How to train your AI on important Keywords | Vapi Tutorial
Вставка
- Опубліковано 7 лют 2025
- In this video, we'll tackle the common issue of AI misinterpreting brand names, personal names, and other important keywords.
We'll show you step-by-step how to use Deepgram's keyword boosting feature to ensure your assistant recognizes the correct terms consistently.
Watch as we demonstrate the process using the Vapi platform, including setting up API calls to update your assistant's transcriber settings. This tutorial is perfect for developers and AI enthusiasts looking to enhance their voice assistant projects.
Try Vapi (my coffee thanks you!)
vapi.ai/?via=j...
My resource hub:
hub.integratic...
Work with us 👋🏼
integraticus.com/
My Links 🔗
👉🏻 Subscribe: / @jannismoore
👉🏻 Instagram: / jannismoore
👉🏻 LinkedIn: / jannismoore
👉🏻 More ways to reach me: integraticus.c...
#aiagents #vapi #aicaller
Love it Jannis, this sort of control seems extremely important to increase the user experience 👌
Was Waiting for This -Thank you so Much!
Thank you for this!!!! Love all of your Vapi content
Jannis, we need a video on pricing models - monthly, rev share, per appt booked, one time set ups, etc... and share some context and case studies of usage how it's worked for your clients or people you know using Voice AI agents to get results.
Amazing! Vapi also released a "Blocks" feature which is related to your conversational Pathways video that you posted a few weeks ago. Would love to see a tutorial on that as well :)
Yep, and you'll see a video about that very soon. :)
Thanks jannis... You are the best
Jannis is the ONLY way to go when it comes to Voice Callers and Automation.
I highly agree
Thank you Jannis. I'm currently trying to find a solution for transcribing foreign names correctly so that the Assistant can repeat them correctly. That's a step in the right direction. Unfortunately, it would probably not be possible to save and weight all foreign names here. However, I am optimistic that a solution will be found in the near future.
You could try to leverage a custom LLM with a name guesser. You could see that like a separate prompt that takes the utterance in account along with the caller language/locale to try to determine the name properly.
This would slightly delay the assistant response, but could yield more accuracy.
I haven’t tested that - just a thesis I came up with
Interesting - and thank you for making us aware of the keyword boosting / supression. however, 1 thing though: I would expect to pronounce EZLivin as Eazy Livin or E-Z-Livin, not as-livin as it is pronounced right now. If that was the case, how could we do this?
In that case, you’d just write the name as you speak it, use that as a keyword boost, and then do a search replace within make.com before saving the transcript
Good video bro
Jannis do you have any videos that focus soley on mastering the prompts in Vapi
I do: ua-cam.com/video/Kw8uQZdJ57s/v-deo.html
Hey! Love your content. Just wanted to know if I want to create backend using vapi which will serve my frontend. How can I achieve that? I also want to track the events. Tracking events is not supported by python sdk yet. So implementing that is not possible. Let me know how can I achieve it.
Can you tell me more whether you're talking about a custom backend using their SDK or some separate framework on top of it? Didn't get that part
@@jannismoore A custom backend, which can serve to multiple frontends. Basically making APIs on top of VAPI to serve my applications
Question: Can Vapi integrate with Botpress? It would be greatly appreciated if you could create a video to demo it.
I guess it could. I barely worked with Botpress as of now. I know it works with Voiceflow though
Hi Janis, love that your channel focuses on voice AI content! I’ve been using Bland AI since February & founded an AI voice company called AuroraVoice AI for phone agents 5 months ago! I am now building the full-stack SaaS so that clients can manage their agent, and I wanted to know if you would be able to share the systems for client fulfillment like how you help the client in a professional way like when they have issues and you sign a huge client? Thank you so much if you would be able to share this or speak about it? Solution Fulfillment and like processes diagram for creating it, dealing with the client and getting them set up? Do they get their own API keys or do you guys cover it in the cost? Also confused on how much to charge and I need like to see a model of how to structure it from other agencies :/ haven’t been able to get the systems yet. I am niching to luxury spas in California
Hey, your questions seem that you really want to go into selling those solution (Awesome!).
You might want to check out the Accelerator of Liam Ottley - I guess this will give you more then enough details to show you how you can get started :)
Do you still prefer gpt 4 or should i go with 3.5 turbo?
Hey man! I think I prefer the Gpt 4, it's features are amazing!
GPT4o is great for the assistant itself, even though the performance is slightly worse than with 4.
For knowledge base retrieval (as a tool), I still prefer 3.5
Hey Jannis,
Do you have a strategy to help the Vapi agent collect email address(es) accurately? When I’m collecting the data for a cal API booking request, it has to be name and email.
Even when I have the agent ask to confirm, it has a poor accuracy rate. This is very problematic for appointment booking. I need some accurate way to make sure I’m getting the emails properly logged. Any ideas?
The best case is to use an SMS for email verification.
You can do the booking over the phone and then send a confirmation sms with an additional booking link
One guy just told me that AI Voice Agents are illegal because they are robocallers.
Is that true?
Could you make a video with vapi and google calendar for making appointments
Already on it :)
Great, thank you
You guys need to understand one thing ! The multilingual side, the performance is not the same then English. We still have problem for booking appointments because the ai can’t even say the numbers correctly or name etc. That’s unfair when I see theses updates apply only in English voice
That’s what you have squads for :)
Also: My community is mainly English speaking so it’s more relevant for me to make English content.
We do have clients in other countries, and using dedicated, language-based agents, we honestly don’t have many issues with booking scheduling
@@jannismoore If you take a look at the community on vapi, you will see there is not really much videos on other languages than English. Why is that ? Because they have so many issues about the voices, the functions, cause the ai hallucinate ( ai talk to himself in another language etc ) or miss pronouns certains keywords for appointments. Only the English version is really good.
@@Emilie-974 We have German assistants that do 200-300 calls a day and they have a 90%+ success rate. If you refer to major languages, I honestly don't see too many issues. Of course, recognizing names etc is always a big thing, but you can handle a lot with the right prompt engineering
One guy just told me that AI Voice Agents are illegal because they are robocallers.
Is that true?
It’s as true as his answer is false.
There are illegal parts to it, e.g. if you use it for cold calling in certain countries.
Inbound is usually perfectly fine, but even outbound is ok if you have some sort of permission to call the user.
@@jannismoore Is it legal if I find the user number in lets say Apollo and then upload those numbers to CSV and then AI Voice Agent calls those numbers?
@@lmnnf Only if it's a B2B situation, not a consumer. Businesses can call other Businesses but because businesses get marketed to a ton, they get so annoyed and won't even engage any AI callers and hang up. So, take that for what it's worth, it won't apply across the board, but many SMS or Voice calls to cold B2B leads will hate this approach. Remember, volume will work over time. Just like cold email - people hate it, but if sending enough, you will get results.
Can I sell it as a service
Of course :)