How to Build an AI Voice Agent in Under 10 Minutes (Retell AI Tutorial)

0:06In 10 minutes, I'm going to show you how

0:07to build this excellent system

0:08step-by-step.

0:10To get started, head over to Retail AI

0:13and log in to your workspace dashboard.

0:15Once logged in, navigate to Agents tab

0:18on the left sidebar and click on Create

0:20an agent in the upper right corner.

0:22In the agent creation model, select

0:24single phone for simple conversational

0:26setup.

0:30First, select your underlying LLM model

0:33under the model dropdown menu.

0:37Options include ChatGPT 4.1,

0:40Claude, Gemini, etc.

0:44So, we are going to use ChatGPT 4.1 now.

0:49Next, configure the voice.

0:51Click on the voice selector, where you

0:53can preview different voices and

0:56integrate custom voices like Eleven

0:58Labs,

0:59Artificial Intelligence, or MiniMax.

1:05Select your preferred voice, for

1:07example,

1:09Andrew,

1:10and hit Save.

1:12>> Hi, I'm Andrew, your sturdy AI support

1:14shoulder.

1:27>> Next,

1:28set your primary spoken language. You

1:30can choose different languages. We are

1:31going to choose English.

1:39Next, let's configure the knowledge base

1:42under Functions.

1:43Click Add new knowledge base and give it

1:46a name.

1:47Here, we can add our knowledge bases as

1:50web pages or upload documents, or we can

1:54manually add text files.

1:59I'll be uploading my client's document

2:01as knowledge base.

2:04And I'm going to eat a knife like HVAC

2:07Echo.

2:12This allows the agent to ground its

2:15responses strictly on your business

2:17policies and accurate technical details.

2:21You can see our knowledge base here.

2:23Click on it.

2:27Under speech settings, you can customize

2:29this experience further.

2:32Add a subtle background sound like

2:35office sound,

2:37some outdoor,

2:38fully realistic.

2:46Adjust response evenness and

2:48interruption sensitivity.

2:52Then you can fine-tune real-time

2:54customer system settings and tool

2:56settings like property rules and others.

3:10You can add data extraction after every

3:12call.

3:14You will get call summary and if the

3:17call was successful or not and the

3:19user's sentiment as well. You can always

3:21add

3:22other options.

3:30Now, let's write the system prompt.

3:33I have already made a prompt.

3:35You have to divide your prompt into

3:36clear visual sections for clarity.

3:41Identity and role,

3:43tone and style, conversational

3:45guidelines, etc.

3:48In the identity and role, define the

3:50agent's name, company name, and primary

3:52responsibility.

3:54In the tone and style, instruct the

3:55agents to speak naturally and keep

3:57responses short and concise.

4:00In the conversational guidelines, ensure

4:02the agent asks one question at a time

4:05and collects customer information

4:07step-by-step.

4:09If you want this prompt, I will leave a

4:11link down in the description.

4:15Here, to give our agent calendar powers,

4:18we'll connect Calendly.

4:23We can go to the functions and add the

4:25check calendar availability and

4:28booking option as well.

4:37So, you will need a Calendly API key. To

4:41get an API key, go to Calendly, navigate

4:44to event types, and create a new event.

4:49We are going to name the title HVAC

4:52appointment.

4:54My angle.

4:59We are going to change the duration.

5:03And you can do with this

5:06to your likes.

5:09You can adjust the

5:11calendar availability option as well.

5:35To get the API keys, go to settings,

5:37then

5:39API keys,

5:41and generate a new API key named retail.

5:45You can name whatever you

5:47And copy this key.

5:53We can click on the new supplier

5:55options, so we

5:57don't have to

5:59reconfigure the API over again.

6:04Now copy and add back to retail AI and

6:07paste your key.

6:08And you would have to

6:10enter your event ID as well. You can

6:13find the event ID in your

6:15search bar.

6:22Now we will do the same

6:25on

6:26booking

6:26booking option.

6:30We'll

6:30paste the API key

6:32and event ID as we did last before.

6:44Now we do our agent fully configured,

6:46let's test it directly inside the Web

6:48API Explorer

6:49using the run test feature.

7:02>> Thanks for calling Air Cool HVAC

7:04Services. This is Ethan. Our main office

7:06is currently closed, but I can

7:08absolutely help you get a technician

7:09scheduled. What seems to be the trouble

7:11today?

7:14>> My AC seems to stop working and it's

7:17boiling here.

7:17>> Thanks for checking.

7:19>> Can you send someone?

7:22>> Thanks for checking. It looks like your

7:24AC might need some attention. Let me

7:25take a look at our schedule to see when

7:27we can get a technician out to you. How

7:28about I check the availability for

7:30tomorrow?

7:34>> Okay, tomorrow sounds fine.

7:36>> Thanks for checking. I'll look into our

7:38schedule for tomorrow morning between

7:398:00 a.m. and noon or in the afternoon

7:42from 1:00 to 5:00 p.m. Let me get that

7:44info for you now. It looks like we have

7:46openings tomorrow morning between 8:00

7:48and 12:00 and also in the afternoon from

7:501:00 to 5:00. Would you prefer the

7:52morning or the afternoon slot?

7:56>> Morning will be better.

8:02>> Thanks for confirming. I'll go ahead and

8:03book the appointment for tomorrow

8:05morning between 8:00 and 12:00. Please

8:06hold on a moment while I g at that

8:08scheduled. It looks like I need your

8:10name, email, and phone number to

8:11complete the booking. Could you please

8:13provide those details?

8:18>> My name is

8:19M.

8:22My email is

8:24at gmail.com.

8:27>> Thanks, can you also tell me your phone

8:29number so I can confirm the appointment?

8:31>> 7306354548

8:37>> Thanks,

8:38can you also tell me your phone number

8:39so I can confirm the appointment?

8:41>> 7306354548

8:50>> Thanks,

8:51can you also tell me the service address

8:53where the work is needed?

8:54>> Miami Beach

8:57>> Thanks, now would you prefer tomorrow

8:59morning between 8:00 a.m. and noon or

9:01the afternoon from 1:00 to 5:00 p.m.?

9:04>> 8:00 a.m. will be better.

9:07>> Thanks, I'll go ahead and book the

9:08appointment for tomorrow at 8:00 a.m.

9:10Please hold on a moment while I confirm

9:12that for you. Great,

9:13your appointment is booked for tomorrow

9:15at 8:00 a.m. Our technician will call

9:17you when they're on their way. Is there

9:18anything else I can help you with

9:20tonight?

9:20>> No, thank you.

9:23>> You're welcome,

9:24have a great night and we'll see you

9:25tomorrow.

9:26>> So our agent is absolutely we can also

9:29fine-tune the agent to make it better.

9:34And we We buy a

9:36new phone number for our agent to

9:38connect and make it live.

9:41Or we can use our existing SIP

9:44if you already have one.

9:46You can connect it as well.

9:51Let us give a call.

9:58>> Thank you for calling Air Cool HVAC

10:00Services. Our office is currently

10:01closed, but I'm happy to help schedule

10:03your service appointment. How can I help

10:05you today?

10:07>> Yeah, my AC is not working from last

10:09night.

10:12>> I'm sorry you've been dealing with that.

10:14Let's get you scheduled. May I have your

10:16full name?

10:18>> Sanim.

10:21S A N I N.

10:24>> My apologies, I missed that. Could you

10:26please provide your full name?

10:28>> My name is Sanim.

10:31>> Thanks, Sanim. What is the address where

10:33the service is needed?

10:35>> Brickell Avenue,

10:37Miami.

10:40>> Got it. Brickell Avenue in Miami. What's

10:42the best phone number for the technician

10:43to reach you if needed?

10:46>> 73

10:4706 35 45 48.

10:53>> Thanks, Sanim. I have that as

10:54736-354-548.

10:56May I also have your email address for

10:58the appointment confirmation?

11:00>> sanim@gmail.com

11:05>> Got it. That's sanim@gmail.com. Do you

11:07have a preferred day or time for us to

11:08come take a look at your AC?

11:11>> Tomorrow morning would be better.

11:18>> Let me check our availability for

11:19tomorrow morning. I'm sorry, we don't

11:21have any openings tomorrow morning.

11:22Would Monday, July 6th between 6:00 a.m.

11:24and noon work for you?

11:26>> Yes.

11:28>> All right. I have you scheduled for

11:30Monday, July 6th between 6:00 a.m. and

11:32noon. Would you like me to book that for

11:34you?

11:36>> Yes, all good. Well,

11:39>> You're all set. I have you scheduled for

11:41Monday, July 6th, between 6:00 a.m. and

11:43noon. The appointment is for Santa Met

11:45Brinker Avenue in Miami. If anything

11:47changes, the technician can reach you at

11:48736-354-548.

11:50Thank you for calling Air Cool HVAC

11:52Services. We look forward to helping

11:54you. Have a wonderful day. Thank you for

11:56calling Air Cool HVAC Services. We look

11:58forward to helping you. Have a wonderful

12:00day.

12:05>> So,

12:09and the booking has been confirmed, our

12:13AI is working properly.

12:16We can see our booking

12:18right here for Monday at 7:00 p.m.

12:22We can fine-tune the prompt to make the

12:24agent even better.

12:28So, that's it.

12:30And that's it. And in other terms, you

12:31have built a fully autonomous AI agent

12:34that can answer calls and do customer

12:36interviews and book appointments like in

12:38your calendar 24/7.

12:42If you have found this tutorial

12:43informative, hit the like button,

12:45subscribe for more AI automation, and

12:47leave a comment below if you have any

12:48questions. Thanks for watching, and I'll

12:51see you in the next video.