Shraey Bhatia, Co-Founder & Chief Scientist at Leaping AI, a pioneering company revolutionizing enterprise businesses, call centers, and sales with self-improving Voice AI. Founded in Germany in August 2023, Leaping AI is part of Y Combinator's prestigious W25 batch, driving the future of AI-powered communication.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of the AI Minds Podcast, Shraey Bhatia, Co-Founder & Chief Scientist of Leaping AI, shares his journey of building cutting-edge voice AI solutions that enhance and automate call center performance.

Shraey discusses his transition from academia in Australia to launching Leaping AI in Berlin and eventually scaling operations in San Francisco. He recounts the challenges of building AI-powered voice agents that not only handle customer interactions but continuously improve through self-learning capabilities.

Originally focused on automating voice interactions, Leaping AI has evolved to deliver enterprise-grade AI solutions that enhance customer engagement and streamline call center efficiency. Shraey explains how their AI-driven platform outperforms traditional systems, offering higher accuracy and adaptability in real-time conversations.

The discussion explores the hurdles of voice AI development, scaling from proof-of-concept to production, and the complexities of deploying AI in regulated industries. Shraey also shares insights on voice AI pricing models, integration strategies, and the future of AI-powered customer support.

Throughout the episode, Shraey highlights Leaping AI’s approach to pushing the boundaries of voice AI technology, leveraging the latest advancements to revolutionize the customer service landscape.

Show Notes:

00:00 "AI Innovations Podcast"

06:07 "The Slope of Delusion"

08:03 German Market Disruption Insight

11:12 Early Adopters Bet on New Solutions

14:33 Dissatisfaction with Vodafone Sales Tactics

17:20 AI-Driven Self-Improving Call System

22:14 European vs. American Business Dynamics

24:28 Optimal Team Locations: Europe & US

27:05 "Self-Improving Voice AI"

More Quotes from Shraey:

Transcript:

Demetrios:

Welcome back everyone to the AI Minds podcast. This is a podcast where we explore the companies of tomorrow being built AI first. I'm your host, Demetrios. And this episode, like every episode, is brought to you by Deepgram, the number one text to speech and speech to text API on the Internet today. Trusted by the world's top enterprises, conversational AI leaders and startups, some of which you probably have heard of like Twilio, NASA, Citibank, and my favorite place to listen to podcasts, Spotify. In this episode I am joined by the co founder of Leaping AI Shraey. How you doing today, man?

Shraey Bhatia:

Quite well, how are you?

Demetrios:

I'm good. I'm excited to talk to you because you've had a stint in Germany where I currently reside. We were just reminiscing about the good times there before we hit record and so I know you were born in Melbourne. You for some reason decided to pick up, travel around the globe, probably spent a day traveling just to get to Berlin and did academia in Berlin. What made you want to go that route?

Shraey Bhatia:

I was born in India and then my family moved to. To Australia when I was younger and then after that after my bachelor's I took a gap year to travel Europe. So I was just traveling around Europe and the gap year was working in Germany for a bit for a bartender, to be honest, at that point. And then later went back to Australia, went into academia, did a PhD and then wanted to build a startup, came back to Berlin. I heard there was a cool scene. I wasn't keen in the U.S.

Shraey Bhatia:

at that point, let's go back to Berlin, met my co founders and then we started a journey out of Berlin. Wow that's what. How we ended up in Germany.

Demetrios:

And you were doing academia in Berlin also or you were just. Still.

Shraey Bhatia:

No, earlier I was doing academia in, in Australia, but then Covid came and Australia had pretty strict regulations. See how I heard I could carry on doing my PhD from any part of the world. So then I moved to Berlin and in Berlin, yes, I did meet other researchers as well and did collaboration around. Around Germany from other universities.

Demetrios:

So.

Shraey Bhatia:

But officially it was in Australia. But during COVID I spent a lot of time being in Berlin, so everything.

Demetrios:

I said about you in the introduction was not correct. I'm glad you set the record straight and I want to know what were you working on?

Shraey Bhatia:

It's interesting because we started, I started actually quite early. It was 2016 or 2017 when AI and LLM was not even like LLM was not even the norm. It was like very basic classification and people don't remember, but there used to be frameworks like Theano which if you write a code like it will take you 400 lines for something which is written in one line at the moment. So that's where we started. And I was working on some at that point, Gans were the cool thing.

Demetrios:

Yeah.

Shraey Bhatia:

And then later Bert came out and I pivoted to pre training of models and eventually I was working on fact checking and hallucination. So a lot of claims were being made and which were not fact checked. And then we said, how do we do that? And then later, last part was hallucinations because that's like LLMs were coming out but it were hallucinating. So how do we try to not conquer it, but at least guardrail around it?

Demetrios:

Incredible. That's one of my favorite pieces of the whole LLM journey. I made a shirt that says I hallucinate more than chatgpt when chatgpt first came out. Because I gotta say, the researchers have the best naming conventions in this space. Like catastrophic forgetting. Come on, where do you come up with the idea of that is what we're going to call it? We're not going to just call it the LLM forgets and it never remembers. We're going to say this is catastrophic.

Shraey Bhatia:

So I have an interesting story on that. I remember the first time I heard the word, I was coming out of a techno club in Berlin. And then in the morning I was talking to a researcher. So I let's talk about hallucinations. And now I'm confused. Which one is he talking about? I like the terminology though.

Demetrios:

And what made you want to start a company? Because I imagine you could have gotten a nice cushy job at one of these big companies. The AI researchers are some of the highest paid employees at these companies. And that feels like it would have been pretty on par for your background.

Shraey Bhatia:

That's absolutely right. I asked a couple of my friends in Australia if you want to start a company with me. And their answer was, we can have a really nice life on a beach in Sydney. Why would you want to do that? I don't even make sense. I think it just comes to a fact where I just want to try different things in life. I kind of like felt like in terms of academia, I reached kind of like the highest point what I could in terms of education. I could work, but it won't be challenging enough.

Shraey Bhatia:

Before that I had a whole world travel, so I did that. Now what's left in something which I want to really explore? And that was right. Like I never, I didn't think about starting a company. So that just naturally was right, let's do something which I haven't done it before just to challenge myself. But I didn't expect that. Now it's going to be a one and a half year journey already in it's way harder than I thought. But that's how it is.

Demetrios:

That tends to happen. Have you seen that graph? I think it is the. What is it the slope of delusion or something where you think, it can't be that hard. And when you start getting into it, then you start to realize, oh my God, this is kind of hard. And then when you're really in it, you just are like, this is impossible. Why did I ever think that I could do this? What? Nothing. I don't believe in myself anymore. And eventually you start to make progress and you lose that arrogance and you start to think like, okay, I think I kind of got it.

Demetrios:

Where are you on that slope right now?

Shraey Bhatia:

I think the first year was pretty hard because we actually started in 2023. August. That's where we went all in. Kind of like people quit their jobs and said, we're going to go in. And at that point, probably we were one of the early voice AI companies, but then whenever we made a pitch, no one could believe it that it's going to work. So we said, your IVR systems are not going to work. We're going to change that.

Shraey Bhatia:

They're right. that's going to have five years away. So that was the harder. I thought they thought we were delusional. It's five years away. You're trying to sell us something, some black magic at the moment.

Demetrios:

Yeah.

Shraey Bhatia:

So from there I think first year was a bit of a struggle, but now I feel we have got a pmf, we can just go and capture the market. So that has changed. But for 16, 14 months we're right. Okay, guys, what are we doing over here?

Demetrios:

So did you start the company with an idea? Was it that you saw in the market that something was needed and then that encouraged you to start the company, or did you just know first that you wanted to start a company and you would figure out the product?

Shraey Bhatia:

I reckon for this I'll give All the credit to my co founder and CEO, Kevin. So he was working at BCG in Berlin at that point.

Demetrios:

Uh huh.

Shraey Bhatia:

And he was reading some reports from big enterprises. And this is around the time also inflation started going off the roof in Germany. You know those times when, the gas prices and we had all this stuff going around. And then there was this whole thing that, we are having this big issue that we are not able to make profits. And then he was reading these reports about different companies and that's where it struck him that, this is a market which can be disrupted. And there were a lot of call centers, which is kind of interesting because you cannot outsource in Germany the same way you can outsource in the US because not many countries speak German. So that's where the idea came. you cannot outsource it to Asia for a cheaper price.

Shraey Bhatia:

Whereas in English that is an option. In Spanish there's an option to an extent. And that's where it struck. Okay, this is a market which can be disrupted. And that's how we started building the stack up.

Demetrios:

So you saw that that was something to be disrupted. Then what? Then it was like, the product, you had to go and interview people to see if that was an actual problem or walk me through the formation of the company.

Shraey Bhatia:

I think Kevin decided to build the first, I don't know, first very easy going demo which was now, if I look back, it's laughable, because it had a six second lag between one person speaking and the other person replying. And this is why around 3.5 came out. And it was pretty dumb as well. So he made the first demo and then we both of us, the other two co founders came to the team together in August. He made a demo in June and then pitched it to a big wine customer in Germany. And then the guys were like, all right, this could work.

Shraey Bhatia:

And then we joined together hands and we said, let's build this up. And the initial idea was we'll do voice AI for sales, which was our pitch. And then everybody laughed on it. They're it's not possible. You cannot do sales with an AI. Which to an extent is even true for now. But that was the original idea behind it. And then the first pitch was just a figma design with a voiceover.

Shraey Bhatia:

And then in August we came together and then started building the product August 2023. And then it took us four months to build the first version because it was not a service. But we built the whole platform. So then January 2024. Okay, we are ready to just go for it.

Demetrios:

And you had someone in Germany that had already given you money and you decided to build it. they kind of give you a handshake deal.

Shraey Bhatia:

It was interesting because we were competing with the big ones. I mean, they're. Without naming them, they're big companies in Germany. And then they were like the US ones coming in. But somehow we did this right. Every week we had a call, we're like, what are your needs? We come back with a better demo. Just improving, like it trading with them.

Demetrios:

And so unintentionally or without these folks knowing it, they were a design partner.

Shraey Bhatia:

They were. I mean, even till late. We actually, every new feature we launch, we first launch it for them. I mean, because credit to them, in a country in Germany where it's slightly slow moving, they were the first ones to take a bet that, this could be the next big thing. And they didn't go for the older nlu press one, press two players. But they said, let's go for a solution which doesn't exist. It exists on paper, And then, they did give us a small check and they said, if you can build this up like a pilot money, and then we can go for a SaaS if it works.

Shraey Bhatia:

And that's when we write. Okay, guys, I'll sell some of my Nvidia stock, I don't need salary, and let's try to build this up.

Demetrios:

Before it popped, this was.

Shraey Bhatia:

No, it was around the point, like somewhere in the middle where it didn't go absolutely crazy, but it went somewhat up and I'm right. Guys, I have some cash now to go in for six months and let's build this up and see how we go.

Demetrios:

That's incredible. Well, now's a good point to talk a bit about the product itself before we go on to like chapter two of the journey and when you move to the US and all that fun stuff. But what exactly is it doing?

Shraey Bhatia:

So the idea, it's actually we first line. We always claim It's a voice AI which works for enterprises because there's a lot of demos which do not end up working in a lot of edge cases. And the problem with enterprise is 98% is not enough. 97% is not enough. Like 99.5% is not enough for them. So it has to take care of all the edge cases. So the product is It's a platform where you can design your voice AI agents.

Shraey Bhatia:

And the way we envisioned was it's not going to be one prompt, it's going to be different agents working together to reach a bigger goal. So that was the vision number one. And that's how we develop the first part of the product.

Demetrios:

And it's primarily for when me as a customer calls the company or is it for the company and internal, how does it work? Like where does it face?

Shraey Bhatia:

It's B2B. So we sell it to other enterprise and then they're obviously if they're a bigger enterprise, they need a bit of white glove treatment. But the idea is they can design, it's a drag and drop builder. They can design their own how you want to act the voice AI and then they just can deploy to a number or if they already have their SIP lines, we just connect them. And then you as a customer, when you call it up, it connects to our tech, I guess.

Demetrios:

Sorry, if I didn't explain that question enough. Is it primarily for support use cases?

Shraey Bhatia:

So going back, circling back, we start for sales. And then we figured out, we had our first YC interview at that point and we figured out no, sales is not going to click it. So we went for support with an upsell feature. So people are support. And then we thought, upsell is a market even when people are coming for support.

Shraey Bhatia:

Let's say someone is saying where is my delivery? But then, this is on a special. Do you still want this on your order? So that started resonating with people that, there's some sale components. It was support with some upsell. That's how we was the initial part of the product.

Demetrios:

So the interesting thing there, which I a hundred percent agree with, anytime I call Vodafone, I get first of all the worst experience. Let's just put that aside. So they should be using leaping, but aside from that they always are trying to sell me on the new plan. And it's like you're paying too much. Why are you paying too much? We can get you a better deal and we'll even send you out a phone. And so it's I don't know how you're making money on you getting me on a cheaper plan, but okay, I'll take it.

Shraey Bhatia:

Fair.

Demetrios:

That's fair. Now let's move on to chapter two and you uprooting yourself from Germany and moving to San Francisco. What was the impotence for that and how was that process.

Shraey Bhatia:

So I'll go back to the point where we entered the German market. We started gaining traction a little bit, signed a couple of big enterprises. And then we were about to write. should we go and raise some cash? In Germany as well. And that's where we figured out that it's kind of a different vibe. What is going in Germany or in Europe.

Shraey Bhatia:

Then there were some regulations coming in. AI act was coming in. Oh my God. It wasn't that conducive. Even though Berlin is a really cool tech scene, you meet a lot of smart people. So at that point, like Kevin said, hey guys, let's. Why not, this is October 2024. Let's.

Shraey Bhatia:

Why not just go and spend a month in the Bay Area? let's fly there. we might as well die, whatever, Let's just go and do it. So we came here in October last year just to spend a month. And then we just. It clicked. We figured out everyone. Even in. When I go to a pub, you meet so many smart people.

Shraey Bhatia:

You're resonating ideas. You're talking different stuff. You go to an event, you meet like literally there was an event by Deepgram, I remember once or somewhere and you meet people like the ones I'm talking on disc with Discord. You meet them there? I'm I was just having a chat with you on Discord. You were just here, fair. So that happened, you came back and then you said, let's try applying in Y Combinator again. it's been a year.

Shraey Bhatia:

Let's see. Maybe it's a different story. We applied as luck was it. We got, we were selected and then we read. Now there's a clean ticket to just move here. We are part of the YC Winter 25 moved here in January for the program. Still waiting for our own.

Shraey Bhatia:

And that's the plan now to build it out of here. Though we do have a strong German presence still.

Demetrios:

And is it primarily a tool for German companies or is it for any enterprise?

Shraey Bhatia:

No, it is for any enterprise. I mean the idea was we start from Germany because we were there, but now we already doing Italian, Spanish and a lot in the US So we are moving big in English as well. So that's how it commenced. And now we're building the part two of the stack, which is the big thing, which is it self improves. So you have a call, you get whole history of calls. The AI evaluates the call and Then it improves itself for the next call. So that's where we are. That's where the idea came in when we were here in SF while talking to a people where, this part of the stack no one is handling.

Shraey Bhatia:

We are automating the call agent, but we are not automating the supervisor and Q and a guy. Let's just take that out of the equation as well. And that's what is now leaping AI as I say, part two happening.

Demetrios:

And is it that you're giving the agents a goal and then they just figure out the best way to get there through many conversations. Are you doing any kind of simulations? What does that look like?

Shraey Bhatia:

You're right on that. So we do have some evaluation metrics, some are more general and some are agent based. Whatever the goals might be, it evaluates something on a bigger model. Using an O1 or a deep seq and then it critiques itself, finds the violations, what happened? And then it back propagates. Eventually, after a bunch of calls which attack together on a similar issue, it back propagates, improves the prompt, then does simulations again to see if it has improved and thereby for the next few calls then it's trained itself. Not trained, but actually improve the prompts itself to know how you're going to improve it. And then you run an A and B test continuously.

Shraey Bhatia:

To see if it's actually improved or not.

Demetrios:

That is fascinating. And that feels like something that many folks will want to plug into their own agents. Maybe not just necessarily on your platform. Is that a product direction that you're thinking about where, I have a voice agent and I just want to have it self improved. Can I plug in your self improvement SDK or something?

Shraey Bhatia:

We haven't thought about that because we never thought it. we never were thinking as a dev tool or an. We don't even have an API. So we always thought about it as an enterprise product. We don't even have self serve. So that's the direction we went through. And then what happened was as we started scaling to 10 to 15,000 calls a day, I can't hear the recordings and like improve the prompts for my customers. I'm like it's a pain point for me.

Shraey Bhatia:

I can go and hear read every transcript to see what's going wrong. So do I hire four more people to do that or can we hire one person who's going to build a system with me so that it does it itself? And that's where this idea came. Just to take some burden out of us that because enterprises wanted the prompt improvements happening every now and then and that's why it came as part of the product, which is now rolling out in beta versions.

Demetrios:

Now, you mentioned one of the strong value props for the platform in the beginning was because you were selling to German companies and for German companies, it's very hard to offshore in the German language. I imagine if you're selling to US companies, you can no longer use that type of value prop. Do you see that it's easier to sell to German companies because of that? Do you have stats around it? Is that something that you later said, you know what, actually that was an assumption that we validated was not true.

Shraey Bhatia:

So what happened over here though is. You're right on that. I still feel it was easier to sell in the US at the end just because the sales cycle is way shorter. So we were able to sign deals in less than a month, which took us probably three to four months in Germany with, they have to have a meeting and they're the bigger stakeholders over there. It's a cleaner green light in us. We are winning. I just feel now because of these superior stuff, superior product, rather than just on cost cutting. So when we are saying, we are going to increase your revenue because we are also doing some sales for you, we are not just cutting costs for you.

Shraey Bhatia:

We have all this AI improving itself. You can scale it to, I don't know, instead of 1,000 calls, you can scale it to 100,000 calls. You can have local accents. So if someone is calling from Florida or someone is even in Australia, the uk, you can have different accents running to sounded more local as well. Which by the way, is also true in Europe. Like we are. It's a Swiss German guy and he's like, I don't want the German dialect, I want my Swiss dialect. So those things have started coming in as well.

Shraey Bhatia:

So the initial assumption that, Germany has this problem and it may not address us kind of is now false because people are more accepting in the US and in the UK as well.

Demetrios:

Oh, man, that's fascinating to think about the different sales cycles and the different speeds at which these two regions move in. And especially I've felt it living in Germany and recognizing that you do have a noticeable difference. It's not like it is subjective in one way, but it's almost like confirmable because we see with the whole European market, there's not as much investment going into this. There's also a little bit standoffish and that manifests itself in regulations and it manifests itself in folks going a little bit slower when they're buying new products that could take them in a different direction. So you getting to feel it firsthand while launching a company I think is fascinating. And then I know that we also talked before about how when you moved out of Germany, you had to pay to move out of Germany for the company, which I also find absolutely nuts that you have to like almost pay tax on the fact that you are getting out of there. They're like, no, you can't leave, we're going to keep you here forever.

Shraey Bhatia:

That's fair. I mean to be fair though, that the talent in Germany, it's still unmatched. Even in here, in this batch there are I think five or six companies from Germany. There's so many people from Germany I meet here, like who are building this big companies and stuff. And that's sometimes like I love my time living in Berlin and sometimes that makes me feel slightly sad. I'm like, you have like the best universities, best talent and why everybody ends up going to the US and not building it out of Germany. And it's not like Germany doesn't have money. I mean the economy is big, it just needs a bit more, I think risk taking to say, there will be some startups which might fail, but so be it. And it's not just comes down to the German government, maybe German VCs, German investment thinking, guys, let's go for it.

Demetrios:

I recently read something about how the best place to have your engineering team is in Europe. And the best place to have your go to market team is in the US So like the ideal scenario is that your engineering is based in Europe because you have this arbitrage. You also don't have these exuberant salaries like you get in the US you have incredible talent in the US but you got to pay top dollar for it. And in Europe, an engineer, an average salary for an engineer is between like 60 and 90k. And that is wild. When you look at in the US you can pay between like 120 and 250k for an average salary for a senior engineer. And so that same profile, depending on which continent you're at, you're going to be spending a very different amount per month. And then the go to market team being in the US will have that close contact to the customers and be able to also have their ear to the ground.

Shraey Bhatia:

Very fair point. And I see a lot of German startups doing that. We haven't decided on that at the moment, personally, because I also believe as a small team you want to be together. But you're absolutely right. You can get top talent in Germany. Even if you want to go absolutely top, you won't spend more than 100, right. 100k euros or 120 at best.

Shraey Bhatia:

And in the US, to really go top talent, you might even have to break the bank above 250. So I mean, every day I'm here, people actually get LinkedIn messages every other day and they don't even see my profile that I'm a founder and then they're this much salary. I'm like, are you making these numbers up? Because I'm like, if I want to hire on these salaries, I'll go bankrupt. That's a big thing.

Demetrios:

That 500k yc check does not cover your salary. It's not good enough there.

Shraey Bhatia:

Yeah.

Demetrios:

Well, this has been awesome, man. I really appreciate talking to you. Is there anything else that you wanted to mention before we go?

Shraey Bhatia:

No, all good. I think it's been a lovely chat with you and always happy to connect anyone in who's in, in Germany. Because I'm visiting again back in April. So if anybody wants to catch up, have a chat. More than happy. I do spend still quite a bit of time in, in Germany.

Shraey Bhatia:

And if you're looking for, as I always say, the best self improving voice AI. Leaping AI is always the answer.

Demetrios:

That is a very nice term. You should copyright that term. Self improving voice AI. It's the first time I've heard it and it instantly has a ring to it.

Shraey Bhatia:

It's the only one as well.

Demetrios:

There you go. Copyright it now before any somebody else takes it.