 So save changes and hit save again. 31:59.120 --> 32:00.040 Okay perfect. 32:00.080 --> 32:02.480 Now just a quick note because I forgot to mention. 32:02.480 --> 32:05.160 Let's go back to our tool configuration for a second. 32:06.680 --> 32:07.640 So double click. 32:10.260 --> 32:12.340 Scroll down to the body parameters. 32:14.460 --> 32:15.500 And to properties. 32:16.060 --> 32:22.300 So right now we are only sending the question as a body parameter, which is all we need for this basic 32:22.300 --> 32:22.900 setup. 32:23.300 --> 32:29.500 But if you ever want to extend this, like maybe you want the agent to book a calendar meeting, collect 32:29.500 --> 32:33.780 contact details, or pass date and time info, you can add more properties here. 32:34.060 --> 32:38.620 Just create additional fields and then will receive all of them in the webhook. 32:39.300 --> 32:41.540 So for now though, we will keep it simple. 32:42.180 --> 32:43.700 Just the question and answer flow. 32:43.740 --> 32:48.220 So once that's done, click Add Tool and you will see it appear in your agent setup. 32:48.340 --> 32:54.860 So I named mine Solar installers, but you can name yours whatever makes sense for your project. 32:55.580 --> 32:57.340 So now let's move on and finish. 32:57.620 --> 33:00.940 So now let's move on and finish integrating this in. 33:01.500 --> 33:06.940 Once you have added the tool and saved everything, let's do a quick test to make sure the webhook is 33:06.940 --> 33:07.960 working correctly. 33:08.800 --> 33:12.240 So now let's, um, in the webhook settings. 33:15.000 --> 33:21.240 Let's temporarily change it to respond immediately just for the test. 33:23.200 --> 33:26.440 That way we'll get a response right away while testing. 33:27.920 --> 33:29.440 So it's safe. 33:31.720 --> 33:32.800 Now test workflow. 33:35.960 --> 33:37.560 Go to your agent. 33:40.360 --> 33:43.280 So now let's test our agent. 33:43.760 --> 33:45.240 So click on Test agent. 33:47.840 --> 33:54.200 And I'm going to call to talk to it to get the data into our workflow. 33:54.760 --> 33:56.640 So click on this button. 33:58.760 --> 34:00.640 Hey I'm Rob your solar energy guide. 34:00.640 --> 34:01.840 How can I assist you today. 34:04.120 --> 34:05.080 Are available now. 34:08.020 --> 34:09.980 Let me check our current solar panel. 34:13.580 --> 34:15.340 Workflow executed successfully. 34:18.140 --> 34:19.700 Now let's check the executions. 34:21.300 --> 34:22.180 As you can see. 34:25.420 --> 34:26.420 Let's wait a moment. 34:29.020 --> 34:30.460 Let's open our web hook node. 34:32.940 --> 34:37.180 And as you can see, our agent successfully passed all the information. 34:37.180 --> 34:40.940 So the question I asked the webhook to our workflow. 34:42.580 --> 34:42.860 All right. 34:42.900 --> 34:43.380 Great. 34:45.780 --> 34:47.060 Now let's go back to editor. 34:49.620 --> 34:51.900 And let's carry on with the second part of the workflow. 34:51.940 --> 34:53.060 Now this is really important. 34:53.060 --> 35:03.780 Before we add our RAC agent let's change the response field from immediately to using respond to webhook 35:03.800 --> 35:04.200 Note. 35:04.240 --> 35:05.640 You will see why later. 35:06.720 --> 35:07.000 All right. 35:07.000 --> 35:07.480 Great. 35:10.640 --> 35:10.960 All right. 35:10.960 --> 35:11.480 Awesome. 35:12.440 --> 35:13.880 Now you have data in our flow. 35:16.680 --> 35:18.280 So let's search for a agent. 35:18.320 --> 35:21.160 So click on AI agent. 35:22.880 --> 35:23.120 All right. 35:23.160 --> 35:23.800 Perfect. 35:26.280 --> 35:31.040 Now the search for prompt we want to switch over to define below. 35:34.480 --> 35:35.240 Expression. 35:36.440 --> 35:43.560 So remember in the webhook configuration on 11 labs platform we can configure headers parameters body. 35:44.600 --> 35:45.600 So let's go back. 35:46.680 --> 35:47.840 Go back to our agent. 35:54.080 --> 35:54.920 Let's scroll down. 35:55.960 --> 35:57.040 Let's open our tool. 36:01.520 --> 36:03.580 Webhook configuration. 36:04.940 --> 36:09.380 You have headers, parameters, query parameters, and body parameters. 36:11.260 --> 36:12.700 So now let me go back to. 36:13.140 --> 36:21.380 So now when we check any then you have headers parameters query and body. 36:21.740 --> 36:23.340 So here you have two. 36:25.780 --> 36:29.100 So in the prompt field you have to include our question. 36:30.860 --> 36:32.620 So let's drag it and drop it here. 36:33.660 --> 36:33.980 Alright. 36:33.980 --> 36:34.620 Perfect. 36:36.060 --> 36:36.700 Great. 36:36.860 --> 36:42.860 Now of course if you switch over to JSON format you can see how it looks like in JSON. 36:42.900 --> 36:44.540 You have a body and a question. 36:45.980 --> 36:46.340 All right. 36:46.340 --> 36:46.780 Awesome. 36:47.220 --> 36:49.020 So now we have to add a couple more things. 36:49.020 --> 36:50.540 But this is very straightforward. 36:51.220 --> 36:54.100 So first chat model done it many times. 36:54.100 --> 36:58.740 So OpenAI model let's stick with GPT for me. 36:59.940 --> 37:00.540 All right. 37:00.540 --> 37:08.080 Once we have set up the prompt and the question is flowing into an agent, the next step is to connect 37:08.080 --> 37:09.320 it to our vector store. 37:09.880 --> 37:13.000 So I'm going to add a vector store tool to the agent. 37:13.040 --> 37:14.720 So click on add button. 37:15.840 --> 37:16.800 Search for answer. 37:22.840 --> 37:24.240 And you want to use this tool. 37:24.240 --> 37:31.480 So this is what allows it to look up relevant documents based on the user's question. 37:31.520 --> 37:33.000 You have to provide the data name. 37:33.160 --> 37:38.560 So let's call it data plus dot description of the data. 37:39.480 --> 37:41.240 So describe the data in vector store. 37:41.280 --> 37:43.920 This will be used to fill this to description. 37:43.920 --> 37:47.960 Useful for when you need to answer questions about and placeholder. 37:47.960 --> 37:54.000 Whenever you need information about placeholder data description, you should always use this input 37:54.040 --> 37:56.440 should be a fully formed question. 37:56.640 --> 37:58.480 So I'm going to use this description. 37:59.000 --> 38:02.780 Use this data to answer the questions clearly and accurately. 38:03.780 --> 38:07.820 You can specify limit or you can hit this button to let the model define. 38:07.820 --> 38:13.340 This parameter limit is the maximum number of results to return, let's say five. 38:15.460 --> 38:15.980 Great. 38:19.980 --> 38:27.820 And here we have to add a vector star to give our agent access to the documents we vectorize in the 38:27.820 --> 38:29.020 first part of the workflow. 38:29.660 --> 38:31.940 So the files you have created are folder. 38:36.540 --> 38:40.700 So you want to use vector store to work with the data. 38:43.860 --> 38:48.060 Now we have already connected our pinecone account in the previous steps. 38:50.380 --> 38:51.500 In operation mode. 38:53.380 --> 38:57.020 This time we want to retrieve documents as a vector store chain tool. 38:58.400 --> 38:59.480 Just to remind you. 39:05.920 --> 39:09.560 We are using Insert Documents operation as operation mode. 39:10.880 --> 39:12.920 So this time we change it. 39:16.240 --> 39:19.520 To retrieve documents from list. 39:21.520 --> 39:22.880 I choose the same index. 39:22.880 --> 39:24.920 So solar panels demo in my case. 39:25.360 --> 39:29.400 And you also want to add option to namespace. 39:32.240 --> 39:34.000 I forgot the name of the namespace. 39:34.000 --> 39:37.240 So let me go back to icon. 39:38.800 --> 39:40.600 I call it solar panels. 39:40.760 --> 39:43.320 Docs migrate. 39:46.560 --> 39:47.440 And paste here. 39:48.880 --> 39:49.560 Perfect. 39:50.480 --> 39:52.000 Now we need to add Embeddingmodel. 39:53.000 --> 39:59.100 So we are going to use the same Embeddingmodel Laura as we used in the first part of the workflow. 40:00.300 --> 40:01.420 So embedding a model. 40:04.060 --> 40:06.660 Takes embedding pixel perfect. 40:07.540 --> 40:13.540 Now we also need to add model to our Python tool to search for an eye. 40:16.660 --> 40:17.660 And let's stick with GPT. 40:17.700 --> 40:18.380 Follow me. 40:20.260 --> 40:20.860 Perfect. 40:28.580 --> 40:29.620 Now the last node. 40:29.820 --> 40:33.180 So we need to respond back to 11 laps with the final answer. 40:33.700 --> 40:36.300 I'm going to respond to webhook node. 40:36.900 --> 40:43.580 So click on this plus button and this test and intend to return the answer to the original webhook request. 40:45.340 --> 40:46.660 Let's search for webhook. 40:47.820 --> 40:50.860 But this time we want to return this data for webhook. 40:51.460 --> 40:54.260 So we're going to use respond to webhook node. 40:54.260 --> 40:54.320 Note. 40:57.240 --> 41:02.640 And we want to respond to your first incoming item so you can leave it as it is before we give it a 41:02.640 --> 41:03.360 final test. 41:03.360 --> 41:04.760 Let's actually clean it a bit. 41:05.360 --> 41:12.160 So I'm going to rename this note to respond to 11 labs. 41:19.960 --> 41:20.760 All right. 41:20.800 --> 41:24.960 Let's rename this workflow to 11 labs. 41:40.440 --> 41:41.280 Alright, great. 41:42.760 --> 41:43.080 Alright. 41:43.080 --> 41:43.560 Awesome. 41:46.200 --> 41:49.160 Actually this is Rogue Agent, so let's rename it to RAC. 41:51.200 --> 41:51.480 All right. 41:51.520 --> 41:52.160 Perfect. 41:53.500 --> 41:59.660 All right, so now is the time to give our airvoice agent a final test and check if it can actually 41:59.660 --> 42:04.580 answer questions based on the information we stored in our database. 42:04.980 --> 42:07.540 So let's see if everything's working end to end. 42:08.820 --> 42:11.420 So now I'm going to test the workflow. 42:11.980 --> 42:20.500 Just to let you know if you are using a test URL in a web node and not a production URL, and you and 42:20.500 --> 42:24.900 your workflow is not active, you need to hit test workflow. 42:25.580 --> 42:29.260 Every time we want to ask a new question your voice agent. 42:30.420 --> 42:31.940 So let's test workflow now. 42:35.500 --> 42:38.340 Now let's go to our Google Drive folder. 42:40.540 --> 42:44.420 And solar panels installer services document. 42:51.640 --> 42:52.760 Now let's say. 42:55.120 --> 43:01.120 I'm going to ask why you should choose solar panel installation. 43:02.120 --> 43:02.480 Alright. 43:02.480 --> 43:06.480 So let's go back to our agents on Living Labs platform. 43:14.200 --> 43:14.840 This one. 43:15.560 --> 43:17.080 Let's start the conversation. 43:21.480 --> 43:23.480 Hey, I'm Rob, your solar energy guide. 43:23.480 --> 43:24.720 How can I assist you today? 43:25.800 --> 43:26.040 Why? 43:26.040 --> 43:28.480 I should choose your solar panel installation. 43:31.800 --> 43:34.720 Let me check our specific offerings and advantages for you. 43:47.840 --> 43:52.860 We're known for our expertise and quick service, typically responding within five working days. 43:52.900 --> 43:58.020 Our team of renewable energy specialists will guide you through the entire process, and we back our 43:58.020 --> 44:00.340 work with a five year workmanship guarantee. 44:00.500 --> 44:06.220 Plus, we offer flexible payment options and can create custom installation packages based on your specific 44:06.220 --> 44:07.260 budget and needs. 44:10.180 --> 44:10.540 Perfect. 44:10.540 --> 44:10.860 Thank you. 44:12.180 --> 44:12.620 Yes. 44:12.620 --> 44:18.980 So as you just heard, the agent was able to answer the question accurately using the information we 44:18.980 --> 44:24.500 previously vectorized and stored in our Python database. 44:26.980 --> 44:34.900 And of course, if you want to double check where the answer came from, you can go into executions 44:34.900 --> 44:35.740 in any turn. 44:38.820 --> 44:40.260 So let's open executions. 44:42.740 --> 44:43.780 This is the last one. 44:45.540 --> 44:53.080 And as you can see our agents successfully retrieved the information from the Python vector store. 44:54.440 --> 44:57.840 It didn't provide the answer just by using the LM. 44:58.360 --> 44:59.360 So like ChatGPT. 45:02.000 --> 45:02.360 All right. 45:02.360 --> 45:02.800 Awesome. 45:03.880 --> 45:04.480 All right. 45:04.480 --> 45:10.320 In this part of the tutorial, we are going to make a couple of small changes to get things closer to 45:10.360 --> 45:11.680 a production ready setup. 45:12.560 --> 45:18.760 And the first thing we'll do is switch the manual trigger for a schedule trigger so the workflow can 45:18.760 --> 45:23.000 run automatically without you needing to start it manually every time. 45:23.480 --> 45:26.480 So let's search for schedule trigger. 45:39.880 --> 45:42.160 Now let's remove the manual trigger. 45:45.400 --> 45:49.260 Over here And connect it to the Google Drive node. 45:55.220 --> 46:00.140 Now I want this workflow to be activated every day. 46:01.340 --> 46:02.780 Let's say at 7 a.m.. 46:05.460 --> 46:05.740 Right. 46:05.780 --> 46:06.420 Perfect. 46:06.540 --> 46:10.180 So basically this workflow will run on the schedule you define here. 46:10.820 --> 46:12.140 Once you activate it. 46:12.140 --> 46:14.220 So the next step is to activate the workflow. 46:17.060 --> 46:17.780 Confirm. 46:19.660 --> 46:20.260 Perfect. 46:23.140 --> 46:27.300 Now I want to show you a few things on the platform. 46:27.300 --> 46:29.420 So let's go back to our agent. 46:33.820 --> 46:39.460 Yes I forgot to mention that here in the voice section you can test different voices. 46:40.500 --> 46:41.820 So by default we have Eric. 46:43.140 --> 46:51.000 You can also play with different settings, such as you can configure latency, you can change the stability 46:51.160 --> 46:54.840 to higher values, which will make the speech more consistent. 46:55.280 --> 46:57.560 But you can also make it sound monotone. 46:58.840 --> 47:05.600 You can adjust the speed to control the speed of the generated speech similarity, so higher values 47:05.600 --> 47:11.760 will boost the overall clarity and consistency of the voice, and very high values may lead to artifacts. 47:11.800 --> 47:15.040 Adjusting the value to find the right balance is recommended. 47:16.160 --> 47:18.520 Right now. 47:18.520 --> 47:25.360 Here in the analysis, you can define custom criteria to evaluate conversations against. 47:25.360 --> 47:29.120 You can find the evaluation results for each conversation in the history tab. 47:30.160 --> 47:35.240 You can also collect data so you can define custom data specifications to extract from conversation 47:35.240 --> 47:36.080 transcripts. 47:36.080 --> 47:41.320 You can always refer to the documentation to find more details about the settings. 47:41.880 --> 47:48.580 Now I'm not going to dive deeper into all the settings in this tutorial, but what I wanted to show 47:48.580 --> 47:49.660 you is a widget. 47:50.660 --> 47:55.260 So here you have a code snippet we can copy. 47:56.460 --> 48:01.140 So add the following snippet to the pages where you want the conversation widget to be, which looks 48:01.140 --> 48:01.820 like that. 48:02.020 --> 48:06.020 Also, you can enable feedback collection, so callers will be able to provide feedback continuously 48:06.020 --> 48:08.860 during the conversation and after it ends. 48:09.300 --> 48:12.620 Information about which agent response caused the feedback will be collected. 48:15.700 --> 48:19.980 Here you can customize the appearance of this widget so you can change the background color, text, 48:19.980 --> 48:20.900 color, etc.. 48:22.940 --> 48:24.460 You can also change the avatar. 48:25.700 --> 48:28.620 Add your favorite color or brand color. 48:34.340 --> 48:40.660 Now, if you want to add this beautiful voice widget to any website, all you need to do is paste the 48:40.660 --> 48:49.000 snippet code into a site's HTML, ideally somewhere between the body tags, so this makes sure the widget 48:49.000 --> 48:49.960 loads properly. 48:50.560 --> 48:55.800 And I'm going to quickly show you how this works using one of my website as an example. 48:56.160 --> 48:56.560 All right. 48:56.560 --> 48:58.040 So this is one of my websites. 48:58.040 --> 49:02.840 As you can see there is no voice widgets showing at the bottom of the page. 49:03.080 --> 49:06.520 Now I'm going to show you how easy it is to add to any website. 49:06.760 --> 49:09.920 This website is built on WordPress and hosted on Hostinger. 49:10.280 --> 49:13.120 So first I will head over to my WordPress admin dashboard. 49:13.480 --> 49:18.960 Now the exact place where you paste the snippet code can vary depending on the platform you are using 49:19.480 --> 49:20.400 on WordPress. 49:20.440 --> 49:23.600 It's usually done through theme settings or with a plugin. 49:24.000 --> 49:30.160 Now, if you are using a different platform like Webflow, Wix, Squarespace, or even a plain HTML 49:30.160 --> 49:36.520 site, you will usually want to paste it somewhere inside the body tags, ideally just before the closing 49:36.520 --> 49:40.780 body tag or in the footer section if the platform provides one. 49:41.220 --> 49:44.860 So now when I go to theme options. 49:48.580 --> 49:50.300 Inside I have integration tab. 49:53.100 --> 49:58.380 And when I scroll down this is where I can paste custom code. 49:58.380 --> 50:02.740 So the exact place where you paste the snippet code can vary depending on the platform. 50:02.740 --> 50:09.580 As I said, and as I mentioned earlier, the best place to add the snippet is somewhere inside the body 50:09.580 --> 50:10.100 tags. 50:11.140 --> 50:14.460 So I'm going to copy the code snippet from 11 labs platform. 50:16.180 --> 50:18.100 Go to my WordPress website. 50:21.260 --> 50:22.820 And paste code right here. 50:25.060 --> 50:28.500 And then save changes. 50:30.940 --> 50:36.620 And now when I refresh the website, the voice agent widget appears at the bottom of the page so I can 50:36.620 --> 50:37.340 talk to it. 50:37.660 --> 50:38.500 So let's try. 50:40.480 --> 50:40.960 three. 50:42.760 --> 50:44.640 Hey, I'm Rob, your solar energy guide. 50:44.640 --> 50:45.920 How can I assist you today? 50:47.040 --> 50:47.560 Hi, Rob. 50:47.560 --> 50:49.640 What's what services do you offer? 50:50.880 --> 50:52.880 Let me check our service offerings for you. 50:52.880 --> 50:55.240 I'd be happy to tell you about our main services. 50:55.240 --> 51:00.720 We specialize in solar panel installations, battery storage solutions, and EV charger installations 51:00.720 --> 51:03.200 for both residential and commercial properties. 51:03.200 --> 51:07.960 We handle everything from initial consultation and system design to professional installation and after 51:08.000 --> 51:09.080 service support. 51:09.080 --> 51:11.440 Would you like to know more about any specific service? 51:11.880 --> 51:12.240 Alright. 51:12.240 --> 51:12.760 Thanks, Rob. 51:12.800 --> 51:13.160 Bye. 51:14.720 --> 51:14.960 Yes. 51:14.960 --> 51:20.840 So remember we trained our RAC agent based on the information available on this website. 51:21.760 --> 51:22.840 So from this company. 51:23.720 --> 51:24.040 Right. 51:24.040 --> 51:24.520 Great. 51:25.680 --> 51:25.880 Yes. 51:25.880 --> 51:30.440 So now visitors can actually talk to the agent and ask questions. 51:30.480 --> 51:34.640 For example I could vectorize some key information about the services we offer. 51:35.200 --> 51:41.660 Add that to our knowledge base and the agent will be ready to answer those queries in real time. 51:43.620 --> 51:45.540 Now, back in my WordPress dashboard. 51:46.500 --> 51:49.980 As you can see, I'm using a built in integration panel. 51:50.100 --> 51:58.460 I don't actually see the full HTML code with the body tags directly, but this box is designed to inject 51:58.460 --> 52:03.500 whatever, whatever code I paste here into the body section of this site. 52:04.180 --> 52:11.180 So even though I can't see the actual body tags, this works just like placing it before the closing 52:11.180 --> 52:13.420 body tag manually. 52:14.300 --> 52:16.140 So this would look like that. 52:16.220 --> 52:19.980 Um, let me just paste here this. 52:20.380 --> 52:24.780 Yes, we have opening body tag for the website content. 52:25.580 --> 52:29.220 And this is where the widget code will go right here. 52:29.500 --> 52:31.060 And then closing body tag. 52:31.260 --> 52:31.580 All right. 52:31.580 --> 52:35.980 So now you know how to add the voice widget onto your website. 52:36.020 --> 52:36.580 Perfect.