WEBVTT 00:01.360 --> 00:02.160 Header. 00:02.160 --> 00:08.920 In the last tutorial, I showed you how to build an AI voice agent in retail AI that gets triggered 00:08.960 --> 00:17.080 when someone submits a form like for example, on your website, it replies using a single prompt. 00:19.720 --> 00:29.680 So that setup works fine, but it does have its limitations, like hallucinations or doing the wrong 00:29.720 --> 00:34.360 thing, especially as your use cases gets more complex. 00:35.520 --> 00:39.440 So basically, we created a single prompt voice agent. 00:40.080 --> 00:48.000 We added a prompt that tells the agent what to say, the tone, the style, and how to behave, and 00:48.000 --> 00:49.640 to make it a bit more dynamic. 00:49.640 --> 00:58.600 We passed in a few variables, but everything still runs through that one single prompt, and that's 00:58.920 --> 01:01.380 what we call a single prompt. 01:01.420 --> 01:02.540 A voice agent. 01:05.860 --> 01:15.860 So even if you use the best large language models like GPT four or cloud 3.5 or 3.7, and single prompt 01:15.860 --> 01:24.020 agents can still run into issues like hallucinations so they can hallucinate or doing the wrong thing, 01:24.060 --> 01:28.940 especially especially as your use case gets more complex. 01:29.820 --> 01:37.500 So single prompt voice agents are great, but as your logic grows, they can become less accurate and 01:38.300 --> 01:43.780 which can often lead to hallucinations or issues when handling multiple function calls. 01:45.260 --> 01:53.100 At this time, we are going to build a more advanced AI voice agent using something called a conversational 01:53.100 --> 01:53.580 flow. 01:53.980 --> 01:57.140 So it's a relatively new method of building a voice agents. 01:57.740 --> 02:00.810 So you have control of every step. 02:01.490 --> 02:04.010 So what does that actually mean? 02:04.090 --> 02:09.330 So think of it like designing a conversation using a flowchart. 02:09.850 --> 02:10.330 So. 02:12.770 --> 02:21.250 Instead of including everything into one big prompt, we break it down into smaller parts. 02:22.650 --> 02:27.090 So each step handles a single task. 02:28.050 --> 02:35.330 So this approach is way more powerful when you are building something for a reward because it keeps 02:35.330 --> 02:36.450 the AI focused. 02:36.770 --> 02:44.970 So each part of the flow does one thing, and you can create a separate path for things like checking 02:44.970 --> 02:49.610 calendar availability, sending an email, booking a meeting. 02:50.770 --> 02:54.090 So it only makes your agent way easier to manage and scale. 02:54.130 --> 02:57.370 And you can add logic at every step. 02:57.430 --> 03:03.070 So trigger automations in n810 and control how the conversation flows. 03:05.590 --> 03:12.190 So in this video I will show you a quick demo and then we will build it together step by step. 03:13.070 --> 03:17.630 And by the end of this tutorial you will have a very powerful voice agent. 03:18.390 --> 03:18.790 All right. 03:18.830 --> 03:22.470 Now I'm going to play a demo video I recorded earlier. 03:22.750 --> 03:24.790 So it includes a walkthrough. 03:24.790 --> 03:28.150 So you will get a clear idea of how the system works. 03:28.910 --> 03:29.630 Hi everyone. 03:29.910 --> 03:37.150 Today I'm going to show you a very powerful AI voice agent I built using retail AI and N810 that can 03:37.150 --> 03:44.070 completely replace receptionist appointment setters, virtual assistants, and you can use it for many 03:44.070 --> 03:47.030 more use cases and tweak to any industry. 03:47.630 --> 03:49.190 And this is not just a prototype. 03:49.190 --> 03:54.790 This system is production ready so you can implement it into your business or sell it to your clients. 03:54.830 --> 03:55.910 Now let's do a quick demo. 03:55.910 --> 03:58.380 So I'm going to call this physical therapy clinic. 03:59.900 --> 04:00.860 Just bear with me. 04:02.660 --> 04:03.140 Hello. 04:04.620 --> 04:04.940 Hello. 04:04.980 --> 04:05.940 This is Sophia. 04:06.100 --> 04:07.220 How can I assist you today? 04:07.540 --> 04:07.700 Hi. 04:07.700 --> 04:09.460 I'm looking for a chiropractor, please. 04:11.060 --> 04:15.500 Would you like me to book an appointment for you with Kate, our chiropractic specialist? 04:15.540 --> 04:16.340 Yes, please. 04:18.260 --> 04:19.500 Can I have your name, please? 04:20.460 --> 04:21.100 I'm Damien. 04:23.500 --> 04:24.740 What's your email address? 04:25.500 --> 04:31.100 It's n I n d e m o at gmail.com. 04:33.060 --> 04:34.860 I also need your mobile number, please. 04:35.260 --> 04:40.020 It's 07 5648 2602. 04:43.780 --> 04:48.020 Can I ask what symptoms you have and if there is anything else you would like me to pass to Kate? 04:48.820 --> 04:53.260 So my knee is very painful and it's a sport injury. 04:55.400 --> 04:56.720 Thank you for the details. 04:56.720 --> 04:58.360 What day and time worked best for you? 04:58.920 --> 05:02.120 Can we do on Monday at 2 p.m., please? 05:03.560 --> 05:05.040 I'm checking the availability for you. 05:05.080 --> 05:06.160 Give me a moment, please. 05:30.960 --> 05:31.480 Hello. 05:33.520 --> 05:35.320 Thank you for your patience, Damon. 05:35.520 --> 05:39.920 It looks like the time slot on Monday at two in the afternoon is not available. 05:39.960 --> 05:45.120 Would you like to try for a morning slot at 11 in the morning, or an afternoon slot at three in the 05:45.160 --> 05:46.600 afternoon on the same day? 05:47.080 --> 05:49.080 Yeah, I can do at 11 a.m., please. 05:50.760 --> 05:51.360 Great. 05:51.800 --> 05:56.790 So just to confirm, I have you scheduled for an appointment with Kate, our chiropractor, on Monday, 05:56.790 --> 05:57.870 at 11 in the morning. 05:57.910 --> 06:02.990 Your details are name, daemon, email, end date and demos at gmail.com. 06:02.990 --> 06:06.390 Mobile 07 5648 2602. 06:06.430 --> 06:09.030 Appointment Monday at 11 in the morning. 06:09.030 --> 06:10.150 Is everything correct? 06:10.510 --> 06:10.830 Yes. 06:10.830 --> 06:11.750 Everything correct. 06:11.750 --> 06:12.270 Thank you. 06:13.950 --> 06:15.310 Your appointment is confirmed. 06:15.310 --> 06:17.350 You will receive a confirmation shortly. 06:17.390 --> 06:18.190 Have a nice day. 06:18.350 --> 06:18.830 Thank you. 06:18.870 --> 06:19.390 Bye bye. 06:22.790 --> 06:27.990 Now, as you can see, the meeting has been successfully booked and we got an email. 06:28.670 --> 06:32.070 So I'm using my test email for a specialist. 06:32.070 --> 06:37.910 So a chiropractor, um, and we got a summary email with all the details. 06:39.030 --> 06:46.110 So we've got a client name, contact details and the reason for the appointment. 06:46.470 --> 06:51.510 I designed this invoice agent specifically for a physical therapy clinic. 06:51.870 --> 07:00.330 It handles all appointment bookings for three specialists chiropractor, massage therapist and physiotherapy 07:00.570 --> 07:01.730 rehab specialist. 07:02.210 --> 07:09.530 Each of these specialists has a separate calendar in Qualcomm AI voice agent has human in the loop functionality 07:09.970 --> 07:16.610 so it detects when a caller is frustrated or requests a real person, and seamlessly transfers the call 07:16.810 --> 07:20.330 to a human agent or directly to the specialist. 07:20.850 --> 07:27.330 Also, it provides a summary of the conversation before the transfer, ensuring that the human agent 07:27.330 --> 07:28.610 has full context. 07:30.490 --> 07:38.730 Because this system is built using the conversation flow method, it cuts down AI hallucinations by 07:38.730 --> 07:42.730 90%, so it stays on track and doesn't make things up. 07:43.210 --> 07:49.490 You have full control over what the agent says, but if you want, you can also give it more flexibility 07:49.490 --> 07:56.160 to handle open ended conversations, which isn't possible to achieve with other tools without coding. 07:57.960 --> 08:04.560 The agent is also trained on the company's knowledge base, so it can instantly answer frequently asked 08:04.560 --> 08:10.920 questions and provide accurate information without needing an intervention. 08:11.200 --> 08:16.640 So here's what happens when someone calls the clinic looking to book an appointment. 08:17.120 --> 08:22.480 The AI voice agent picks up the call and guides them through booking an appointment. 08:23.080 --> 08:26.320 Then it checks the specialist's availability in real time. 08:26.680 --> 08:31.120 Once they pick a date and time, the system locks in the appointment. 08:31.160 --> 08:39.040 Automatically, the client receives an email confirmation and the specialist gets an email summary of 08:39.040 --> 08:45.440 the conversation, including the client's contact details, their symptoms, and any extra notes or 08:45.440 --> 08:46.600 comments, they added. 08:46.920 --> 08:56.020 Each specialist has a dedicated AI Agent built in N8N that processes and analyses all the information 08:56.260 --> 08:58.820 before sending a detailed email summary to them. 08:59.620 --> 09:02.540 This system is designed to be super adaptable. 09:02.660 --> 09:05.020 You can tweak it for different industries. 09:05.300 --> 09:11.540 For example, instead of a clinic, you could use it for routing calls to different departments in a 09:11.540 --> 09:12.100 business. 09:12.500 --> 09:15.500 Automating scheduling for beauty salons and spas. 09:15.860 --> 09:19.980 Handling customer inquiries before escalating to a human. 09:20.460 --> 09:23.700 Processing appointments for any service based business. 09:23.900 --> 09:26.940 Handling inbound support for real estate agencies. 09:27.260 --> 09:30.660 Automating customer service for e-commerce stores. 09:30.740 --> 09:31.460 And much more. 09:33.780 --> 09:36.260 Honestly, this tech is a game changer. 09:36.780 --> 09:43.420 Whether you're a clinic, an agency, or any business that handles a lot of calls, an AI voice agent 09:43.420 --> 09:47.460 like this can save time, money, and frustration. 09:48.720 --> 09:56.360 All right, so now let's, um, get our hands dirty and start building this voice agent from scratch 09:56.400 --> 09:57.160 in little AI. 09:57.840 --> 10:05.840 And in the meantime, we'll jump will jump into an A-10, uh, to build a agents, to integrate the 10:06.120 --> 10:16.200 score analysis function and also to send a confirmation email with the notes the AI captures during 10:16.200 --> 10:17.280 the conversation. 10:17.280 --> 10:19.200 And let's create an agent from scratch. 10:19.320 --> 10:23.560 This time we are going to build a conversation flow agent. 10:23.560 --> 10:28.880 So this step, this type of AI voice agent is best for tasks with complex transitions. 10:28.880 --> 10:30.280 So let's start from blank. 10:30.840 --> 10:38.560 But you can also use pre-built templates such as for patient screening, real estate qualification, 10:39.040 --> 10:39.920 and others. 10:41.440 --> 10:42.560 So start from blank. 10:50.270 --> 10:55.710 Alright, so the first thing is to configure our welcome node. 10:56.030 --> 11:00.750 So this is so basically this is the conversation node. 11:01.270 --> 11:06.910 So we'll use this node most of the time just by clicking on conversation right here. 11:08.470 --> 11:10.550 But for now let's delete this one. 11:15.950 --> 11:23.190 So in this welcome node we need to type what we want our voice agent to say as a welcome message. 11:24.110 --> 11:27.470 So by default we have hello this is customer support department. 11:27.510 --> 11:28.750 How can I help you today. 11:29.710 --> 11:31.030 But we can change it. 11:31.030 --> 11:34.110 And I'm going just to copy and paste. 11:35.030 --> 11:36.030 Hello, this is Sophia. 11:36.030 --> 11:37.430 How can I assist you today. 11:42.030 --> 11:42.790 All right. 11:42.830 --> 11:45.310 And here we can add transitions. 11:45.790 --> 11:50.730 So it's basically like a path we want a AI voice agent to follow. 11:50.930 --> 11:58.410 So by default we have user needs to return the package, or user needs to check the order status. 11:58.690 --> 12:06.610 And because we are building a voice agent for a physical therapy clinic with with three specialists, 12:07.450 --> 12:13.930 for chiropractic care specialists, for therapy specialists, and for physiotherapy and rehabilitation 12:13.930 --> 12:14.730 specialists. 12:15.530 --> 12:17.850 So we need to add three transitions. 12:18.170 --> 12:21.090 So the first will be for chiropractor specialists. 12:21.130 --> 12:23.250 So I'm going to copy and paste it. 12:23.970 --> 12:25.050 The user needs to talk. 12:25.090 --> 12:28.330 Chiropractic care specialists and change it. 12:28.370 --> 12:29.370 Change this one. 12:33.810 --> 12:40.250 The second transition would be to massage therapy specialist. 12:41.330 --> 12:43.330 I'm just going to copy and paste it. 12:45.720 --> 12:50.520 And the transition I'm going to add by just clicking on this plus button. 12:53.200 --> 12:57.520 And using it to talk to physiotherapy rehabilitation specialists. 12:58.200 --> 13:06.720 So after our voice agent says this line, it analyzes the customers response. 13:06.720 --> 13:15.040 And based on that it transitions the flow so it moves the flow to the appropriate next node. 13:15.280 --> 13:22.600 So in that case it would move the conversation just following this path to the next node. 13:22.600 --> 13:24.120 So to the appointment handler. 13:24.760 --> 13:31.160 And then it will say would you like me to book an appointment for you with Rob our physiotherapist and 13:31.160 --> 13:33.920 rehab specialists and so on. 13:34.520 --> 13:42.320 So we are going to as a third transition add user needs to talk to physiotherapy and rehabilitation. 13:42.680 --> 13:44.900 So this time I'm going to type it manually. 13:45.380 --> 13:52.140 During the talk, the user needs to talk to. 13:56.060 --> 13:56.780 Ezio. 13:57.460 --> 13:58.180 Therapy. 14:02.180 --> 14:04.580 And rehabilitation specialist. 14:11.340 --> 14:12.140 All right. 14:16.100 --> 14:16.780 Great. 14:16.940 --> 14:21.740 So the first one is is for chiropractic care specialists. 14:23.020 --> 14:25.820 So we are going to add the same type of node. 14:25.820 --> 14:27.380 So conversation node. 14:30.100 --> 14:31.820 Let's move it up. 14:32.820 --> 14:38.580 And then we are going to connect them together. 14:41.060 --> 14:44.730 And then let's Ask another question. 14:45.650 --> 14:46.130 So? 14:46.810 --> 14:48.330 So, would you like to book? 14:53.770 --> 14:54.730 An appointment. 14:57.930 --> 14:58.330 With. 15:00.970 --> 15:02.410 Our chiro. 15:02.930 --> 15:03.650 Practic. 15:06.370 --> 15:07.130 Specialist. 15:07.250 --> 15:10.970 You have two options to choose from in the conversation node. 15:11.290 --> 15:17.490 The first one is prompt, but 90% of the use cases will use static sentences. 15:17.490 --> 15:23.010 This is the exact sentence we want our voice agent to say. 15:23.610 --> 15:29.770 The prompt gives an LLM more flexibility, but is less accurate. 15:30.410 --> 15:41.050 So again, because if you want to use a prompt, really allow for more randomness and consequently this 15:41.050 --> 15:43.870 can cause more hallucinations. 15:43.910 --> 15:47.070 Going to stick with static sentence option. 15:48.950 --> 15:53.910 All right, so now we are gonna add transitions to this node. 15:54.110 --> 15:58.350 So in other words instructions to our voice agent. 15:59.670 --> 16:00.190 So. 16:02.270 --> 16:11.470 If the user wants to book an appointment you can say something like if the user wants to book an appointment. 16:11.990 --> 16:16.190 So this is the first branch and the second transition. 16:16.710 --> 16:25.550 If the user the customer is not interested in booking an appointment but has some questions, we can 16:25.550 --> 16:31.270 ask something like we can add something like if the user have questions answered shortly based on the 16:31.310 --> 16:32.110 knowledge base. 16:32.630 --> 16:41.140 So for the AI voice agent built using conversation flow method, we can also add knowledge base. 16:41.660 --> 16:50.300 So the agent can go through the knowledge base and find the relevant answer for the user questions. 16:50.980 --> 16:55.580 Let's rename this node to Appointment Handling. 17:02.020 --> 17:11.140 And now we need to add another conversation node if the user want to book an appointment. 17:12.980 --> 17:16.340 We want to ask them about the details. 17:17.180 --> 17:18.660 So our agent can. 17:21.540 --> 17:24.660 Capture capture of of all this information. 17:25.940 --> 17:29.940 And then and then use it to to book an appointment. 17:36.140 --> 17:43.520 And this time we want to ask About their name and as a transition. 17:45.640 --> 17:52.520 If the user answered the question, we want a reverse agent to move to the next node. 17:53.600 --> 18:00.520 So let's add another conversation node and connect it right here. 18:10.680 --> 18:12.320 And let's rename this node to. 18:15.280 --> 18:17.680 Collect details. 18:19.840 --> 18:22.240 And let's say we want to ask about their email. 18:24.120 --> 18:26.160 So we can always follow up. 18:27.400 --> 18:28.720 So what's your email address. 18:30.360 --> 18:33.680 And as a transition let's add the same condition. 18:33.680 --> 18:36.880 So once the user. 18:40.540 --> 18:42.620 Answer the question, move to the next node. 18:45.820 --> 18:51.380 Then we need another conversation node to ask about their phone number. 18:52.340 --> 18:54.020 So we're going to do the same thing. 18:55.460 --> 19:03.740 So switch to the static sentence and ask about the mobile phone. 19:03.740 --> 19:05.900 So I also need your mobile number please. 19:08.140 --> 19:10.020 And let's rename this node to. 19:12.100 --> 19:14.180 Collect email address. 19:16.420 --> 19:18.060 This one to collect. 19:20.700 --> 19:21.460 Number. 19:26.660 --> 19:27.380 All right. 19:27.420 --> 19:34.740 Now while you are building your conversational flow, retail actually gives you a really handy way to 19:34.780 --> 19:36.500 test your agent in real time. 19:37.290 --> 19:40.650 So over here you've got the test. 19:42.930 --> 19:44.290 So let's just click on that. 19:50.490 --> 19:51.930 And here you have two options. 19:51.930 --> 19:52.650 Test audio. 19:52.650 --> 19:57.770 So you can actually you can literally talk to your agent like you are on a call. 19:58.090 --> 20:01.250 That's great if you want to hear how it responds with voice. 20:01.410 --> 20:02.370 So let's test it. 20:04.890 --> 20:05.450 Hi there. 20:05.610 --> 20:06.890 Hello, this is Sophia. 20:06.890 --> 20:07.890 How can I assist? 20:08.370 --> 20:09.930 Hi, I'm looking for a chiropractor. 20:10.450 --> 20:14.330 Would you like me to book an appointment for you with Kate, our chiropractic specialist? 20:15.970 --> 20:16.170 Yeah. 20:16.170 --> 20:18.450 So as you can see, it simulates the conversation. 20:18.450 --> 20:23.970 But honestly, what I use most of the time is the test feature, so it's just quicker and more. 20:24.010 --> 20:28.010 So when you click on this manual chart, you can manually chat with the agent. 20:31.130 --> 20:34.410 And you can start the conversation from any note in the workflow. 20:35.130 --> 20:37.550 So let's start from the welcome node. 20:42.870 --> 20:47.710 Now let's move on to the calendar availability node. 20:56.030 --> 21:01.590 As you can see, it's actually tries to call the function to check availability. 21:02.550 --> 21:08.190 This is how you can test your agent while you are building your conversational flow. 21:08.430 --> 21:14.310 So it's super useful way to debug and make sure everything works as expected before you actually go 21:14.350 --> 21:14.870 live. 21:15.830 --> 21:22.830 So now at this step, once our voice agent has captured all the key details like the name, email address, 21:22.990 --> 21:30.550 mobile number, you can use that to ask the customer about their symptoms and record that information 21:30.550 --> 21:30.910 too. 21:31.710 --> 21:37.380 So that way when the agent sends a confirmation email with the full summary of the conversation to the 21:37.380 --> 21:38.180 specialist. 21:38.500 --> 21:40.420 It can include the symptoms as well. 21:40.940 --> 21:44.700 So this helps the specialist prepare properly before the appointment. 21:45.340 --> 21:48.700 So let's add the conversation note again. 21:52.620 --> 21:53.780 And this time. 21:55.820 --> 22:00.460 We are going to add this as a static sentence. 22:00.740 --> 22:02.980 So can I ask what symptoms you have. 22:03.060 --> 22:07.660 And if there is anything else you would like me to pass to Kate. 22:08.340 --> 22:15.100 And as a transition of the previous note, we want to use the same condition. 22:15.300 --> 22:17.220 So I'm just going to copy and paste it. 22:20.020 --> 22:20.220 Okay. 22:20.220 --> 22:20.780 Great. 22:21.820 --> 22:22.580 Just to. 22:26.580 --> 22:27.340 Fix it. 22:27.340 --> 22:29.260 And it's called a number. 22:33.800 --> 22:35.240 Now let's connect these nodes. 22:40.640 --> 22:44.600 So the transition will be exactly the same as in the previous nodes. 22:46.080 --> 22:52.520 So once you answer the question move to the next node going to place it here. 22:55.120 --> 22:57.920 And then I'm going to add another conversation node. 22:59.360 --> 23:01.280 And this time I'm going to. 23:04.160 --> 23:07.680 Say something like thank you for the details. 23:07.680 --> 23:09.720 What day and time works best for you. 23:11.280 --> 23:13.360 And also let's say. 23:16.080 --> 23:16.600 Here. 23:18.640 --> 23:19.560 Ask for. 23:22.680 --> 23:23.960 Additional information. 23:28.720 --> 23:31.720 So then it will be easier to reference to this node. 23:34.830 --> 23:35.630 And here. 23:35.630 --> 23:40.030 Don't forget to switch over to this static sentence mode. 23:40.790 --> 23:43.630 And as a transition, we want to say something like. 23:46.710 --> 23:51.430 If user provides a valid date and time, move to the next step. 23:52.510 --> 24:01.390 So only if the user provides the valid date and time, move to the to the next step where we will check 24:02.710 --> 24:03.590 availability. 24:04.150 --> 24:12.910 So now if the user wants to book an appointment and provided all the relevant information such as contact 24:12.910 --> 24:19.870 details and the date and time that works best for them. 24:21.070 --> 24:25.990 We want our AI voice agent to check availability. 24:26.470 --> 24:31.890 And in that case, and this time we are going to use different type of node. 24:33.370 --> 24:34.450 The function node. 24:35.290 --> 24:36.410 So let's click on that. 24:38.730 --> 24:41.850 And now we have to add a tool. 24:43.530 --> 24:44.970 So click on add button. 24:46.410 --> 24:53.450 And as you can see retail AI has built in integration with Calcom. 24:54.010 --> 24:55.890 To check the calendar availability. 24:57.290 --> 24:58.490 So let's select this one. 24:59.090 --> 25:03.810 And here we need to provide the API key from Calcom. 25:04.250 --> 25:06.530 And event type ID from Calcom. 25:06.570 --> 25:13.930 So in the previous tutorial I showed exactly how to find the API key and the type ID. 25:14.290 --> 25:18.690 So if you are not sure or you have forgotten, just go back and check that video. 25:19.330 --> 25:20.490 So I've added mine. 25:20.530 --> 25:24.370 Just going to save it and add. 25:27.170 --> 25:27.530 All right. 25:27.530 --> 25:28.970 So this node is configured. 25:28.970 --> 25:34.040 Now let's Connected to the last node in our flow. 25:36.360 --> 25:40.280 So if user provides a valid date and time, move to the next step. 25:41.720 --> 25:42.760 So to this node. 25:44.880 --> 25:52.920 Now once the user provides the date and time and our agent is ready to check the specialist calendar 25:53.280 --> 26:04.440 to see if that slot is available, and we want our agent to sound as natural, to sound as natural and 26:04.440 --> 26:05.880 human like as possible. 26:06.240 --> 26:11.680 So on the right hand side you will see more configuration options for this node. 26:12.520 --> 26:18.760 So let's go ahead and toggle the speak during execution feature. 26:19.280 --> 26:25.000 And in this node again let's change to the static sentence. 26:25.560 --> 26:29.120 And here you have to define what to say when calling the function. 26:30.220 --> 26:35.180 so there will be no silence when our agent will analyze the calendar. 26:35.700 --> 26:36.660 So let's say. 26:38.860 --> 26:40.820 I'm checking the availability for you. 26:40.860 --> 26:42.060 Give me a moment, please. 26:42.900 --> 26:45.620 Now we are gonna add two transitions. 26:46.540 --> 26:53.980 One to check the calendar for a time that matches the user's schedule, and another to handle cases 26:53.980 --> 27:00.580 where the user asks to speak with a real person or sounds frustrated for any reason. 27:01.740 --> 27:04.300 So I'm going to add, um. 27:07.060 --> 27:13.220 Find the time on on a calendar that fits the user schedule as the first transition, and as the second 27:13.220 --> 27:13.580 one. 27:14.340 --> 27:21.340 I'm going to describe it as, uh, if the user requests to speak with a real person or express frustration, 27:22.900 --> 27:24.060 move to this node. 27:25.060 --> 27:27.780 So here we need to add another type of node. 27:27.850 --> 27:35.570 So the transfer call node so that if the user asks to speak with a real person or gets frustrated, 27:35.970 --> 27:39.570 the agent can try to hand off the call to a human. 27:40.650 --> 27:46.170 So let's click on that node and let's connect it to this transition. 27:48.530 --> 27:57.490 Now the configuration settings on the right hand side, just enter the phone number you want the call 27:57.530 --> 27:58.730 to be directed to. 27:59.170 --> 28:07.890 And if the transfer doesn't go through, we will add a fallback conversation to let the caller know 28:07.890 --> 28:10.770 and and end the call politely. 28:11.610 --> 28:14.290 So we are going to add another conversation node. 28:19.610 --> 28:22.930 Move it right here and connect this transition. 28:25.530 --> 28:32.070 And in that case we want our agent to say something like, I'm sorry, the specialist is currently unavailable. 28:32.110 --> 28:34.990 Please call later to talk directly to the specialist. 28:35.350 --> 28:37.510 Thank you for your patience and goodbye. 28:38.390 --> 28:39.990 And we want to end a call. 28:40.150 --> 28:43.270 So we are going to use the. 28:45.430 --> 28:46.910 Ending call function. 28:47.830 --> 28:48.670 So is this one. 28:53.790 --> 28:55.630 And just connect these nodes together. 28:58.550 --> 28:59.990 All right so let's zoom out. 29:01.310 --> 29:06.150 So this is the conversation flow we have built so far. 29:11.030 --> 29:11.270 Okay. 29:11.270 --> 29:12.150 So let's carry on. 29:13.990 --> 29:22.470 Now let's add one more conversation node with the condition I connected. 29:24.830 --> 29:34.290 And the a condition, we want to describe the transition as if you agreed on the day and time with the 29:34.290 --> 29:36.290 user, move to the next node. 29:39.250 --> 29:46.490 So as the user and the agent agree on the date and time, the agent should book an appointment. 29:47.170 --> 29:47.850 So let's go. 29:47.890 --> 29:53.530 Let's go ahead and add another function node. 29:54.770 --> 30:02.450 But this time let's add book on the calendar built in Chalcombe function. 30:03.250 --> 30:10.370 You can also build your own custom function, but it's more complex. 30:11.370 --> 30:12.490 So select this one. 30:13.250 --> 30:14.810 You can leave the description as it is. 30:14.810 --> 30:18.490 So when the users ask to book an appointment, book it on the calendar. 30:19.250 --> 30:24.490 But you have to provide API key from the and event type ID. 30:25.480 --> 30:26.560 so let's do it now. 30:27.120 --> 30:29.000 So I'm in my calendar now. 30:29.760 --> 30:31.160 Let's go to settings. 30:33.840 --> 30:34.920 API keys. 30:36.280 --> 30:36.680 Add. 30:40.960 --> 30:42.160 A name. 30:48.840 --> 30:49.880 Never expires. 30:50.040 --> 30:50.720 Save. 30:52.000 --> 30:52.840 Copy it. 30:56.280 --> 30:59.680 And paste it here to event type ID field. 31:01.000 --> 31:03.000 You have to go to our calendar. 31:06.560 --> 31:10.520 So back event types. 31:13.000 --> 31:15.760 I have 15 minute meeting. 31:17.680 --> 31:19.400 So I'm going to click on this one. 31:20.800 --> 31:22.600 And from the URL bar. 31:23.420 --> 31:24.500 Copy the number. 31:29.300 --> 31:30.340 And paste it here. 31:33.300 --> 31:36.500 Select your time zone and hit save. 31:39.380 --> 31:40.340 And click add. 31:42.460 --> 31:48.900 Alright, so now we have configured book Appointment node to use this function calling. 31:49.860 --> 31:53.060 And when we want our voice agent to book an appointment. 31:54.060 --> 31:55.340 So let's connect it. 31:57.900 --> 31:58.900 Now let's say. 32:02.100 --> 32:07.420 We want our agent to speak during execution and we want it to. 32:07.460 --> 32:12.100 We want it to say something like your appointment is confirmed. 32:12.140 --> 32:13.940 You will receive a confirmation shortly. 32:14.340 --> 32:14.980 Have a nice day. 32:14.980 --> 32:15.620 Goodbye. 32:19.180 --> 32:21.380 So now the appointment is confirmed. 32:22.090 --> 32:28.930 And before we end the call, we want to send the details of this conversation to Natan because we need 32:28.930 --> 32:30.450 to automate what happens. 32:30.490 --> 32:31.650 What happens next? 32:32.130 --> 32:38.970 So that includes things like summarizing the call and notifying the specialist, and sending a confirmation 32:38.970 --> 32:49.330 email to the specialist with a summary and notes from the conversation, or saving the info for for 32:49.330 --> 32:49.810 later. 32:50.290 --> 32:58.810 So we are going to create a custom function and use a webhook to pass all the conversations data from 32:58.810 --> 33:01.850 retail AI directly into our workflow. 33:03.250 --> 33:10.290 So let's create a function, click add button and create custom function. 33:10.850 --> 33:12.330 Now let's jump into Nathan. 33:13.530 --> 33:15.490 So let's create a new workflow. 33:16.450 --> 33:17.530 Let's rename it. 33:19.090 --> 33:21.830 Now we are going to add a webhook node. 33:21.830 --> 33:24.430 So let's search for webhook. 33:26.550 --> 33:29.590 So this webhook will trigger our automation. 33:30.910 --> 33:34.190 So we are going to copy this production URL. 33:36.870 --> 33:43.150 Let's go back to our retail account and paste URL in this field. 33:45.790 --> 33:51.030 Now as a description let's say sense users answer to the questions. 33:51.750 --> 33:55.710 And let's name it ten days 2010. 33:58.830 --> 33:59.870 Now let's save it. 34:01.950 --> 34:02.910 And let's add it. 34:07.230 --> 34:07.630 All right. 34:07.670 --> 34:08.110 Great. 34:08.110 --> 34:11.110 Let's add the transition to this node. 34:11.150 --> 34:14.190 So we want to send all the details if the. 34:20.100 --> 34:22.260 If the appointment is successfully booked. 34:24.060 --> 34:25.420 And let's connect these nodes. 34:31.260 --> 34:32.140 Alright, great. 34:32.620 --> 34:34.620 Now the same transition for this node. 34:35.860 --> 34:37.380 So let's add a transition. 34:39.500 --> 34:41.140 If appointment successfully booked. 34:42.780 --> 34:44.100 Now let's end the call. 34:44.300 --> 34:47.140 So we are going to use ending function. 34:52.340 --> 34:52.700 All right. 34:52.700 --> 34:53.180 Awesome. 34:54.460 --> 34:59.500 Now in the webhook node we want to change the method to post. 35:00.460 --> 35:06.020 And now we want to add a agent to analyze all this data. 35:07.220 --> 35:09.340 So let's add a agent. 35:13.220 --> 35:14.820 And let's rename it to Kate. 35:19.160 --> 35:21.360 So this is the name of our chiropractor. 35:22.280 --> 35:24.120 Now you want to add the prompt. 35:25.280 --> 35:26.800 So let's define below. 35:28.120 --> 35:28.640 Let's switch. 35:28.880 --> 35:33.600 Let's switch to expression and let's expand it. 35:34.480 --> 35:36.280 So this is the prompt I have prepared. 35:37.880 --> 35:43.200 So you are an AI assistant specialized in summarizing call transcripts. 35:43.520 --> 35:48.720 Your task is to extract only the most important and relevant details from the conversation. 35:49.240 --> 35:51.840 Always focus on the client's name and contact info. 35:52.320 --> 35:53.320 Reason for the call. 35:53.360 --> 35:53.880 Symptoms. 35:53.880 --> 35:55.000 Issue or concern. 35:55.320 --> 35:57.120 Additional notes from the client. 35:58.440 --> 36:02.680 So you want our agent to summarize the transcript. 36:03.560 --> 36:09.680 So you will get this JSON when you test your voice agent. 36:11.520 --> 36:20.190 While this workflow is active, the current date and time is now Generate a structured appointment confirmation 36:20.190 --> 36:26.590 email using HTML, ensure clear bullet points and proper formatting for readability, so this prompt 36:26.590 --> 36:28.910 will be available for you in the resources. 36:29.830 --> 36:32.550 So example output A new client has booked an appointment with you. 36:32.950 --> 36:33.790 Client name. 36:34.310 --> 36:40.830 So the extracted name from this transcript, email extracted from the transcript, etc.. 36:41.710 --> 36:47.390 So ensure proper line break breaks and HTML formatting for a clean display in emails. 36:48.550 --> 36:50.230 And we have one tool. 36:50.830 --> 36:52.670 So use email tool to send emails. 36:54.150 --> 36:54.950 Alright great. 36:54.950 --> 37:02.270 So you can just you can simply copy and paste this prompt into your agent or you can modify it to fit 37:02.270 --> 37:02.950 your needs. 37:03.790 --> 37:04.710 Let's close it. 37:08.590 --> 37:10.030 Now let's add chat model. 37:12.430 --> 37:13.870 I'm going to use OpenAI. 37:18.450 --> 37:20.090 GPT four should be fine. 37:21.090 --> 37:23.490 Make sure you OpenAI account is connected. 37:25.370 --> 37:29.370 Okay, as a memory, I'm going to add simple memory. 37:32.170 --> 37:32.890 Alright, great. 37:32.890 --> 37:36.610 And as a tool I'm going to add Gmail. 37:37.610 --> 37:38.570 So Gmail tool. 37:41.090 --> 37:47.970 We want to send an email I want to send the email to this email address. 37:49.850 --> 37:58.010 But you would use the specialist's email address so Lehman's gmail.com. 38:00.250 --> 38:03.250 As a subject your appointment booked. 38:05.650 --> 38:09.490 Now for the message field just click on this. 38:09.530 --> 38:13.210 Let the model define this parameter feature. 38:13.210 --> 38:19.070 It will automatically and intelligently write an expression for you. 38:19.070 --> 38:25.230 So GPT four will be able to define this parameter. 38:25.990 --> 38:26.790 All right great. 38:26.790 --> 38:27.910 So let's close it. 38:29.190 --> 38:33.030 And now let's make this workflow active. 38:35.790 --> 38:36.550 Confirm. 38:38.590 --> 38:40.190 Now let's go back to retail AI. 38:40.870 --> 38:45.350 Here in the global settings you can switch to other languages. 38:45.710 --> 38:51.230 You have Spanish, French, Chinese and so on. 38:52.190 --> 38:55.470 You can select your favorite voice. 38:57.030 --> 39:04.390 You can add your custom voice for example from 11 labs you can change the gender, you can change accent, 39:04.950 --> 39:05.790 and so on. 39:06.710 --> 39:07.870 Let's close it for now. 39:13.510 --> 39:19.820 And here you can change the large language model for a global prompt. 39:21.220 --> 39:24.420 GPT four is fine and is very cost effective. 39:26.020 --> 39:27.300 So let's choose this one. 39:28.740 --> 39:32.380 And here we are going to enter our global prompt. 39:33.380 --> 39:41.780 So we are going to guide this AI voice agent and provide more details on the context of the of its role. 39:43.380 --> 39:45.900 So I'm going to paste the prompt here. 39:45.900 --> 39:48.260 And you can also find it in the resources. 39:48.780 --> 39:54.180 So here Sophia, a customer support representative at Physical Therapy Clinic, offers physiotherapy, 39:54.220 --> 39:58.180 chiropractic care, massage therapy and rehabilitation. 39:58.220 --> 40:04.500 Your role is to assist potential clients in scheduling appointments with the appropriate specialist, 40:04.700 --> 40:09.540 including physiotherapists, chiropractors, massage therapists and rehabilitation experts. 40:11.340 --> 40:22.400 Always use the current date and time, and we specify our time time zone and also pass the general current 40:22.640 --> 40:28.680 date time variable and assume that all appointments are for future dates only. 40:29.360 --> 40:37.280 So as objectives assist in scheduling an appointment for physiotherapy, blah blah blah and best practices 40:37.280 --> 40:42.800 during the call, ask one question at a time to keep the conversation natural and engaging. 40:43.200 --> 40:44.160 Listen actively. 40:44.760 --> 40:50.880 Now when collecting client information, ensure to gather their name, email address and mobile number. 40:50.880 --> 40:52.080 So this is very important. 40:53.120 --> 40:57.080 And waiting for each detail before proceeding to the next step. 40:58.240 --> 41:02.440 So when presenting available time slots clearly state them in spoken language. 41:05.280 --> 41:11.750 And if the user selects an unavailable time slot, provide two alternative options one in the morning 41:11.750 --> 41:15.350 and the one in the afternoon on the closest available day. 41:15.750 --> 41:18.270 And keep your responses clear and concise. 41:18.310 --> 41:19.430 Avoid rambling. 41:20.910 --> 41:21.830 Okay, great. 41:22.590 --> 41:25.070 Here you can choose the knowledge base. 41:26.710 --> 41:28.950 So basically you can add the knowledge base. 41:29.350 --> 41:32.070 As I as I showed in the previous tutorial. 41:32.590 --> 41:36.590 You just simply go to your AI dashboard. 41:37.270 --> 41:46.870 So under Knowledge base section you can just click on this plus button and give the Knowledge base name. 41:51.230 --> 41:52.870 You can add a web page. 41:53.950 --> 41:58.590 You can upload files and you can add text manually and then hit save. 41:58.590 --> 42:08.110 And when you go to your, uh conversational flow agent, you can add the knowledge base right here. 42:09.190 --> 42:12.010 So it will have access to the documents. 42:12.330 --> 42:18.930 Here you can play with some speed settings such as you can. 42:19.170 --> 42:21.330 You can add the background sound. 42:21.850 --> 42:27.290 You can adjust responsiveness so control how fast the agent responds. 42:27.290 --> 42:35.570 After users finish speaking, you can add some interruption, sensitivity, etc. I'm not going to dive 42:35.610 --> 42:37.130 deep into the settings. 42:37.170 --> 42:44.570 Now this is really important because by adding the post analysis feature, you can automatically pull 42:44.570 --> 42:47.050 out key details from the conversation. 42:48.370 --> 42:53.610 So let's add text and click save. 42:53.890 --> 42:55.330 And here you can choose. 42:55.650 --> 43:00.530 And here you can switch between different LMS for this post call data retrieval. 43:01.370 --> 43:04.170 But let's stick with the GPT four. 43:04.690 --> 43:12.920 So now the final step in this setup is making sure that you We have a fallback plan in case things don't 43:12.920 --> 43:14.040 go as expected. 43:14.040 --> 43:17.080 So like when a human agent is not available. 43:17.600 --> 43:23.400 And for that, we are using a special type of node which is called a global node. 43:23.840 --> 43:29.200 So you can enable this feature in every conversation node. 43:30.200 --> 43:39.480 When you click on the conversation node, like any of them, you can see that there is a global node 43:39.520 --> 43:40.200 option. 43:41.400 --> 43:48.400 So you can toggle it and it will allow other nodes jump jump to this node without edges. 43:48.920 --> 43:52.400 So this node doesn't need to be connected to the rest of the workflow. 43:56.000 --> 44:03.000 The global node is basically a part of your agent that can be triggered from anywhere in the flow. 44:03.600 --> 44:11.180 So no matter when the conversation is and in this case It kicks in whenever the user says something 44:11.180 --> 44:15.660 like, I want to speak to a real person or they sound frustrated. 44:16.540 --> 44:21.700 So in this note, the agent responds with something like, I'm transferring you to a human agent now. 44:22.340 --> 44:24.980 And then it tries to hand off the call. 44:25.500 --> 44:31.260 But just in case the transfer fails, maybe no one is around. 44:31.500 --> 44:39.140 We've got a backup message ready, so the agent politely lets the caller, uh, know support is not 44:39.140 --> 44:41.980 available at the moment and ends the call nicely. 44:41.980 --> 44:49.940 So this way we can keep everything professional and user friendly and and avoid bad customer support 44:49.940 --> 44:50.740 experience. 44:50.860 --> 44:53.780 So this is our conversational flow agent. 44:53.820 --> 44:56.380 As you can see we have a quite a few nodes. 44:56.740 --> 44:59.220 And this is the second part of the workflow. 45:00.060 --> 45:08.770 So this is our global node right now for the other two specialists in our physical therapy clinic. 45:09.370 --> 45:12.970 So the massage therapist and the physiotherapy rehab specialist. 45:13.330 --> 45:16.170 The conversation flow is going to be exactly the same. 45:16.530 --> 45:23.850 All you need to do is set up a separate calendars for each of them in Calcom, and then create dedicated 45:23.850 --> 45:26.490 agents in Naa10. 45:30.010 --> 45:33.770 To handle the color summaries and send out confirmation emails. 45:34.370 --> 45:37.170 So really, it's a simple copy and paste job. 45:37.410 --> 45:42.290 Just duplicate what we have already built and adjust the names and calendar links. 45:43.130 --> 45:46.530 And of course, this setup is not just for small clinics. 45:46.530 --> 45:52.690 You can easily scale it up for bigger businesses with five, ten, or even more team members. 45:53.090 --> 45:55.330 It's flexible and ready for production. 45:56.610 --> 45:59.090 All right, I hope this was helpful. 45:59.810 --> 46:02.450 Now go ahead and build it out for your practice. 46:03.130 --> 46:06.130 Thanks for listening and see you next time.