Building an AI Phone Assistant in 2 Hours? | Vapi x Make Tutorial
Вставка
- Опубліковано 1 чер 2024
- In this video, I am going to build an AI phone assistant for restaurants in less than two hours! Using Vapi and Make, I'll guide you through creating a custom solution that can reserve tables, save transcripts, and even remember details about callers. From setting up the chatbot to integrating with calendars and CRM systems, we cover it all step by step.
Link to the resources:
hub.integraticus.com/building...
Work with us 👋🏼
integraticus.com/
My Links 🔗
👉🏻 Subscribe: / @jannismoore
👉🏻 Instagram: / jannismoore
👉🏻 LinkedIn: / jannismoore
👉🏻 More ways to reach me: integraticus.com/links/
00:00 Getting Started
00:34 The call result
01:44 What to expect
02:25 Explaining the flow
04:13 The infrastructure explained
12:58 Getting Started with Vapi
28:00 Transient-based assistants
38:00 Automate with Make.com
47:19 Create the Lead Database
55:19 First Dynamic Call
59:31 Give the caller context
01:04:32 Second Dynamic Call
01:17:00 Full demo call
01:19:45 Final Call Example
#vapi #aiassistant #aivoice #aiagents #coldcalling #vapiai #conversationalai #makecom
Killing it bro
The VAPI King himself 👑
Love it. Good job. Amazing video❤
Looks sick brother!
Excellent work !!!
Thank you so much for this video!!
You are so welcome!
Awesome bro !
Sehr nicer Input Jannis, Grüße aus Deutschland!
Thanks Jannis - This was great but wished you had spent just another 5 minutes and integrate the Google Calendar piece. Will be signing up if you can send me the link. Thanks for going over the functions as I could not really find anything on this. For another tutorial, it would be nice to know how to do add a knowledge base without using another tool or add-on. By the way, the resources are not in the Hub.
I really appreciate the input! I've published the resource slightly after it was released (It's live now).
To make up for it, I've also added a little bonus to it that includes a bonus blueprint that includes a Google Calendar integration within the function call scenario. You will find everything via the following URL: hub.integraticus.com/building-an-ai-phone-assistant-in-2-hours-vapi-x-make-tutorial/
@@jannismoore many thanks Jannis - really appreciate it. Just got things downloaded and saved - will pay with it today and let you know how it goes.
we have all the tech in the world
but show me real business with this implementation
:)
We implement dozens every week.
I’m not sure what exactly keeps you from believing this isn’t really a thing for actual businesses, but you’re most welcome to share your concerns.
@@jannismoore I never said I believe in anything. I just want to see real results in business to test the product. Not sure what the big deal is there :D
Thank you for your videos and testing this incredible system, I have a question, does it work for the Brazilian Portuguese language? and the second question is would it be possible for me to connect with a Brazilian number (+55)?
Awesome video thank you, would the cost per minute of calls be cheaper if you use Twilio or would it still cost $2 per minute?
$2 per minute is way too pricy for any ai calling platform.
Vapi charges you an average of $0.13 per minute, and in any case, even the priciest platforms won’t go over $0.65.
@@jannismoore thanks Jannis just signed up for your Acdemey 🤙
Hi Jannis, how can we connect the bot to the restaurant's POS system to handle calls and send orders directly to the kitchen? Thank you for the great content.
Hey Paulo, this mostly depends on the POS system you use and how they allow external services to communicate with it.
In most cases, they have some sort of API or Webhook integration, which you can check within their documentation.
Depending on the endpoints they offer (The different features for communication), it requires a different setup.
Feel free to share some more details about it or send me a message via integraticus.com
Hi,
great video
do you think you can integrate whatsapp ? So people can call or they can text. But all transcript go inside whatsapp ?
WhatsApp has an API so I assume you can feed in the transcript to WhatsApp. You most likely need some custom code for that.
Just amazing the knowledge high value - is there a limit for incoming calls? - let’s say 100 calls how you will set up limits and cycles?
There's no direct limitation from Vapi, but there is some on Twilio. In general, I recommend not running more than 50 calls per second as otherwise Twilio rate-limits your calls. With 100 calls, you can simply batch the calls using make.com (This is a setting when you do it using a Google Sheet table)
Thanks
Great video man, this was really helpful.
Just wanted to make a request if you could do a tutorial video for food ordering and appointment booking for Attorneys and doctor please
Hey! Do you refer to simple food ordering by leveraging the transcript or by actually implementing a third-party CRM to submit the orders to?
@@jannismoore simple food ordering
What a legend
how many clients do you have? still havent answered my qs man!
Impressive stuff" Having an issue getting the transient assistant to work, prompt is fine and got all the variables setup but the initial vapi server url / webhook => webhook response is confusing me... Can you make a tutorial please on how to get the vapi/make handshake to work? Thanks bud and keep it up
Good point. There are a couple of entry points for creating the transient-based assistant. Vapi actually released an updated version of their documentation that explains the different possibilities of the server URL very clearly. I suggest checking out this URL: docs.vapi.ai/server-url/setting-server-urls
If that still doesn't help, feel free to share some more details here
At timestamp 40:18 where you add the webhook URL. I was just looking at this. Setting the URL here means that you can only run one assistant - is there another way using any of the other server URL spots ?
If you set a URL there, all of your assistants will run through it.
You can also connect a URL to the phone number, as well as to the assistant.
@@jannismoore Ok - thanks. But then does that mean you have to have every assistant hook up to the same Make Scenario ?
I enjoyed your video and it was very informative. I like the functionality, but I was surprised as to the low quality AI speech. Is there a reason the bot sounded that robotic?
I used azure for the example. It’s not the most natural-sounding, but I like it for covering multilingual conversations.
It’s always a game between latency and quality so if you value quality more, I suggest to stick to the 11labs voices.
Please do a tutorial with Google Calendar availability. 🔥🔥🔥🔥
Added to my list! You only want the availability covered or as well the scheduling and cancellations?
@@jannismoore A full walkthrough on how to setup google free/busy calendar api, check for availabity, book the time, and even reschedule (if they were to call back and say they want to reschedule, or if we call them and they say they cant make it but want to reschedule). Ideally for an outbound and inbound call scenario
@@jannismoore A full walkthrough on how to setup google free/busy calendar api, check for availabity, book the time, and even reschedule (if they were to call back and say they want to reschedule, or if we call them and they say they cant make it but want to reschedule). Ideally for an outbound and inbound call scenario
@@HelloWorld-ii2uh Yes please
@@HelloWorld-ii2uh I got you! Will be released next week
top top top
Call comes to Vapi then to Make from Make goes everywhere else 🎉🎉 can you also add your URL for data gathering on top of maker?
Of course you can. :)
Hello professor, is it possible to use N8N instead of using MAKE?
Of course it is :)
Can you do a video on handling errors? In my experience, the AI agents in VAPI will tell me that an outbound call was successful, when in fact it was not. This creates the terrible scenario where the caller thinks an action was successful when it was not. Imagine showing up to your birthday celebration only to find out there was no reservation made!
I have a video on how to debug Vapi errors: ua-cam.com/video/u-8xdblVY_4/v-deo.html
Unfortunately this won't be much of the help for the issue you describe as that's mostly related to the right prompting. If the assistant does something but it hasn't booked the call, the instructions weren't clear enough.
I watched the video. I'm interested in the tool you will like to provide...
Which one exactly? The ones mentioned are all available via my resource hub (Link in the description)
But how do you integrate it with other calendar softwares that are not google calendar?
If they have a predefined integration on make.com or Zapier, you can use those. Otherwise you still have an option to build your own if they have an API available.
Calendly, for example, is one of the few services that does not have an API endpoint to schedule meetings, so in that case you need to become creative with browser automations.
Hey Jannis, I am getting this error on the "If it doesn't exist" google sheet:
RuntimeError
403: PERMISSION_DENIED - The caller does not have permission
Origin
Google Sheets
For context, if I ignore the error I am still not able to fetch the assistant. How should I troubleshoot this?
Now the error is: "assistant-request-returned-invalid-assistant"
You should see more details of what’s wrong within the response data of make
@@jannismoore I figured out the google sheets error, but now it just says invalid assistant (no error on make)
@@tommychryst5640 that means there is something wrong with the assistant JSON. I did a debugging bud that also explains that in depth so I suggest checking it out.
To give you a tldr: you can run the scenario ones within make and then check the HTTP module response - this one includes details about why the call failed. You can access this with the bubble that appear on the top right of the integration after it ran
In the First Google Calendar
Error
Validation failed for 2 parameter(s).
- Missing value of required parameter 'timeMin'.
- Missing value of required parameter 'timeMax'.
What Min and Max values are being requested ?
Just another question, do you want techical issues here or on the Discord Server ?
Min and Max both come from the function call variable that was set during the tutorial (The specific time). If it says it's empty, I suggest to check whether or not the webhook URL was actually included into the right function serverURL field, and if the request was actually a function call in the first place. When you press "Run" and then cause the function to fire, you can see all of the details the function sent along with and if it was a function call in the first place.
many thanks Janis. I will relook at it. It is a little confusing which we hook goes where.
45:44 You lost me here bro, what do I do with the blueprint where do I paste it
Once you’re inside of the scenario, you can click on the three dots at the bottom of the screen and then click on import blueprint.
There you can add the blueprint you downloaded from within the resource hub
The reason your functions did not transfer over is that you forgot to press save for each function. Once you save a function you can then create another property.
Yeah I noticed that too. They’ve changed the functions layout twice in the last two weeks, which was kind of unexpected. Let’s hope it stays like the one we have now.
does is handle interruption like humans?
You can set a number of words the caller (human) has to say to interrupt the assistant.
Let’s say you set it to 2, then the caller must say two words until the assistant stops talking
Why paste the whole code in the webhook response?
Because it's the easiest way to return a transient-based assistant that you can give dynamic values. IF you can live with no dynamic adjustments within your prompt, you can simply use the visual interface that Vapi provides.
@@jannismoore thanks. make the google calendar video!!
@@ILDFI Already added this to my list. I assume it'll be out next week.
Great video 🦾 how much you sell this service ?
There is no one-size-fits-all solution and therefore no fixed pricing, but if you got a rough idea, feel free to drop me a message via our contact form with some key aspects of what you're looking for, and we will come back to you with more details: integraticus.com/contact