Це відео не доступне.
Перепрошуємо.
BUILDING an AI Call Center | VAPI Squads Full Guide
Вставка
- Опубліковано 17 сер 2024
- In this VAPI tutorial, we’ll be building an AI Call Center using the newly released VAPI Squads feature. VAPI Squads enables you to create teams of AI callers, making each and every VAPI assistant specialized and higher performing. The specific example that we’ll be building is a customer support AI call center, where we’ll have separate voice AI agents for transferring calls, managing accounts, troubleshooting, and appointment setting. We'll also be creating VAPI tools using Make.com to enable powerful actions for AI calling agents.
Get your questions answered with a FREE strategy session:
cal.com/shadow...
📚 Resources used in this video:
Get access to the Make, prompts, JSONs used for API requests, and more:
tally.so/r/w2rP8b
🔗 My Links -- Follow for golden nuggets outside of UA-cam!
For business inquiries, contact ibrahim@shadow-ai.co
👉🏼 Subscribe to the channel: / @theibrahimmunir
👉🏼 My Website: shadow-ai.co/
👉🏼 Voice AI Newsletter: thevoiceaihub....
👉🏼 LinkedIn: / imunir20
👉🏼 Twitter: / imunir20
Affiliates (thank you for the support!):
Vapi AI: vapi.ai/?ref=im
Bland AI (25% off first purchase): app.bland.ai/s...
Make: www.make.com/e...
⏰ Timestamps
00:00 AI Call Center
01:54 Introduction
02:54 VAPI Squad Breakdown
08:05 Example Call Flow
14:07 Setting Up Assistant Prompts
33:03 Creating the Call Center Squad
47:31 Creating the Tools for the Assistants
1:05:15 Account Tools Logic Explained
1:09:24 Account Tools in Make
1:22:26 Appointment Setting Tools Explained
1:26:48 Appointment Setting Tools in Make
1:56:24 Demo Call
00:04 Setting up AI Call Center with VAPI Squads
02:26 Creating an AI call center for customer support
07:08 AI Call Center Tools and Call Flow
09:25 VY Squads efficiently handles account management queries.
14:07 Building static assistants on VAPI dashboard
16:29 Customizing advanced settings for AI call center
20:46 Building a system prompt using GPT-3 template
22:41 Creating effective prompts for AI call center
26:57 Detailing the task in AI Call Center setup
29:05 Creating tools to access and update account information
33:25 Setting up API request using Postman
35:38 Modifying squad members and creating destinations for transferring calls
40:35 Setting up AI Call Center Squads with Assistant IDs and Destinations
42:57 Setting up API key and authorization for successful request
47:37 Creating tools through the API
49:52 Configuring tool messages and conditions for AI call center
54:40 Setting up web hooks for scenarios
56:58 Creating separate scenarios for each tool with specific details and web hook URL
1:01:33 Testing and troubleshooting make automations
1:04:14 Setting up tools on the VAPI end
1:08:34 Building an AI call center to handle account validation and updates.
1:10:54 Creating a Json formatted response for VAPI tool calls
1:15:43 Setting up filters for valid and invalid accounts
1:18:06 Updating records with the new AI Call Center tool
1:22:26 Implementing the appointment Booker with check availability tool
1:24:30 Efficient appointment scheduling with predefined time slots
1:28:39 Implement CLA model for cost saving
1:30:51 Configuring datetime and Json parsing for AI call center
1:35:40 Implementing logic for weekends and weekdays
1:38:02 Running and testing a specific module
1:42:47 Processing date and time data
1:45:20 Determining available time slots using the cloud module
1:50:07 Building a webbook response for handling available time slots
1:52:32 Building an AI call center using VAPI Squads for managing appointments and calendar events.
1:57:11 Troubleshooting a non-working remote control
1:59:25 Scheduling an in-person appointment
2:03:44 Encouraging to build AI systems and gain experience with AI callers.
This is absolutely incredible. I'm jumping careers (from a successful high profile career) to concentrate on this - and have been checking out loads of videos and tutorials (and post grad studies) - but this is by far the best explanation of this. Thank you Ibrahim - best of luck to you with this dude.
You've got another sub!
@@mabsfreeman1187 Put a smile on my face. Thank you!!
@@theibrahimmunir Pleasure!
This is amazing man! The information you gave was super high level, and I could understand it as a beginner. I really respect you for making this and I look to watching more of your content in the future. Keep it up man you're doing great!
I appreciate the kind words! Thanks for the support
Awesome! I love it. Thanks Ibrahim for putting it together as a case study example and easy to understand. Appreciate it..!!
Appreciate the support Richard! Let me know if there's anything else you'd like to see.
It even has the Indian accent for the realistic experience 🤣🤣🤣
This is nuts 🤯, thanks for putting this guide together
Of course, no problem man!
Always crazy stuff from Ibrahim, He is on another level !
Appreciate it as always Sami!
Super useful content! Thanks Ibrahim
@@stefanocondorelli314 Of course Stefano! Thanks for watching.
This is fantastic! Thanks for sharing this valuable information. I’m currently developing AI phone services for local businesses and actively researching optimal solutions. Your insights are timely as I prepare to launch my email campaign to my ideal customer profile
Appreciate it Trejon! I’ve benefited from your content as well so happy to return the favor!
awesome great video! ... thank you for doing this
Thanks Mansoor!
Great Video. What do I have to do in Postman that it runs permanently? Or is it running automatically?
Thanks Niklas!
For your question, I am not sure I understand what you mean. You can download Postman and then open the app and use it like that. Then during the project, you can just open it whenever you need to send a request. It doesn't have to run permanently, it can run whenever as needed.
Let me know if there's anything else.
Please do future video on transient assistants :)
Hey Cameron, thanks for the suggestion! I have a video on how to create transient assistants for outbound calling, but if you're referring to transient assistants in a Squad, I can definitely keep that in mind!
@@theibrahimmunir Yes sorry for squads. Use case is multilingual call center that can be reused (templatized / transient)
@@cameronyking For sure man, multilingual use cases are definitely on the to-do list. Thanks for the suggestions
I have a question; how does it work with the phone numbers? Like so I a phone number for each assistant? I just can't figure out how they actually transfer on the phone. Thanks
also when I go to squads it say I have an incorrect configuration.
Hey Aydin, so in this demo, we had all 4 assistants on the same phone number, so it didn't have to do any "real" transferring.
I haven't actually tried what you're referring to inside VAPI, so I can't say for sure 100% that it's possible with VAPI's platform. But to test it out, what I would do is to make sure your phone numbers have an assistantId connected to them (can do this with a PATCH request to update assistantId field for the phone number and then setup a transferCall tool (for the initial assistant that will pick up the call) where you can transfer to different phone numbers depending on the situation.
See the following links for more info:
docs.vapi.ai/api-reference/phone-numbers/update-phone-number
docs.vapi.ai/call-forwarding
Hope this helps.
Very nice
Thanks Brian!
I cant get the airtable and webhook integration part of it working. What can I do to fix this?
I would double check the serverUrl to make sure it matches the Make webhook URL. Also make sure the tool description is well written and clear. And don't forget to run the Make scenario when you're trying to test the webhook.
Make sure you can connect to the Airtable as well, as long as you follow the settings I have on my scenario, it should work. I would suggest fixing the webhook issue first since that could be causing some of other issues though, and then take it from there.
@@theibrahimmunir Thanks a lot for your reply. I am in the formula tab of the airtable integration but it isn't the same as in your video. I am refering to 1:11:25 .
I don't have custom webhooks whenever I try to type the formula and I was wandering if just putting it up in text woild work or not.
@@TariveAdhikari Might be happening because the webhook hasn't been tested, it doesn't know what data to expect.
I would make sure that you follow the steps from 59:48 to 1:03:54. That portion of the video covers how to test the webhooks and make sure that the webhooks are catching your data from VAPI correctly. If you do this part correctly, the webhook data should show up when you're using the Airtable module.
This is currently the greatest and most valuble tech tutorial to be found anywhere. Thank you for your time. A couple of things maybe you can advise on? Firstly I used your templates all the way through to set up a working demo. When I did the Squad patch i got a 400 error stating "property squadid should not exist" Any ideas on this?
Thanks for the support Roni!
For the patch error, try making sure the "i" in ID is capitalized and make sure it looks exactly like mine.
For the second error, I'm not quite sure -- it could be a multitude of different things. It might be something with the assistants themselves, maybe the wrong assistantId was passed in or something else.
@@theibrahimmunir Thanks so much for taking the time to reply, I really appreciate it. The second error I removed as it was down to me not naming the assistants with the same spelling and syntax as they appear in VAPI. I will check up on the other error with what you have advised. On another note, I am a Bubble developer. I have custom software that will complement this system and provide 100x value to your client offerings. I will find a way to contact you with more information on this.
How can this be made for clients? Is there whitelabeling or would it be a case of creating the system and then billing clients on the usage?
Some clients might just want the configuration of the Squad and then they take it from there (so just what I built in this video).
For whitelabeling or a full dashboard solution, this VAPI Squad could be used the AI calling infrastructures and then you just build a UI around it, usually with custom code with JavaScript (maybe could try out some no-code app platforms). Probably want to setup a database as well to store user data, transcripts, recordings, summaries, etc.
2nd option will obviously be a more costly solution in terms of price and upfront work but also once you build it once, you can easily re-sell to other clients.
Hope this helps.
@@theibrahimmunirit would be marvelous if you could make the tutorial about how to integrated it with UI (including how to build in its UI tutorial ) so we could provide it for the clients ....many thanks in advance... you are the best Guru . automatically subscribe
@@firmankrisnadi7248That’s a great idea, I was actually thinking about doing a video on how to wrap this entire product around a nice custom UI. Would be a very value packed video.
thank you ibrahim, support arabic?
Yes, VAPI can support Arabic, just make sure to change the prompt to make sure it speaks in Arabic and choose a good voice that fits with Arabic well
@@theibrahimmunir
How WE Can use it on order confirmation for "COD , E-commerce '' thank again
@@AliAssam01 You can change the prompt to make it for any type of situation like ecommerce, etc. Add in any custom functions for ecommerce with Make. I would suggest looking at my first video for some ideas of what you can do.
Jheezee asalamuleikum! That’s a good question. Program looks good so far.
Can you support me in a vapi inbound project with Funktion calling?
Sure, you can watch my first video which covered an VAPI inbound callers with function calling
However just keep in mind that VAPI is planning to remove functions and will replace it with tools
why are you using Anthropic claude and for what purpose. i didn't understand it
At the time of this video, Claude was running faster and more accurately for me. Sonnet 3.5 also was released recently, and pretty good (just as good if not better than 4o on certain tasks).
Both 4o and Claude are good choices though. If you prefer 4o, feel free to substitute that LLM in. Should accomplish the same results.
Oh i understand it but in this video in the automation check available Appointment you use Anthropic claude for create a message but you give a prompt to get a dates so i don't understand the work of Anthropic claude. Can you clear it to me?
@@UsmanFarooq-r4z Anthropic Claude which works the same way as GPT4o or any other LLM. The "Create a message" module is used to send a prompt and get the answer back from the Claude LLM. In that case, I was prompting Claude to turn my day/time into a more structured datetime.
Hope that helps.
WHAT IS THE DIFFERENCE IN THE USE OF TOOL AND FUNCTION?? CAN DO ONE VIDEO??
They are very similar, the main difference is that tools give you the ability to say certain messages before starting the tool, if the tool call is delayed, if it fails, or once it completes.