WEBVTT 00:00.480 --> 00:01.200 Welcome back. 00:01.520 --> 00:08.240 In this video, I'm going to present you a demo and overview of the WhatsApp AI, customer support and 00:08.240 --> 00:12.320 low ticket sales agent and explain how it works node by node. 00:12.720 --> 00:15.680 And at the end, I've got a task for you. 00:16.200 --> 00:22.440 After watching the video, please download the workflow and try to connect all the credentials and modify 00:22.440 --> 00:23.800 this solution to your needs. 00:24.040 --> 00:29.680 If you have been following this course lesson by lesson, you should already have most of the required 00:29.680 --> 00:39.160 accounts connected like WhatsApp, OpenAI, or Google Drive, so you can simply click on a specific 00:39.160 --> 00:45.920 node and choose the account you want to connect, for example, Google Drive and so on. 00:51.200 --> 00:57.840 Now, just to remind you, you can manage all your credentials in this section for each project separately. 00:58.640 --> 01:03.240 So just go to credentials and you can manage all your accounts right here. 01:04.770 --> 01:11.570 However, as I mentioned in one of the previous lessons, some platforms like WhatsApp may require you 01:11.570 --> 01:17.010 to refresh your credentials from time to time and re-enter the API keys manually. 01:17.170 --> 01:23.330 Also, you have not used Google Gemini in any of the previous projects, so if you are stuck setting 01:23.330 --> 01:30.450 that up or need help connecting any platform, feel free to go to the dedicated section on setting up 01:30.450 --> 01:31.330 credentials. 01:31.650 --> 01:37.970 Next in this project, we'll be using Supabase vector database to give our WhatsApp agent long term 01:37.970 --> 01:38.530 memory. 01:39.090 --> 01:44.370 So to find out exactly how to use Supabase, configure everything and complete the setup. 01:44.770 --> 01:50.290 Please refer to the next lesson called How to Build an Agent That Remembers Everything. 01:50.330 --> 01:52.970 Where I will show you step by step how to do it. 01:53.250 --> 01:54.690 Alright, now it's your turn. 01:55.210 --> 01:56.690 Watch the rest of this video. 01:57.370 --> 01:59.530 Download and import the workflow. 01:59.690 --> 02:03.650 Connect all your accounts and make this WhatsApp agent your own. 02:04.090 --> 02:06.490 Alright guys, so let's do a quick demonstration. 02:07.130 --> 02:16.260 So I trained my WhatsApp I agent using my company's information like the services we offer using the 02:16.260 --> 02:17.660 first part of the workflow. 02:19.140 --> 02:27.700 So now when I click on the workflow, it's waiting for the trigger to fire in the second part of the 02:27.700 --> 02:28.300 workflow. 02:29.220 --> 02:29.940 With this one. 02:34.420 --> 02:36.020 So let's simulate a message. 02:36.420 --> 02:41.820 Imagine a client type something like what services do you offer? 02:51.500 --> 02:52.740 And as you can see. 02:55.940 --> 02:57.300 The workflow has been triggered. 02:58.660 --> 03:04.060 And now agent is working on a response to the user. 03:07.700 --> 03:13.640 And yeah adaptive AI offers a range of services designed to enhance business operations and customer 03:13.640 --> 03:15.480 engagement, including etc.. 03:17.800 --> 03:26.880 So basically the agent created a response using the information stored in a base vector store. 03:29.120 --> 03:36.160 And this is where we pulled all the details from the word file that we pulled from Google Drive using 03:36.160 --> 03:39.400 the first part of the workflow. 03:40.440 --> 03:46.920 This one, and this is the fourth file in Google Drive. 03:48.360 --> 03:48.560 Okay. 03:48.600 --> 03:51.600 So now let's test the voice message feature. 03:52.720 --> 03:54.520 So click on the test workflow. 03:57.320 --> 03:59.040 And I'm going to ask the same question. 03:59.480 --> 04:00.760 What services do you offer. 04:06.440 --> 04:10.080 And we should receive similar response from our agent. 04:20.730 --> 04:23.530 It takes like ten, 15 seconds. 04:25.930 --> 04:26.130 Yeah. 04:26.170 --> 04:29.610 As you can see, adaptive AI provides a variety of services, including. 04:30.290 --> 04:36.530 In this video, I will walk you through how to set up a WhatsApp AI agent that's trained on your company's 04:36.530 --> 04:39.290 data and can retain conversation history. 04:39.850 --> 04:44.490 So you can set up, set this up for yourself, or even sell it for your clients. 04:44.610 --> 04:51.050 So this agent can handle text and voice messages and retrieve relevant information, and then reply 04:51.090 --> 04:54.530 naturally by keeping the context of the conversation. 04:55.210 --> 04:56.410 So let's get started. 04:58.730 --> 05:02.250 So this is the first part of the workflow. 05:03.650 --> 05:09.210 And the workflow starts with when clicking the test workflow node. 05:09.890 --> 05:16.730 And this is manual trigger that lets you test the setup by running the workflow on demand. 05:18.130 --> 05:21.340 Next we have uh, the Google Drive node. 05:22.380 --> 05:26.500 So this node downloads a file from a specific Google Drive folder. 05:26.500 --> 05:31.500 So you can select which document you want to process and store in the database. 05:32.180 --> 05:39.540 So this is where your company's documents, like PDFs or word files are retrieved for further processing. 05:41.740 --> 05:42.140 Um. 05:44.820 --> 05:45.460 Next. 05:45.660 --> 05:51.940 Once the file is retrieved, uh it's passed to Supabase vector store node. 05:51.940 --> 05:56.620 So here, uh, the queue is stored as part of your knowledge base. 05:57.100 --> 06:03.380 So this step ensures that the data is available for processing in the next steps. 06:04.660 --> 06:12.940 So after storing, uh, after storing the workflow moves to recursive characters. 06:13.060 --> 06:15.380 Uh splitter text splitter node. 06:16.100 --> 06:18.860 So this splits the document into smaller chunks. 06:19.060 --> 06:26.550 And splitting is crucial because AI models Perform better with smaller, structured, structured pieces 06:26.550 --> 06:27.190 of text. 06:28.310 --> 06:33.350 So next up the default data loader node. 06:33.350 --> 06:39.790 So it organizes the text chunks and ensures they're ready for embedding. 06:39.790 --> 06:43.470 So this step format the data for smooth processing. 06:45.550 --> 06:50.310 After formatting the chunks are passed to the embeddings OpenAI node. 06:50.870 --> 07:00.070 So this node generates a numerical embeddings for each chunk of text, making it easier to compare and 07:00.070 --> 07:02.390 retrieve relevant information later. 07:03.470 --> 07:08.350 And finally, the embeddings are sent back to the base vector store. 07:09.270 --> 07:15.950 So these embeddings are stored in the database and creates a searchable vector store. 07:16.390 --> 07:23.430 So this is essential for enabling enabling your AI to retrieve the right information when queried. 07:26.910 --> 07:29.480 So this is the second part of the workflow. 07:30.360 --> 07:33.960 So the workflow starts with the WhatsApp trigger node. 07:35.000 --> 07:43.000 This listens for incoming messages like texts or audio and from from your clients on WhatsApp. 07:44.000 --> 07:44.640 One. 07:44.680 --> 07:50.080 Once a message is received, it moves on to the next steps in the workflow. 07:50.920 --> 07:55.080 So the trigger connects to a split out message pass node. 07:55.080 --> 08:00.960 So this node checks whether the incoming messages, uh, is text or audio. 08:01.960 --> 08:13.000 So if the message is uh, an audio, uh, it's passed to the get audio URL and download uh audio nodes. 08:13.640 --> 08:25.680 So this nodes uh fetch and download the audio file file and then uh Google Gemini audio node uh processes 08:25.680 --> 08:28.720 it to transcribe the voice into text. 08:29.720 --> 08:31.260 So Transcription. 08:31.300 --> 08:37.660 Tenants voice messages into text so the AI can process them like regular text messages. 08:39.940 --> 08:40.540 Next. 08:40.780 --> 08:47.060 Once the message is transcribed, or if it's already text, it goes to the Getusers message node. 08:47.420 --> 08:54.180 And this node organizes and formats the message, pulling out important details like the type of message, 08:54.180 --> 08:56.220 the text content, and who sent it. 08:56.820 --> 09:03.420 So it extracts the sender's phone number and message text so they are ready to use in the next steps. 09:04.540 --> 09:13.540 So the information go to AI in this case is AI tools agent. 09:14.420 --> 09:19.260 So next the Postgres chat memory node comes into play. 09:19.380 --> 09:26.020 And this node lets the AI remember the conversation history, including previous messages and context 09:26.340 --> 09:28.900 like the client's name or earlier questions. 09:29.340 --> 09:37.310 So this basically this makes for follow up questions easy to handle without the client needing to repeat 09:37.310 --> 09:37.950 themselves. 09:38.310 --> 09:40.830 And it keeps the interaction feeling natural. 09:41.630 --> 09:45.910 So now we move to supervised vector store. 09:46.510 --> 09:53.150 Uh, this is where the agent searches your company's knowledge base to find the most relevant information. 09:53.670 --> 09:57.150 So it works with, uh, embeddings. 09:57.270 --> 10:06.070 OpenAI model, um, to match the client's questions, to stored knowledge and pull up the best possible 10:06.070 --> 10:06.550 answer. 10:07.310 --> 10:13.870 So after gathering all the data, the OpenAI chat model node, uh, generates response. 10:13.910 --> 10:21.070 Uh, it combines the client's question and the retrieved knowledge and the conversation history to create 10:21.070 --> 10:22.830 a helpful and natural reply. 10:23.910 --> 10:28.950 And the final step is to reply to the, uh, user node. 10:30.590 --> 10:35.990 This sends the response back to the client through WhatsApp.