Piotr Żaczek, Founder at Calloptima | BotCast #15
Meet Piotr Żaczek, Founder at Calloptima, who joined the BotCast #15 to answer our usual questions on building with Voice AI.
Who are you?
I'm Piotr. I'm building AI agents and voice agents for enterprise, especially for larger scale. And I've been doing it for the past two years. And I love it. It's an awesome journey.
How did you end up using voice AI?
So, I wanted to build my startup since I can remember. And I thought I could replace the interviews that people have at Google or Amazon or other FAANG companies and make it with AI. And this is where I started. I built everything myself and invited friends in engineering to try it. And it was an amazing journey. It led me to where I am right now, to incredible interviews. And I'm very grateful for being able to fail at that time.
A moment of joy and pain using voice tech.
The biggest moment of pain was realizing that pretty much nothing works out of the box. I heard over and over again: if I just use X, it's going to work. And after five minutes I realized that it just does not work.
And the joy is in finding the right solution for your particular use case. It differs from geography to geography, and from the users that are using your voice AI application. There are so many different factors that you should take into account. And I find building the product very exciting: making sure that you truly understand what the intent of the functionality is, and that it functions well within the business context as well. I think it's something that a lot of the narratives really miss, probably because we often don't go deep enough into understanding the business context.
What lesson would you share with other builders?
I think all builders should try to build voice AI themselves first, without AI. I really strongly think that, because if you build it yourself, if you're able to code obviously, then you realize what exactly is going wrong under the hood. If you don't have a solid foundation, you trust the LLM that whatever it says is true, and even more so nowadays. And it leads you either to wrong solutions, or to not the best solutions, or to not understanding what might change in your particular use case in the future.
I feel like this lack of foundation is really, really painful. And I hear it from customers as well. I often hear: I just need to build an agent on this platform and it's going to work, right? I can just port it somewhere. And what do I do here? Do I spend an hour trying to translate: okay, I think this and this and that is what you need? Or do I say: you can trust me, because I've gone down this path and it's a painful one, but I know what to do right now.
I think you really learn a lot by building yourself. Even though there's a lot of content online saying just use this, it will help you, try thinking on your own, using your own brain, and not what someone else is trying to put into your mind. Because often you're going to see that the best solution may be a very, very simple one in your case.
Where do you see the industry in 12 months and in 5 years?
I think in 12 months we'll see models being really good at understanding speech in terms of addresses, numbers and the annoying things that speech-to-text still sometimes gets wrong. I think the models are going to get much better, and everything is going to become that cheap. I'm pretty sure about it. And I'm pretty sure we're going to see speech-to-speech models that just work very, very well without a lot of additional work.
What's not going to change is the fact that you still need to build on top of the existing systems. So I think you'll get to 98% much faster, and the last 2% is still going to make a huge difference for the customers. This is where engineers will play the biggest role. That's my belief. And there will be much bigger coverage of the regions that are unsupported right now: if you're outside well-populated countries, voice AI just doesn't work. I think that's going to change.
In five years, I'm pretty sure we'll be like: okay, this is the observability, this is how your voice agents operate in this region, and it's one prompt. You understand everything that happened within this small rural area, and it gives you solid insights into how everything is working. You're going to have self-training loops that are really solid, and models even cheaper, basically for pennies. And I think you're going to have local voice AI that works really, really well on your device. On-device voice AI is so underdeveloped right now, and it's going to be a huge leap within the next five years. I'm excited for that time.
CALLOPTIMA
BotCast powered by https://bespoken.ai 🫶
