Building A $3,500 AI Voice Receptionist w/VoiceFlow + Replit + Twilio

Watch a live demo of Sophia, an AI receptionist built with VoiceFlow, Replit, and Twilio for $3,500. She answers questions about hotel rooms, describes Vieques Island, and pitches wedding bookings with natural conversation and specific details about amenities.
Hello, I am Sophia, your virtual assistant from the Blue Horizon Boutique Resorts. How can I help you?
Hi, can you tell me more about your queen-size, uh, garden view room?
The Blue Horizon Resort offers a queen garden view room which is beautifully designed with quiet privacy in mind. This room features one queen bed with a garden view balcony.
Okay, great. Um, I also want to go to the Vieques Island. Uh, can you tell me more about the island?
Vieques Island is a beautiful destination known for its wide open spaces, country roads, and rolling hills. The primary language spoken on the island is English, and the American dollar is the standard currency. Can I book my wedding at Yes, you can book your wedding at Blue Horizon Boutique Resort. We offer special commitments to make weddings a very special event and a lifelong memory. Our tropical hotel provides a range of features that benefit weddings, including huge open spaces, fantastic ocean views, exceptional facilities, a breathtaking pavilion, and a fabulous circular bar.
Guys, what you have witnessed right there was an AI receptionist. That's right, now you can train AI to talk as well, and you can can even customize the voice so you can have it speak in an American tone or a British tone, male, female, etc. Also, you can train the AI to understand a specific set of information, for example, in this case, everything about a hotel. And now that AI is able to help all of the customers of that business 24/7, 365. I hope now you're starting to see why businesses are adopting AI at a rapid rate, because instead of having to pay a receptionist thousands of dollars every single month, they invest in this AI, pay a one-time fee, and now they have an AI employee that works for their business anytime, any day, and without complaint. And this is the power of AI. And this is why AI adoption is going so quickly. And today, I also want to show you behind the scenes of how we made this AI receptionist for one of our hotel clients. Let's dive in now. Okay, so the first step to creating our AI receptionist is to train it on all of the information that we wanted to know. And in this case, it's all of the policies and procedures related to the hotel client that we have. And to create this custom AI knowledge base, we want to use a tool called Voiceflow. And you can do this for any kind of business, by the way, from chiropractors to gyms to massage therapists to hotels to anything really that you can think of. All right, so once we get into Voiceflow, this is what we're going to see. And if it's your first time logging into Voiceflow, you want to get your own account by pressing Get Started right here. I already have an account, so I'm just gonna go ahead and press Sign In. Once I sign into my account, this is what my dashboard is going to look like, and I can see all of the chatbots that I have created. And I'm gonna go ahead and press up here New Assistant at the top. This is where I'm going to create a new chatbot here, and I'm going to call it Hotel Assistant. You can call it whatever you want, it really doesn't matter. You can even add in an image or logo right there. Under Modalities, I'm going to go ahead and press chat. And you may be wondering, but don't we want this chatbot to talk? Yes, we do. We want to connect it to a phone number, and we're only using Voiceflow as the backend knowledge base. So right here, pressing chat is more than sufficient. And under languages, we're going to select English and then press continue. And then we're going to confirm that we want to create this assistant, and then we'll be taken to the default designer dashboard that we see here. And we just want to go ahead and delete all of the default information here because we don't need any of this information. So once we have highlighted and deleted all of those boxes, we should just be left with the green Start icon. We can just leave the Designer view for now. We can see all of the different tabs that's available to us on the left-hand navigation right here. We're currently in Designer, which is the default. We want to go down to the Content tab right here. This is where we can add knowledge to our agent. And as you can see right now, our agent doesn't have any knowledge, so it doesn't know anything so far. In order to add our data sources, we can press Add Data Source in the middle or here at the top. And there are various different types of data that we can add. We can add plain text, we can upload a file, we can add specific URLs, we can add all of the sitemap or other integrations such as Zendesk. All right, so we want to go ahead and select sitemap for this project because We currently just want to train the bot on all of the web pages of the hotel. And if there are additional information that your client or the business that you're building this AI receptionist for wants to be trained on, you can always add additional documentation. It's generally pretty easy to find the sitemap of a URL. You simply go to their website, whatever it is. So let's say xyz.com, and then do a /sitemap.xml. I'm just going to go ahead and copy the sitemap for the hotel and paste it into the sitemap URL. And then the next part is a refresh rate. This is how often you want the sitemap to refresh. If you select never, it's always going to be that static set of URLs. Let's say if I select monthly, then what this means is that Voiceflow is going to crawl this site URL every single month and see if there are URLs that has been taken out or new URLs that has been added in, and it will learn all of the new information from the sitemap. I'm going to go ahead and press continue here. And as you can see, once this sitemap has been scraped, it has found 89 new URLs. And this is what we want to import. And this is what we want this, this bot to be trained on. We want it to be trained on all of the information for this specific business. Now I know ChatGPT has really been a revolutionary technology, but a lot of businesses have been asking, can I use this in my specific business? Can I train a ChatGPT that knows all of the information about my business? And yes, now you can, you can, you can leverage the language model that OpenAI has made and customize it for specific businesses or clients that you may have. And it's really applicable to any business in any industry. And And as we can see here, all of these web URLs have been loaded and the status here you can see is a green checkmark, which means that the AI has successfully learned all of the information on all of these web pages. Now before we wrap up with our knowledge base, there's just one more thing that we need to do. What we need to do is click on the gear icon here at the top. And once we click on that, we'll see our knowledge base settings. This is where we can select the language model that we wanted to The default is GPT-3.5 Turbo, which is also the cheapest. If you do want to use GPT-4 in your, in your bots, it is going to be significantly more expensive. We can also use Claude as well. Generally, I just stick with GPT-3.5 Turbo for something like a voice or chatbot. Now you're going to have 3 different levers here to be able to pull. One is the temperature, so Temperature is a value between 0 to 1. If you set it at 1, then the— you're giving AI the permission to be very creative. It can start making up stuff. If you set it to 0, it means that it's only going to stick with the information that it knows. I'm going to leave it around 0.1 so that it's still not that creative and it's going to stick with the knowledge sources that we have trained it on. The number of tokens will determine the length of each response, so the higher the amount of tokens, the longer the AI response can be. I'm going to leave it again at the default of 128 tokens. While there is complicated math that you can use to calculate how many tokens is equal to a word, and for the sake of this tutorial, we don't really have time to go through that math, but a rough rule of thumb that you can remember is roughly 4 characters equals 1 token. Okay, roughly, roughly 4 characters equals 1 token, and that can kind of give you a rough estimate of how, how many tokens you want each response to be. And lastly, here you have the chunk limit, which determines how many data source chunks will be passed to the LLM as context to generate the response. And here I'm just going to turn it up to 6. And then lastly, you have the system prompt. So the system prompt is the role that you want the AI to take on. This is the overarching instruction that you want the AI to keep in mind no matter what user interaction this AI has. And this is where you describe the role, the task that you want it to do. So I'm going to go ahead and delete out the default system prompt. Okay, great. I have put in my system prompt and you can download this system prompt in the descriptions below if you want to get this exact prompt and follow along with this tutorial. Now in my system prompt, I put that you are Sophie, a virtual assistant designed for assisting with inquiries related to the Blue Horizon Resort. Your role is to accurately address inquiries related to the resort's accommodations, amenities, bookings, and policies. As you can see, I'm describing the name, the role, the task that I want this specific AI bot to take on. For the sake of time, I won't read out all of this entire prompt to you, but you can download it below if you want this exact one. Now that I have copied and pasted in the system prompt, I'm just going to go ahead and click Save. Now this is all we need to do in terms of the knowledge base after we have trained the AI on all of the materials that we wanted to learn. The next step that we want to do is to create the connection between the phone number to the AI, and there is a lot of code involved in doing this, but the good news is I'm going to give you all of this code, and all you need to do is to copy it. So in the description below, you'll find a link to Replit. And Replit is a platform where you can host and deploy code. And it's absolutely free to sign up, there is a free tier, so you don't have to pay anything to get your AI receptionist all set up. This is what you're going to see once you click on the link that's in the descriptions. And once you have signed up for your own free Replit account, all you need to do is to press Fork and Run here at the top. After you press Fork and Run, you just need to give a name to, to what you want this, you know, quote-unquote project or folder to be called. I'm just gonna call it Voiceful Twilio Hotel Demo here. And then you just go down here and click Fork Replit. And fork just basically means copy in development terms. After I have copied over and made a copy copy or fork of this entire project setting, you're going to see the first thing that it pops up is for you to configure the missing values. And you do not want to click on Dismiss. You want to click over here on the blue button where it says Manage Secrets. This is really important because if you don't fill in the secret values, this bot is not going to work. Now in the secret values, you're going to see a value called the VoiceFlow API key. So we want to go back to VoiceFlow here And we want to get our API key. In order to get our API key, we're going to press the 3 triangles right here, and this is going to take us to the Integrations tab. Under the Integrations tab, we can see 2 different tabs, Web Chat and API Key. API Key is the one that we want, and we can see our API key, which is hidden right here. We don't have to expose that key. We just have to go ahead and press Copy API Key right here. Go back to Replit and paste in the value that we just copied right over here. Once we have copied it, we just have to go ahead and press Add Secret. And then once we have added this, what we want to do is to stop and run the Replit again. This will allow the whole project to reconfigure and configure it with the key in mind. Now, if you're wanting to customize this voicebot for your own project, there is just a slight amount of code that you need to modify, but I promise you it is really easy. From the left-hand navigation here, you'll see all of the folders and files that you're going to get once you fork this Replit. And you just want to click on source, then click on IVR, then click on handler.js. Once you have this open, you'll see this code right here. There is no coding that you need to do, don't worry. Just right over here, you're going to put in your system prompt. This is the prompt, same prompt that we just put into Voiceflow, which we clicked on content, remember we then we clicked on the gear icon and we put in the system prompt. You just need to copy and paste that prompt and put it into Replit as well. Once you have put that into Replit, you can also set a default message. This is what the bot is going to say as soon as someone calls, and you can see we set it here to say, hello, I am Sophia, your virtual assistant from the Blue Horizon Boutique Hotel. How can I help you? And if it doesn't know the response, we have set it to, sorry, I can't help you with that. Do you have any other questions? You can change these texts to whatever you want it to say for your own virtual AI receptionist. After you have modified these prompts, just go ahead and stop and run it again so that everything is saved. And that's it for this step. Now let's go on to step number 3, which is configuring your Twilio account. So to get your own Twilio account, simply go to twilio.com, twilio.com, and you'll be taken to a site that looks like this. You can get started for free, and once you sign up and sign into your account, you're going to get into your Twilio dashboard. And your Twilio dashboard will look something like this. And once you're logged into your Twilio account, If you don't already have a phone number that you want to connect this AI receptionist to, you will need to buy a phone number. If you have an existing phone number, you will have to see if it can be ported over to Twilio. A lot of providers do let you port your number to a different provider. Sometimes that does take a couple weeks though. So in order for me to get my AI receptionist over to and connected to this phone number, I'm going to go ahead and buy a phone number. So just go and click on the first tab once you're in Twilio and on your dashboard. Click on Phone Number again and then go down to Manage and then press Buy a Phone Number. From here, I can select the phone number that I want to buy, click Buy, enter in my credit card number, and once I have purchased this phone number, I can go back up to Active Numbers. And this is where I'll see all of the phone numbers that I have purchased from Twilio. I'm going to select the phone number that I want to be connected to my my AI receptionist bot that I just created, I do so by clicking on the phone number. And then from here, I'm going to see all of the configurations that I can put in. I'm going to go back to my Replit account, and from here under web view, I'm going to see a website, and I'm going to go ahead and click on the green website right here. And once you do so, you'll see a a dev URL. This is the URL that you want to copy because this is the, the link basically that will link your Twilio to your Voiceflow. You can think of Replit as the in-between. We just linked the Replit to Voiceflow by entering in our secret API key. Now we need to link the Twilio to the Voiceflow, and we do so by copying this dev URL. Going back into Twilio, and this is where we want to put in our webhook URL. It's really important here that once you put in this URL, you put IVR/IVR/launch. Okay, it's really important that you add in the last bit right over here after you have copied over the the dev URL from Replit. Okay, so again, copy this URL into, into Twilio as such, and then put /ivr/launch. And after you have done so, go down, scroll down to the bottom, and press Save Configuration. And that is it, guys. Once you have saved this, you can give that phone number a call and it will work work as expected. If the bot does give some wrong answers, what you will need to do is to modify your knowledge base, retrain it, give it additional system prompts, or feed it more information until it gets to the right responses. And if you do have a business or know someone who has a business who want us to build an AI receptionist for you, simply click on the link below and book a free strategy call with me or my team. Thank you so much for joining me for today's tutorial, and I shall see you soon.
More from Helena

How to Use Claude Code to Get UNLIMITED Leads in 6 Mins | Best Lead Scraping Tool

These AI Videos Has 21.5M Views — Here's the Exact System (Fable 5 + Higgsfield)

NotebookLM 2.0 Update: Analyze Your Entire Business in Minutes! (Full Tutorial)







