Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
How to Run a Profitable Services Firm in the AI Era | Sridhar Muppidi (ello.ai)
1 Plays3 seconds ago
Most service businesses are adding people to grow. One Hyderabad founder did the opposite and cut his team from 1,000 to 600 with zero drop in revenue, a live case study in AI and the future of work.
Sridhar Muppidi, co-founder of [x]cube LABS and ello.ai, breaks down exactly how, and what it means for anyone building in the AI era.
Few people have built through as many tech cycles as Sridhar Muppidi, who started in the dot-com boom, co-founded the cloud telecom firm PanTerra Networks, and shipped India's first game on the Apple App Store before scaling [x]cube LABS into a bootstrapped agency that has created over $5 billion in value for clients like Amazon, Sony, and Dr Lal PathLabs. His newest bet, ello.ai, builds enterprise voice agents that let customers talk to software in Hindi, Telugu, and a dozen Indian languages instead of clicking through menus.
In this conversation with host Akshay Datt, he argues that AI came for high-paid coders before blue-collar work, that today's cheap AI tokens are a subsidy with a reckoning coming, and that the engineer over 40 is now the most valuable hire. With Indian IT shedding entry-level roles and the AI and the future of work debate intensifying, his ground-level view is timely and contrarian.
👉How [x]cube LABS cut headcount from 1,000 to 600 people with no drop in revenue using AI coding agents
👉Why AI displaced high-paying programmer jobs before blue-collar work, reversing the 2012 consensus
👉Why Sridhar calls AI model companies drug dealers and predicts a token pricing reckoning 👉What separates writing code with AI from orchestrating coding agents, and how to brief them like a real team
👉Why almost every piece of software built in the last 40 years must be rebuilt for AI agents and voice
👉Why a profitable quick commerce idea his team built in 2015 slipped away, and the lesson he took from it
Subscribe to Founder Thesis for weekly founder conversations and follow Akshay Datt on LinkedIn https://www.linkedin.com/in/akshaydatt/ for daily insights.
#SridharMuppidi #xcubeLABS #elloai #FounderThesis #AkshayDatt #VoiceAI #AIandJobs #IndianStartups #AIcodingAgents #FutureOfWork #EnterpriseAI #ConversationalAI #AITokenEconomics #SaaSPricing #TechLayoffsIndia #AIagents #StartupPodcast #IndianITjobs #AIautomation #DeepTechIndia
Disclaimer: The views expressed are those of the speaker, not necessarily the channel
Recommended
Transcript
The Need for Critical Thinkers in AI
00:00:00
Speaker
Who's writing code today? We produce over 1.5 million software engineers every year. We need more liberal arts, critical thinkers, more than computer science engineers. How is software as an industry shaping or evolving in the world of AI? Out of 110 or unicorns, half of them are SaaS businesses. There is going to be huge price pressure. Almost every software is potentially going to be rebuilt.
00:00:23
Speaker
Sridhar Muppidi is the founder of Xcube Labs, which has built software worth over $5 billion dollars for giants like Amazon and Sony. He is now building LO.AI, a voice agent platform for enterprises. What are your regrets in this journey? We were always there at the right time, right place. Just didn't execute. We got distracted and tried to solve somebody else's problem. We could have been billionaires instead of millionaires.
Introduction to Sridhar Muppidi
00:00:54
Speaker
Sridhar, welcome to the Founder Thesis podcast. ah I will let you do your own introduction because you are doing so many things. So ah tell me a bit about, but like give me like a headline of who is Sridhar.
00:01:07
Speaker
ah ah Well, I'm trying to figure that out
The Evolution of Purple Talk and Xcube Labs
00:01:10
Speaker
myself. My name is Sridhar Mopadi. I'm a co-founder of a company called Purple Talk. I'm also the chairman of the the company. We are a group which was formed in 2008 when Apple was about to announce their SDK. And since then, we've been building products, apps. you know We are also a consulting business. So we have various businesses. ah you know Some failed, some succeeded. One of our biggest businesses is called Xcube Labs. Xcube Labs is a dev shop. It's a consulting business. But...
00:01:42
Speaker
ah I would say we are different from other consulting business because we we we take on more challenging problems than others. ah And ah because as as engineers, as designers, ah we want to be excited about things we do. So Xcubelabs is that, and that's one of our biggest businesses.
00:02:02
Speaker
We've done various other, we spun off various other businesses from Purple Talk. One of them is called Nukar Shops. If you are in Delhi, Hyderabad, other airports, most of the point of sales billing systems are ours, as well as, you know, over 12,000 other billing systems at on yeah in in india This Nukkad is this N-U-K-K-A-D that Nukkad is?
00:02:26
Speaker
Yes, Nukkad. Yes, I was a big fan of the TV show Way Back, you know, that reveals the age I am. I have no idea which which TV show this is.
00:02:37
Speaker
exactly Good one. So, ah yeah, so ah so that's, ah so Nukid Shops is one of our businesses. And then we also have video gaming, which is Yes, No, ah where we build ah mobile casual games, primarily puzzle genre, and we're quite good at it And traditionally, we've also licensed some big IPs like Star Trek, Adventure, and Kingsman. So we build some big games.
00:03:05
Speaker
Now we're building more focused puzzle genre games. So that's one of
AI's Impact on Software Development and SaaS
00:03:10
Speaker
them. We also have a marketing automation platform called Upshot, ah which provides analytics from the biggest airlines in India to, you know, around the world.
00:03:22
Speaker
ah So that's one of our bigger businesses. And... ah For past year or two, we've been heavily invested in AI, ah building tools, building products. One of the products we've launched ah is called Ello. It's a conversational voice platform, ah which allows people to build any kind of conversational voice agents.
00:03:44
Speaker
And on top of that, we just launched another platform called Mina, which is a meeting participants. And it's tied to almost all your workflows and tools. So because of those integrations, it can be smart, right? And ah I mean, great thing about AI is you can rent whatever intelligence you want, right? You can be the smartest person or the dumbest person, it's up to you.
00:04:07
Speaker
so ah So Meena provides that to our users. So that's a new product I'm working on. So within PurpleTalk Umbrella, I head up the products. ah My other co-founders handle other businesses.
00:04:20
Speaker
ah Give me an indication of like ARRs for these, like ARR for SQ, which is a dev shop. How big is that? And what's the help on there? I'm assuming that's the... So overall as a group, I would say overall as a group,
00:04:36
Speaker
can't go specifics, but
From Spinning Off Businesses to Investing in Startups
00:04:38
Speaker
as a group, we do probably about $43, $25 million. dollars The biggest would be Xcube. Biggest is Xcube, yes. But we other businesses are doing quite well for us.
00:04:48
Speaker
Yes, like around $60, $40 range. I come from a different generation of entrepreneurs. um I started my career way back in 99, professionally earlier than that. But 99, I had my first job. I started my first company in 2000.
00:05:03
Speaker
And what was that company? like Just give me a little bit of that history. Okay. It's called ah eMatrix. But the point I was trying to make was ah most people don't know the dot-com boom and the telecom boom, right, which has happened. The telecom boom is basically, you know, from AOL to everybody's stocks were like skyrocketing. And, you know, they've been making insane investments. Nobody understood why they were theyre laying those so many cables and you all of that.
00:05:32
Speaker
If you look back, while it didn't work out for them, it worked out for us. It it made everything easy. And that brought about the whole digital revolution, which eventually happened, right?
00:05:43
Speaker
So those things, ah the turnaround is 25 years, 50 years, what you what you laid there. The thing about GPU investment is four to five years.
00:05:54
Speaker
right It resets every four to five years. So when you are making that kind of investment and it's not like a one time thing, you make that investment and then you have power, your cooling charges and all of that. So it's a we're talking about a completely different thing when we talk about investment in this generation.
00:06:11
Speaker
And it's at the moment, it is ah unsustainable unless people start charging what it costs them. And But it's going to take time, right? At the moment, they are able to spend because they can. But it's such an amazing technology.
00:06:28
Speaker
You know, we're all, I mean, as somebody who's been a programmer and product manager and all that, it's I can't imagine a world without AI now. Does your token spend exceed a million dollars annually?
00:06:40
Speaker
Yes, we are going to get there. Wow. ah in In like in a couple of months, like within this ah year, you will cross a million dollars off. No, no, we will definitely. maybe ah We will definitely get there. we are probably So when you talk about token spend, then we are talking about two places where you are spending tokens, right?
00:06:59
Speaker
You are spending tokens to ah build something, you know, and you're spending tokens within the product you use for providing your chatbots, for providing conversational AI, ah providing problem solving tools, whatever, right? So as a combo, we are definitely exceeding a million already.
00:07:18
Speaker
Our spend to build products is not there yet, but it's going to get there because, you know, You know, we we need we need that. Tell me something, like, how is your ah framework on what product to spin out? So in a way,
Challenges in SaaS Pricing and Software Transformation
00:07:34
Speaker
Nookershop used spin out of Purple Talk. I'm assuming the person who runs it also has equity in that business, right? He's like a co-founder there. Of course. Yeah, yeah. yeah He's a see significant owner, yeah. so So is there a playbook on spinning out businesses, like what business you select? Is it largely dependent on the person that, okay, this guy is really great and I should ah support him and ah like he'll build a new division for us? Or is it based on opportunity or like how does that whole thing happen? Because you've spun out multiple businesses, you have that upshot. um
00:08:08
Speaker
I don't know if Yes, No was also a spin out or not. Yes, No was a spin off. Yes, No was also a spin off. Yes, No also spin off. So oh ideally, the point is, you know we live in a world, if you want to build something successful,
00:08:27
Speaker
90% of the time, at least in the you know startup world, you should be able you should raise money if you want it to succeed. Of course, there is always 10%. You can get away by not raising money, but you know the scale at which you can go, like excluding some outliers like Zerodas and others, I think ah that's difficult. right So now, a if you want to build a fundable company,
00:08:54
Speaker
ah it you know It's like ownership, the founder's commitment, all those elements come in. ah We were unfortunately not built that way. right Reality is we you know we we could keep some significant share, but there' still an investor would come in and basically look at life.
00:09:12
Speaker
ah this is all dead, ah you know, ah investment, right? It's like pre pre whatever they come in. So for a company like us, while we can give a blank check and keep investing in the product till they become profitable, and that's what we've done with all our businesses.
00:09:31
Speaker
ah we kind of weren't able to raise outside money, not as much as we should, right? Because you were a development shop, like you were essentially a services business. but not not Not necessarily the ownership and the how much you own, right? At the end of the day, an investor is going to come in and like, I'm not okay you owning more than say 10% of this company. And you're like, hey, I've already put in a million dollars or $2 million. dollars Why would I own only 10% that company? But that's not the case today.
00:09:59
Speaker
Today, it's the opposite. ah Investors will not invest in a business where the founder doesn't have enough skin in the game. Exactly. So that's what I mean. That's the problem, right? Which is basically we as a company, so there are, right, when you spin off, how much does the the person who's actually running versus how much the company owns is the tricky part, right? right So we were structurally designed. Right.
00:10:21
Speaker
Like Nukar, the co-founders there would have like a... Yeah, wouldn't have as much as the industry would expect them to. And our position would be, have already put in $3 million dollars on it. And, you know, how could I take less, right? So so it becomes tricky.
00:10:37
Speaker
ah And that's the reason we kind of slow down on that business model, horizontal growth business model we're investing. So you feel that the your funding is not enough? Yeah.
00:10:51
Speaker
The point is it is enough to a certain level. It's just that then then what happens is ah it's only your conviction, right? At the end of the day, you want to get other people on board and you know ah and use their network to help it grow and things like that.
00:11:10
Speaker
uh so and then there's so many other things you want to do right it's like because at the end of the day a company for a company to succeed we got to put in four to five million dollars and that's like a lot of money uh why you know so so that those are the things to look at so lately at least for a few years now we changed our model we rather invest in companies versus uh ah you know, ah spin off our own company. So we've just been actively, when we mean we see great entrepreneurs, we just cut the check, right? that that's That's our thing now, which is, that's easier for us versus we, you know, and then also on the consulting side also, we've done that.
00:11:53
Speaker
A lot of startups, which are in town to become billion dollar companies, we picked up like one, two percent in those. So that that also happened. Essentially, what you're saying on the spin-off model is that ah it will need a lot more capital to have a meaningful outcome. It can be ah like a lifestyle business or whatever, like better than a lifestyle business. But if you really want a meaningful outcome, then you need ah VC money in that business. And a VC will not...
00:12:23
Speaker
typically invest in this spin-off structure because the guy running the business has like a maybe will teens or something like that kind of a stake and you have the lion's stake. And obviously when there's such a large strategic investor in the cap table, then other people will not come in.
00:12:40
Speaker
ah So it is better for you to just take
Nookershop's Strategic Pivot
00:12:42
Speaker
small stakes as like angel investments or whatever, something like that. Exactly. if you So that's the thing, right? So when you do that, then you know how much are you committed, right? It's like you giving, say, $50,000 or $100,000 check versus you giving $3 million, dollars right? It's easier for us to do $50,000, $100,000, which is what we prefer right now.
00:13:03
Speaker
We do smaller smaller checks. We pick up very small percentage too. So that way they have freedom to grow. And our policy is simple, which is to find great entrepreneurs and ah if you like what they're doing or within our company too, you know. So anybody who but wants to work on something great.
00:13:20
Speaker
And that's been our kind of USB where like, you know, people just, when they, they would rather just ah take some money from us and build something. ah And if it doesn't work, they would rather just come back to us again. Right. So that's been, that's kind of been working for us.
00:13:34
Speaker
When you say we invest like the Purple Talk as a company is investing money or you are coming in as an angel investor? Yeah, Purple Talk as a company but gives a small check usually.
00:13:45
Speaker
What are like some flagship clients of XQ or what kind of products are you building? Are you, for example, building products for consumer internet companies or SaaS companies or is it like banking tech? Banking tech typically tends to be the biggest spender for software health.
00:14:02
Speaker
Yeah. So, I mean, we have diverse set of clients, right? From healthcare to gaming to like one of our biggest clients is Panini, which is ah a collectible card company, largest collectible card company in the world. We handle most of the digital.
00:14:18
Speaker
We, you know, clients like Lalpat Labs, right? They're the biggest diagnostic center in India. ah And so we have... Indigo who uses one of our you know technologies.
00:14:31
Speaker
So their yeah ERP has been built by you basically, like for Lal Path Labs. Yeah, a bunch of things. We are working with them. and you know so So there's different people, different clients where we solve different problems.
00:14:45
Speaker
ah we've We've been doing this for so long, like this is our 18th year being in the digital transformation to now being AI native. that weve we've solved We've created over over five to six billion dollars worth of value for people. right We work with entrepreneurs who are still studying when we started building their product and they ended up being billion dollar companies. right ah you know So we've done various those kinds of ah products and companies.
00:15:19
Speaker
Internally within XQ, I kind of want to understand what has the adoption of AI given to XQ? ah Does your code now get written by a coding agent with human quality checks? Is that what happens? or like like you know what What has AI given to XQ as a business? ah So about ah not 100 percent, because there are clients who would rather, you know, there would be policies, AI policies, right? Because at the end of the day, they don't want their code to go up on the cloud to open AI or cloud, anybody to look at their code, right? to To be honest. So ah most big companies would rather not
00:16:02
Speaker
ah would rather we not use AI. And even if we use AI, it's probably, yeah. so that That sounds like, it sounds, i mean, I'm saying old-fashioned, even though this would have been absolutely correct way to think about it just nine months ago. But today it sounds old-fashioned. I mean, who who's writing code today, right? like So no so the that's the thing. So there is ah there are two aspects when you're talking about code, right? There is Writing code using AI is different from orchestrating using coding agents.
00:16:34
Speaker
right writing code is like hey okay so the difference is i'm writing a product i need to write these set of functions okay then i basically go to my uh llm or whatever tool i'm using hey write these functions this is how i would end up using it so as an engineer will basically tell my agent to do those three c four functions and then i copy them and put them in into my system and then test it out Right.
00:17:01
Speaker
That's the old way of doing it, as you said. Right. That's how people were doing it four months back or six months back. New way of doing it is ah being able to set a vision, brainstorm what you want to do and set the framework you want your coding agent to follow, which is okay these are the things I need you to come to me for approvals. These are the things I need you to make the decisions. And this is what I want to do.
00:17:28
Speaker
Stage one, these are the things I want you to deliver. So you are basically more or less talking to another high-end engineer or multiple engineers who you think ah would have to deliver something, right? So there you are basically artistician. Also, you are handing it over to, you know, these these coding agents, right? that So that part, companies, big at least big enterprises, don't yet want you to it.
00:17:54
Speaker
they They're okay if there is a, you know, like something like Gemini and others who basically say, you know what, we're not going to use your code. It's ah it's under a... ah So those are assurances they're looking for.
00:18:07
Speaker
And we are getting those assurances from some of these bigger players. And if and when that happens, you know, we can we can orchestrate everything or hand it over. And the second aspect...
00:18:19
Speaker
or the opportunity over there is like a legacy course, right? You got this monster drawer, which you've written about 10 years back, 15 years back. ah Then you can't just hand it over to a coding agent and who knows what it would do to it, right? Versus there you will still do piecemeal things. Versus a brand new project, then you can...
00:18:41
Speaker
hand it over to a coding agent. So there is like when you are working on this large scale projects, you you have these options on, you know, ah what you want to do. So do you want to be the orchestrator or do you just want the coding agent to be the orchestrator while you become like a more, a you know, ah somebody with the judgment?
00:18:59
Speaker
Fascinating. This team of agents approach would essentially make one person equal to a team of five, something like that?
00:19:11
Speaker
It depends upon the project to project. ah It could be team of 100. It could be you know team of 100 over two years. you know So it it it depends upon ah how you orchestrate it These models are getting incredibly powerful.
00:19:29
Speaker
Has the pricing evolved in the world of AI? Like traditionally, software pricing was time and material kind of pricing that this project has this many man hours and maybe whatever additional on top of that, something you will charge. Is it still the same or has the pricing model itself also evolved in the world of AI?
00:19:47
Speaker
Oh, that's a good question. it's ah as As I said, it's an evolving field right now. We're all trying to understand how that works because at the end of the day, you do need a ah people who are using AI or sort of like almost leaders who are not cheap, right? so So, you know, you saying that I'm going to charge you the same amount I used to charge when you are paying maybe double or triple to your people, right? so So there's that.
00:20:16
Speaker
But then the number of people would have gone down, right? Yeah, exactly. So the the point is ah the traditional time and material won't work anymore because your expense has gone down, ah gone up.
00:20:27
Speaker
That was a labor arbitrage. Like the more and man hours you can... stuff into a project the more you will earn but that's no longer the case that's no longer the case we you know that's no longer the case because now you still have to bill based on that but you can't be billing your 50 an hour anymore you have to 150 200 dollars an hour because at the end of the day you are pay paying your people that kind of money So the ah the the pricing benchmarks have shifted up. It has to, you know, but but that's hard to explain, right? At the end of the day, how are you going to basically say, hey, I didn't have very smart people working on it. Now I have smart people working on it.
00:21:07
Speaker
So yeah, that's it that's a that's a tricky ah problem to solve. And typically the token cost is a pass-through cost or like I'm curious about that.
00:21:18
Speaker
At the moment it's not. ah But as I said, so there are two sets of expenses. There is the expense within the project. Then you look at it like your cloud expenses. Why would you pay? The client will pay, right? The same way.
00:21:31
Speaker
the token expense within the product is going to be theirs. But the tokens you use to build the product, ah at the moment, it's not passed through. But if the prices keep going up the way they are, ah then we'll see.
00:21:43
Speaker
So you've been building software ah for companies, both B2B, B2C, internal-facing software, like, say, what you told me for LALPATH Labs, as well as software which is directly consumer-facing. um How is software as an industry shaping or evolving in the world of AI? Multiple things.
00:22:07
Speaker
ah the This is an argument which me and my colleagues, we we keep having have having, right? Is the easter AI good enough where can we build production quality products and launch within weeks, right?
00:22:24
Speaker
And I think consensus is no, it's not, you know? It needs to still need the back ends. It still needs to have the framework. It's getting close, but it's not at the moment at that level.
00:22:41
Speaker
what almost everybody who is arguing and is no, it's not also argues that in six months, in one year, it will be right. So, so basically, even though we are not, we're probably going there or getting there. So with that in mind,
00:22:58
Speaker
ah Do we, India employs what? ah sixty million Six million people in IT-enabled services, you know. ah We produce over 1.5 million software engineers every year, right? 1.5 million people every year our learning to be programmers or, you know, in some form or other.
00:23:21
Speaker
Every engineer in college is you know has like two, three computer-related programs. Some of them exclusively only do IT and computer science programs, right? Do we need that many computer science engineers?
00:23:37
Speaker
ah Maybe not. right that's the That's the world we are entering in. So ah that's a problem. We need more liberal arts, critical thinkers, more than computer science engineers. right Because we need people who solve problems, people who can think pink, not just some tunnel vision.
00:23:55
Speaker
ah you know Hey, give me code and I'm going to look through it. So that's the that's the thing where we are heading towards. ah So that's a problem overall for ah just Indian industry as such. I don't really have a solution. i am ah I'm excited about technology, but I also think we need to be realistic. There is something coming up for India and people need to prepare themselves.
00:24:22
Speaker
ah The second aspect is... massive ah SaaS businesses out there, right? Which is like, they're like, literally, India has, you know, out of 110 are unicorns, half of them are SaaS businesses. Globally, there are hundreds of unicorns doing large scale software as services.
00:24:44
Speaker
Now, The argument has been, ah can ah can a a large enterprise ah build a a CRM or a Salesforce management or SAP kind of thing using Claude or orchestrate it? And why would you pay millions and billions of dollars to some SaaS company when you can probably get away building it yourself?
00:25:08
Speaker
And you own it. Right. That's that's the argument which is going on. And the only right now counter to that is that's not your expertise. Why would you want to build, you know, just because you want to take some A to B something, you are not going to build a road for it if the road already exists. Right. You just pay the toll for it. Right. Just because you don't want to pay the toll. ah So that's that's the argument, right? Why would you build it if it's not your core business? If it's not going to add value, we're just building it to bring the price down.
00:25:39
Speaker
ah But that said, can you get away from charging as much as you're charging? So what's going to happen in SaaS business, I think, is there is going to be huge price pressure, right?
00:25:52
Speaker
ah Somebody who was charging you $300 a month subscription per seat on something, right? would be asked the question that, hey, what is your cloud cost for all my 100 people? Right. I'm paying you $30,000 right now. What's actually your cloud cost? Oh, that's $3,000. Right. So if it is $3,000, why am I paying you $30,000 a month, almost 10x more? I will probably give you 50% more. So I'll give you $4,500 period. Right. That's the argument. That's the conversations ah companies start happening, are having with SaaS businesses.
00:26:29
Speaker
So that unless the SaaS business actually is not only software, but it's data, right? So if you are a sales force, but you also own the database of clients and the phone numbers and all of that, you own meteorologists, you know, you own all of that, then then it changes. So this is the doom part of it, right? You know, there is going to be job loss. There is going to be, you know, put price pressure on people who build products.
00:26:58
Speaker
ah But ah the opportunity would be, ah you know, almost every software is potentially going to be rebuilt.
00:27:10
Speaker
Essentially, you're saying software will become system of record. Like that's where your data will reside? Yeah, where the data resides, where the business logic resides. But yeah um right now, the software were built for humans, built for it as an API. They need to be changed. They need to be changed to be agent first, right? So in the olden days, or not olden days, but we we have this thing, this whole subject. Yeah, it seems like olden days. No, I get that.
00:27:38
Speaker
yeah we we we we We use this word, right? User experience. We've we've been talking about UX, UX design, UX flow. Now we all all we talk about is agent experience, right? what How would agent consume this content? How would the agent ah interact with this? And what is the...
00:27:57
Speaker
ah how What do I do to make it better? right so So that's where ah you know the world is heading. And that presents, if you are a builder, that presents an opportunity to rebuild everything. right Everything which was built in last 40 years needs to be rebuilt to be conversation first and to be agent first.
00:28:16
Speaker
right And that is the opportunity. So if you're a builder, those are the opportunities for you. So it's not all... you know, doom, there is all this great opportunities. You just need to be ahead. You just need to start thinking, how would agents consume? How would use humans consume? How do I make their lives easy?
00:28:34
Speaker
Right? Once you start thinking, there's like the million opportunities in front of us. Your POS solution, that Nookker shop solution, ah so ah is that also at risk? That's a very easy thing to replicate now, right? Yeah, there's an interesting story around Nookker shops, right? It's... um oh you know And there's an insight I would love to share, which is when we were building Nookut Shops, it was an internal hackathon. right So a bunch of engineers, we you know we we ran a hackathon, say, hey, pitch great ideas. you know And somebody pitched this app, we're saying, hey, let's give this mobile app to, this was 2015, I think.
00:29:18
Speaker
this so build this mobile app where we give shopkeepers this mobile app and then they can you know ah they will have their inventory and they can serve their local market. right So ah Your Swiggy model, your Zepto model, it's just that people can come and pick which store the closest to them and pick that and then pick the items and the guy will deliver, right? It was a super simple like idea.
00:29:45
Speaker
And somebody internally pitched it and we all liked it. And we you know we we were looking for something to invest in. So we decided, okay, this is the idea. We'll go ahead and invest.
00:29:57
Speaker
And there were two problems we faced immediately when once we launched it. One was there was a lack of inventory in information, right? So when it somebody places an order, ah the there is almost 80% chance that the item they wanted wasn't there in the store, right? Because the inventory information wasn't apparent and the shopkeeper wasn't keeping it up to date.
00:30:24
Speaker
So there was that problem, right? The second problem was quality of service. Now, quality of service problem was People, the person who, we cant be there was no way for us to assure that they will deliver in 30 minutes because it was their responsibility.
00:30:39
Speaker
So they didn't deliver. ah They will deliver at their convenience, one. ah The second one is the person who delivers Who is he?
00:30:50
Speaker
You know, who is this person who's coming to my house? Right. What kind of, you know, ah background check happened on him? And, you know, so those two problems were like immediately hit us. Right. The moment we launched it, we, you know, we we got a lot of shopkeepers excited. we you know, we did some aggressive promotion and people were placing orders. But these problems were very apparent. And this was a margin business, like you were taking a margin from the order?
00:31:16
Speaker
No, yeah ah we yeah, we were taking a small margin, but basically, we yeah we were thinking subscription versus margin versus, the business model was still evolving, right? At that point, ah the the CEO of Lookit Shop, so the person running the Viksha Club,
00:31:35
Speaker
ah He came in and pitched to us, say, you know what, Strider? This is not going to work. you know The data shows that data shows that you know the quality of service is a major problem and inventory a major problem, right? So what do we do?
00:31:52
Speaker
right And he came up with this thing saying, let's do a dark store. Let's set up like our own Kirana store. and know That way we know exactly what they're going to have.
00:32:03
Speaker
And then deliver because our pitch was like, we will you will get your things in 15 minutes. You you've went through five years of quick commerce evolution. Like five years of that evolution in five weeks. Yeah, right. So ah this this was 2015. Yeah, this was 2015. And yeah this was two thousand and fifteen and He pitched it.
00:32:22
Speaker
And my response to him at that time was, you know, that's a good idea. um mean, it's for us, you know, to set up a shop is like four, five lakhs, 10 lakhs, you know, instead of going through another shopkeeper.
00:32:35
Speaker
ah But, you know, we won't think I was thinking, OK, first of all, scale. You know, if I need 100 shops, how would I do it? I wasn't thinking raising funds, one. the more More than that, I was thinking, you know, Big Basket is there. ah You know, they're all coming after these poor mom and pop Kirana stores.
00:32:54
Speaker
Our job is to empower them and help them, you know. I wasn't thinking I'm building a unicorn or a startup or whatever. I was thinking I'm going help these people really, right, more than anything. And I was like, you know, what what we need to do is let's build them an amazing point of sale system where we do inventory management so that the information actually feeds them. Let's give them on the back end supply chain, you know, so that they they can place the order and the supplier will deliver. so When the the fork was supposed to be taken, my experience or whatever basically said, you know, oh we should go this side, while the actual person who was talking to customers, customer facing, because I was shop facing, he wanted to go the direction which would have been a billion dollar path, right? So, ah and, you know, so the the thing is,
00:33:49
Speaker
That was something which hit us later on where we, you know, obviously it grew. I mean, it became a decent business helping these people and all that. When you say you wanted to help them, it became like a ah like a shop that they could spin up, ah like a Shopify.
00:34:09
Speaker
easy use, stripped down, stroppy pie. Yeah, right. We figured they can do online delivery. They can do, ah you know, we will help them get the best products from their suppliers, the cheapest margins, you know. So we we went on that direction.
00:34:24
Speaker
ah So and we tied up with the people who can give them credit lines, people who can, you know, ah supply them at a discounted price, you know, things like that. Right. So we've we've done various things on that direction versus The actual problem is the customer themselves, right? or We have like, hey, we have all this data too. But the problem is like,
00:34:46
Speaker
You know, I've started making tea and I don't have milk, right? You know, can I just open this app and would somebody deliver milk, right? That was the problem we we but we wanted to solve. Instead, we got distracted and tried to solve somebody else's problem.
00:35:01
Speaker
What's really the the takeaway in this? That you should listen to the guy who's... Or what? Takeaway is chase money, right? you you won't regret You won't regret chasing money. You don't regret, you know, so ah don't confuse...
00:35:16
Speaker
ah entrepreneurship with, you know, social entrepreneurship and those aspects. You know, if you are there to build business, think money, right? You have to think, ah how do you make money?
00:35:29
Speaker
And how do you solve a problem, you know, so that it could make money, right? Those are the aspects. we were thinking we got distracted we were thinking mom and pop let's help them out you know let's make sure they survive a little bit longer right is that that's the wrong reason to get it uh the reason is you're solving somebody's pain problem which is they need something immediately they need somebody they can trust delivering it they need whatever they need can be delivered. right Those are the problems we were supposed to solve. And that's a big takeaway, which is, you know we got distracted. We didn't understand why we were doing what we were doing.
00:36:09
Speaker
And yeah, the this co-founder doesn't let me forget it. And ah how did PurpleTalk start? like like You started it alone, you had co-founders. Just tell me a bit about the origin story, like your plus PurpleTalk origin story.
00:36:23
Speaker
ah Oh, that's a good question. So I'll go a little bit further back. Right. ah um As I was talking earlier, I've started my career during the dotcom boom.
00:36:38
Speaker
ah i I did my BSc computers. i I was an average student, was decent in math. ah but overall average.
00:36:49
Speaker
And then I discovered computers, right? Then I discovered programming. and And then I realized i was exceptional when it comes to code, right? ah ah Because, and the realization came in because it's the,
00:37:05
Speaker
i I was surrounded by some very smart people and, you know, top person outlaws and all that. And they used to struggle. And for me, everything was apparent and easy, right? ah So that kind of just, hey, what happened here, right? The people I used to look up to are struggling with this while it seems like common sense to me.
00:37:30
Speaker
ah So when I discovered that and... you know And I doubled it all. It's like I went all in and pick became pretty good. right I used to contribute to the Apache project, which is like what almost all the be websites run on and you know things like that. So I was an early contributor there in the open source community.
00:37:53
Speaker
I started Hyderabad. I'm from Hyderabad, India. I started Hyderabad's first Java user group way back in 97. And I used to help a a bunch of companies. And I was still studying, doing my basic computers. So ah that group, people used to come from around, ah ah you know, in Hyderabad, all the big companies used to reach out.
00:38:18
Speaker
and ask questions. And i e I knew the answers. I used to answer. I used to do detailed responses for everyone. So while I was studying, one of these companies reached out and said, hey, would you be able to come in as a consultant? And I was like 19 years old or something like that. and And they're like, sure, why not? you know so And then then I came in and solved some problems for them. And then they were like, would you want to work full time? And I said, yes. So I started working while I was back in the study.
00:38:51
Speaker
of It also got to my head a little bit. I thought I was whatever. ah But you know, I started that way and i think in 99 or 2000 I started while I was working. I left that company because I wanted to work with some of my college other friends, people I was with and studying and they were still studying. They got out of college, I got out of college and we're like, hey, let's do something together.
00:39:16
Speaker
and they will they were freshers so they could they couldn't get a job. So I couldn't get them a job in my company or the one I was working for. So I said oh I'm gonna quit and start a company and work with my friends.
00:39:28
Speaker
So I had an idea of building a ah live support system for websites. This was 2000, right? there was There was nobody doing it at that time. Now obviously Freshworks to everybody else does that online support but 2000 nobody was doing it. so I built this product along with his friends over the weekend.
00:39:48
Speaker
very beautiful looking interface. ah And ah we put it out there. And there was a Israeli-American company, which was also, it's it's called Web Telecom, I think. And it was also building something similar. They had like, they raised like 10 million VC money.
00:40:06
Speaker
And they didn't have a product after after like six months or one year of them them having like a 20 member sales team and everything, no product to sell. And they looked at our products and said, okay, this is a great product. So they reached out to us saying, would you want to sell this to us?
00:40:23
Speaker
And we were like, okay, sure, why not? And, you know, because we we we were like a bunch of kids who were right out of college. We had no idea what we were doing. ah you know i mean So we said yes and they said how much? We said two million and you know surprisingly they said yes. And I was like that that was like a shocker. It's like whoa, what is this? What happened here? So anyway, so ah they said yes and they gave us a couple of hundred thousand dollars, I think two hundred thousand dollars and we gave them the software.
00:40:58
Speaker
What we didn't know, what they didn't know was what we built was a prototype, not a product. ah So it worked with two people. It didn't work with 100 people.
00:41:09
Speaker
Right. And of of course, you know, we didn't have the testing infrastructure. This is something we built over a weekend so that we could test and showcase. and And then these guys decided what we gave them was a production software.
00:41:24
Speaker
Nobody bothered to test and deployed it with about 200, 300 customers, right? And it stopped working. And that kind of was like scary because, yeah, i mean we we were buying all these expensive cars and everything, but you know nobody was looking at the product.
00:41:40
Speaker
ah And of course, I mean, point is at the end of the day, the moment those problems came in, this was over Christmas, I think, of 2020. Oh, sorry, 2000.
00:41:54
Speaker
We were working all night for like three days continuously and fixed everything. It wasn't, at the end of the day, it all wasn't rocket science. It's just that we didn't know there were problems. If we knew, we would have fixed it, right?
00:42:05
Speaker
And once they came to us, we will you know we fixed everything. took us like three, four days. By then, the problem was these guys we didn't know were out in the market trying to raise money. They had investors, checks coming in, you know, all of that. They were like out of money.
00:42:23
Speaker
We didn't know any of that, right? So ah by then, the they they figured this is not going to work. Even though it started working, they just went out of cash. And then the dot-com bust was about to happen.
00:42:36
Speaker
So that company file filed for bankruptcy. So they didn't pay us all the money. They just paid us $200,000. But that was our big lesson from, you know, what I mean by software, right? or what What right now your cloud code gives you as a prototype. That's what people need to know.
00:42:51
Speaker
It's not a production quality thing, right? So it it doesn't mean it won't get there. It's just that you need to know the difference. You need to orchestrate it. You need to test it.
00:43:02
Speaker
And we didn't know. And ah that was a huge lesson, right? It's like, you know, we built something, we got paid and we didn't deliver, right? But that gave us, you know, ah my first company before I started this, my first company, i had I started working on a gaming framework called PopX. And then I i did... um ah I did some voice stuff. or We were working on the SIP protocol, which was still in draft stage at that stage, at that point.
00:43:34
Speaker
ah So we were working on that. So Vivo AP was new. ah So because I had now experience building a kind of a startup where, ah you know, we build this life support,
00:43:46
Speaker
ah I had this idea idea that can I do the same thing with phone? ah Can I move support to phone where you know we can do phone calls and all that because we had that experience. So we started this company called E-Matrix and then it it was called Pandora Pantera Networks. where we did phone-based support system, call center solutions, then collaboration tools and video conferencing, you know. So Pantera Networks... This would be like competing with an Avaya kind of a... It did. It did. and But we were like one of the first ones to do it. Pantera Networks is one of the largest companies out there in the Valley now, ah which does these products. Your co-founder was in the US or like how did that happen?
00:44:34
Speaker
Yes. So the the person who was... heading this other company which went bankrupt. So I pitched it to him, saying, hey, you know, I know i know it involved caught between us on this, but I have this telecom idea, ah you know, and you already have experience selling it to these customers.
00:44:55
Speaker
How about we collaborate? You can be the CEO, I will be the CTO. So that was a pitch. And and then, yeah, we we I mean, of course, you know, he had some great ideas because he came from that industry.
00:45:06
Speaker
So we collaborated and the we built this company, Pantera Networks. And yeah, that became a pretty big company. D.E. Shaw invested in us. ah And yeah, it's it's, as I said, one of the largest companies in that space.
00:45:20
Speaker
ah We've never exited. It's been running for about 25 years now, but ah but and it's still still quite active. so Do you still have stake in Oh, yeah.
00:45:33
Speaker
I'm probably one of the larger shareholders there. then Why did you walk away from that? Why aren't you running it today? 2007, I got a call at 3 o'clock in the morning saying that a fire department was using us at that time and the phone calls were dropping.
00:45:53
Speaker
so interesting So, yes, I fixed the problem, but I didn't want to be in a space where the, you know, we, you know, fire departments were using us our phone service and I had to deal with something which had nothing to do with me, my product.
00:46:08
Speaker
But, you know, you have to figure out because the service provider we were using had issues. Right. so So it just became a place I wasn't having fun anymore.
00:46:19
Speaker
I wasn't, ah you know while it was amazing piece of technology and all that, we weren't enjoying. ah So that's when, you know, I quit.
00:46:30
Speaker
And then I wanted to do something simpler. ah And because we had a lot of experience with the video video technology and all that. So I started this nonprofit group called Education for Free.
00:46:43
Speaker
which basically, there's a but I'm sure there's a website called educationforfree.org, which basically provides so sort of Indian state-run public schools, ah sort of like a substitute teacher using video conferencing. This was in 2007, right? Almost back.
00:47:00
Speaker
twenty hours years back ah Bitrate was a problem, broadband weren't common. ah So we had this amazing compression technology which nobody had because of our relationship with people. And so we used those codecs to build it, which eventually with the codecs with Skype and ah WhatsApp and others ended up using for video conference. So we were the first ones who used it in a production grade software.
00:47:27
Speaker
And we launched that product. We gave it away to a bunch of schools. We had like thousands of, ah you know, we got some PR, thousands of teachers registered because the idea was like, can we inspire kids in rural India, right? Where somebody like you and me can be in a city and basically saying, you know what, kid, I used to sit in that same corner as you, right? And I'm a doctor now, I'm whatever.
00:47:49
Speaker
And you could be that, just work hard, right? You know, so it's... The whole idea was, can we create role models for people and while also provide, because at the end of the day, education is all about exposure, right? the you know While you can pick up a lot of things from books, but the exposure, knowing the right people, that's going to help. So we we figured that's that's the direction we wanted to go. So we had a lot of tie-ups.
00:48:15
Speaker
And that was doing quite well. and But then while that was happening, me and a couple of other people who left with me when we were doing Pantera, we were building this.
00:48:26
Speaker
Apple announced they're going to open up their Apple SDK so that developers can build apps on it. ah By then, me and my other co-founders were already playing with the jailbroken phone. We were already building games and apps and having fun with it. So we were all crazy about iPhone at that time.
00:48:44
Speaker
And we're like, okay, if this is happening, then we are going to go all in. So we we started Purple Talk. And were the first company in India to release a game in the Apple App Store.
00:48:59
Speaker
Okay. So the gaming business was Purple Dog's original business. Yes. that That's how we started it. and then But the reality was there was no real way to make money at that time from games.
00:49:10
Speaker
So we started building games for other people. ah While we were building games, people reached out saying, hey, can you do this app for me and all that. So that's how that business grew. And then we doubled down on it because we we felt like, OK, you know, if we're going to do this professionally, then we need expertise. We need to learn. So we bought in the talent we needed. We learned ourselves so that we could go after the space properly.
00:49:33
Speaker
Hmm. Okay, interesting. So why have ah like Xcube and Yesnome is also into game development as an outsourced company, right? Not really. Yesnome is not into outsourcing. It's Xcube primarily.
00:49:49
Speaker
ah Okay. So Yesnome is when you are launching the games yourself and outsource work is all with Xcube. So that's how the split is. Yeah. so what what happened was... Because I started my career way back in 99, 98, building games, so I always wanted to get back into games. So when we started in Purple Talk, we wanted to build games.
00:50:09
Speaker
But then because we weren't making money, we became sort of a consulting firm while we were doing great there. ah I always had that hunger that I want to build games, right? And because that's the reason I quit Pantera Networks. so ah So I started YesGnome or spun off YesGnome to just build, focus on building games. So we started licensing IPs like Star Trek, Adventure Time, Kingsman, Medigus, Cur, you know, stuff like that. So we like those games, build games on top of them.
00:50:40
Speaker
And this is after we build games for other people through Xcube. where we build the expertise of building large games. So we had experience. It's not like we were just... And then we started building games and we're doing all right there.
00:50:55
Speaker
Okay. ah What is the way you monetize your games? Like through ads or in-app purchases? Yeah. So predominantly ads right now. ah But the and maybe about 20, 30 percent, 20 percent comes from in-app purchases. But majority food would be ads.
00:51:11
Speaker
And is your heart still in that business? Because to me, it seems like your heart is in LO. That's a good question. ah Gaming is incredibly fun, right? Because so ah designing a system, solving those problems, it's it's immersive, you're you're excited. It's just that I i feel like I've done that long enough.
00:51:35
Speaker
right I've been doing this ah for almost 15 years now. ah So i i I thought, you know, there is something amazing happening around us and I would be stupid if I don't immerse myself in that world.
00:51:54
Speaker
ah So that's when, you know, about a year and a half back, we decided to take a step back from active game development, which my team is still doing. We we still have a pretty large team, almost like 70 members where we are doing it. But I'm not like active decision maker there anymore because I feel like ah ah there are better, smarter people doing that within the organization. Maybe you also see better outcome for LO than for the game studio?
00:52:24
Speaker
Not really. I think ah gaming could be a printing machine or money printing machine if you do it right. Right. So ah that's not it, though. ah I'm at a point in my career where...
00:52:38
Speaker
ah technologies excite me, right? ah new Newer problems excite me. And right now I'm just seeing opportunities everywhere, right? Which wasn't the case ah even five years back, right? So ah if you are a builder, this is like an incredible time to be because there's so many problems. Because think of it this way.
00:53:02
Speaker
oh we None of us knew we needed groceries in 15 minutes. right None of us knew. ah But the moment it happened, now we don't know how to live without it.
00:53:14
Speaker
right Because ah the kind of things you can do is only happens because of that time. right because ah you know ah and like There was this whole conversation about or top-up versus you know ah your whole monthly bags and all that.
00:53:32
Speaker
Doesn't matter. Everybody just buys everything from Zeptoes and Swiggy's and Blinkets and others. Right. Same thing with the AI. ah None of us knew we needed this kind of building capability.
00:53:45
Speaker
Right. they This speed. But now that we do, how we solve problems is completely changing because I can think of a problem. I don't need to wait.
00:53:56
Speaker
two days for somebody to solve and show me, by then my head my brain has moved on. Now I can be captive because I can be there for 48 hours continuously and I can feed through. This is like something which was never possible.
00:54:11
Speaker
And because of that, we will been we will build higher quality products because people are like only immersing themselves in one thing at a time. Of course, it's also, you know, stressful. It's also you're always anxious and feeling like the world has mowed on without you. But yeah, it's also exciting.
00:54:33
Speaker
So, voice as a field is pretty crowded. There are a lot of companies in India doing voice AI, especially the multilingual opportunity ah is something which Indian startups feel they have an opportunity in.
00:54:50
Speaker
um What's really the gap in that market? So, it's basically intelligence isn't there, right? There are multiple things about voice, right?
00:55:02
Speaker
ah there are, you for voice to work properly, you're using three different models. There is the text-to-speech or speech-to-text, and there is your LLM um where you orchestrate what you need to respond and all that, or tool access, and then there's text-to-speech, right? And when you're working with this,
00:55:22
Speaker
ah almost a lot of things which you need to get from your speech to text which is basically emotion uh you need to get the uh like i'm taking time to think through the gaps uh you need to get uh volume if somebody's yelling at you what's this it's like because once model just turns what you are saying into text. It's just text. I might have yelled at you, but ah as far as the, yeah, yeah there's there's so much which yeah voice actually can communicate, text doesn't, right?
00:56:03
Speaker
ah you know So a lot of nuance in what has been said is getting lost in that speech-to-text model level, right? That's a big problem.
00:56:15
Speaker
Then there is the LLM orchestration level where Intelligence, how much intelligence do you need, right? If it is collection where you basically are calling somebody saying, when are you gonna pay, you don't need intelligence.
00:56:30
Speaker
But if you are, we have one of our customers who's using it for exit interviews. Now, exit interview goes on for like a good 30 minutes where you want to probe them.
00:56:41
Speaker
Think, why are you leaving? What exactly happened? What would you do? You know, things like that, right? And there you need like intelligence, right? You can't use like your, ah you know, Gemini 3.5 flash or something like that. You need like, you know, claw something, right? So it's... a knowing what model to use for what context. And also within that orchestration, intelligence intelligence also changes based on the type of question.
00:57:07
Speaker
Because it ah if a question is around or something simple, something simple decision where you you need to say yes or no versus something where you need to give a thoughtful answer, so you need to change which model you go to, right?
00:57:21
Speaker
Then, as you said, nuance of language, right? If you're speaking to somebody in Telugu and somebody calls you and they will call me and say, Sridhar Garu Bhagon Narap. If it is Hindi, ah it is like Sridhar Ji, Kese Ho Aap, right? So ah the Ji is something which is expected, but ah the translation doesn't know that, right?
00:57:48
Speaker
The same thing again when you go to text to speech. There are elements how you deliver, how much emotion, how much pause you need to give. So while people say it's a solved problem, no, it's not.
00:58:01
Speaker
There are so much which need we need to get tried for people to trust this technology. right So that's where our investment is going in. where we are we We are trying to make these technologies ah not necessarily sound human because at the end of the day, that's not where we want these to go to ah because we want people to know they're talking to a bot.
00:58:27
Speaker
ah because why why deceive people? you know let the Give them the choice, but deliver the value, right? If you're talking to a human versus you were talking a bot, if the bot can deliver the same value, why won't you talk to it?
00:58:40
Speaker
There's no reason for you to deceive. So right now, I think the investment seems to be going into oh, I'm going to make it sound so realistic. No, i mean, you don't need to sound realistic. You need to work. You need to, you know, just work, right? You need to know exactly what their problem is, what the context is, how many times they have called, what is their history.
00:59:03
Speaker
Those tools are important, right? So that's where Elo comes in, where we built that. We built, because I come from, as I said, from telecom background. I built monster telecom systems, right, before... ah got distracted building games for 15 years. ah But, ah you know, that's my bread and butter.
00:59:23
Speaker
And so we built the system. We're very excited. We have a lot of customers using it. What are your customers using it for? ah So two, three things. ah One is obviously, you know, outbound calls, inbound calls, those kind of things are anyway happening.
00:59:38
Speaker
Like customer service or collections or sales? yeah Yeah. Some of those like, you know, lead qualification, you know, stuff like that. Right. You know, as I said, exit interviews, you know, so those are happening anyway. Some in-mong calls are also are happening where they they want, they're curious about something, they want clarification, stuff like that. So those are obviously low-hanging fruit. That's something we are doing pretty good job of.
01:00:04
Speaker
ah But what we noticed, ah because we the people who design the system are all engineers, we and we had that unique insight that all the software is becoming conversational.
01:00:17
Speaker
Now, how do we build something for software, right? So that's when we you know we made sure our software becomes easy, right? Like to imagine you... you go to a really good shop. I mean, you're in Japan, right? you're You go to a shop.
01:00:32
Speaker
a I don't think it happens in Japan, maybe because of the language problem. But ah you you you go to a a shop in US and, you know, you are some of the high end shops here in India. When you're talking to people and they, you know, you can say, yeah hey, I'm thinking of going on this walk or a trail and, you know, and then and the guy talks, to oh, then you're looking for this. This is perfect for you. This is what you should wear. Or you say, you know what, I saw that person with that blue thing and I really like it. And the person says, oh, that's last season. But we have this green thing which just came in. right Being able to connect at your wavelength and selling you, that doesn't happen.
01:01:10
Speaker
Now, imagine... You go to a website and you had that button where a shopping concierge is there, you click on it and the person can talk to at your wavelength, knows who you are, knows your history, what you used to buy and has a conversation where you can basically say, this is what I'm thinking versus search blue shit, right? Which is what the current search works, right? Versus I'm talking to it while it is telling me the screen in the background is changing.
01:01:38
Speaker
where it says this is what you need, right? And this one, I have your size, right? So it knows me and it can sell me as if it can see me, right? That's the world we are entering. and For that, the conversational voice needs to be re-looked at to build for software.
01:01:56
Speaker
And that's what we've done. but We see, you know imagine ah you're trying to book a flight, ah but you don't need to anymore press any button. You can just talk to the thing and the screen keeps changing, but you you're going to have that of conversation. So I think every software is going to become of conversation first. So what you would be selling here is not just the voice agent, but also some sort of an orchestra orchestration ability so that the voice agent can manipulate the software. Exactly. Exactly. We're giving SDKs and tools where software can become conversational, right?
01:02:38
Speaker
ah Where they have memory, where they have context, they have, you know, so it comes with all the tools needed for it to sound, familiar for it two for it to have the same wavelength as you, which is not possible right now.
01:02:52
Speaker
Has any customer deployed this so far? well This is like a pretty good category. I don't think there's any major competition here yet. not no ba Barely anybody is thinking about it. Everybody is using, trying to think about it from the off the shelf. Okay, you know can I just use Google Voice Agent? but ah The orchestration they need, the ability to move around, those elements are not there. Something which can handle memory, something which can handle context seamlessly, nobody has done a good job. So we we have like a lot of customers we're working with to implement this. of you know So we you know hopefully in a couple of months we should have some of those out.
01:03:30
Speaker
ah Indian companies or? A couple of them are Indian companies. A couple of them are US companies. Currently, my experience with the conversational, like voice conversational AI has been pretty janky. Like when the AI starts speaking back to you, if you speak in the middle, you know that it can't handle those back and forth interruptions. You have to wait for it to finish so that it's a smooth experience. Otherwise, if you speak in the middle, then things get derailed and all. How far away is that experience which you and me right now are having, where I can stop you in the middle and ask you to clarify, and you can immediately pivot into that clarification. and
01:04:07
Speaker
ah How far away are you from that? ah Not very far, but because you you you just need to evaluate, right? At the end of the day, how does this communication happen? i i'm like i Imagine I i basically say something like, you know,
01:04:22
Speaker
Oh, I can explain this to you in three points, right? And I talk about my first point and then you interrupt me, right? Cool, I answer that. Then I know i need to start back at second point.
01:04:33
Speaker
Right. An AI agent has no idea. Right. It basically takes everything fresh again. It's like, oh, that's done. Let me. ah So it's like for you, it feels like, hey, what happened? You know, you were telling me about those three points. I just needed a clarification. Why did you stop?
01:04:49
Speaker
Right. So there are those, you know, for us, humans is common sense. It's not because those things are designed piecemeal that way. Orchestration is done that way. It's just at the end of the day, folks like her sitting through and looking at every one of those problems and figuring out solution for them.
01:05:07
Speaker
right ah So we are probably weeks or months away from getting an experience which is going to be amazing. It's just that smart people need to work on it. Is a voice model possible? See, basically an LLM is what? It is fed...
01:05:26
Speaker
shitload of text and through that shitload of text it understands and it builds some sort of comprehension, some sort of predictability that okay, if it is like the sky is and the next word is likely to be blue or whatever. ah Is it possible to feed shitload of dialogue to create a LLM which is not going through this loop of speech to text and then that text is analyzed by an LLM and the response is generated and that response is again text to speech, that whole loop.
01:05:59
Speaker
ah Is it possible that this whole loop is not needed? it No, no, it is ah it is possible already. Some big guys are already doing it, Google to Sarwam2 and others. ah The technology is there.
01:06:11
Speaker
ah It's not rocket science, it's quite easy. ah There are two problems with it right now. ah The whole full cycle conversational models, ah voice models is one is they're expensive in terms of the amount of compute they take, versus you doing the cascading model where like, I'm gonna do speech to text, I'm gonna do LLM, I'm gonna do text to speech, versus just using the whole model. ah is more expensive.
01:06:38
Speaker
why Why is that? It's just the amount of compute they take to process something. there The cost is in training the model or even in the inference? No, no, no. infference Inference is the cost, right? Training is like a one-time excellence. Inference is where you are spending money. So conversation models are expensive. The second problem is they're not flexible.
01:06:56
Speaker
especially with tool integration, because think of it in a such way where I'm in the i'm having this conversation with you and I ask the models like, okay, you know, ah these are the credit cards or whatever, you know, thing. And it needs to be able to go look up a database and fetch this information about my past records. Oh, he's been rejected for this. you know It needs to have access to talking to various tools. It needs to have ability to go to RAC or a vector database to get some information. So there's so much it needs to do.
01:07:31
Speaker
That flexibility goes away when you basically give everything to the model ah except you know and run it through a prompt. So those are the problems which haven't been properly solved yet.
01:07:42
Speaker
I don't mean they won't be. But we are probably a year away, ah you know, from seeing a sophisticated a voice model. But, you know, as I said, ah you know, thanks to Skype, thanks to WhatsApp, in the olden days when we used to do telecom, we were told that ah anything more than 400 milliseconds, human ear will know that there's a delay.
01:08:07
Speaker
But thanks to WhatsApp, thanks to Skype and other tools we ended up using, Nobody cares for a second delay. delay And one second is what you can do right now with these technologies.
01:08:18
Speaker
So the the latency currently is one second from you saying something and then... One, 1.2 second. Yeah. Yeah. So, and people people use fillers and others with like, uh-huh, and you know, absolutely. And then then you throw in. So you could do all of that, but... You know you you can you can get away. you know Nobody cares if it is one second because humans have got used to it.
01:08:41
Speaker
Okay. Right. Interesting. Very interesting. And in six months' time, where do you think that one 2-second number will be at? ah That's a good question. the It all depends upon the expense. ah The delay is primarily at the LLM level.
01:09:01
Speaker
Text-to-speak speech to text is dearly, it's instantaneous, but... No, they come to about 600 milliseconds, right? Both combined. ah The other 600 milliseconds come from LLMs, which could go up or down depending upon how much intelligence you need.
01:09:19
Speaker
yeah ah so So if you are using smaller models, it's a simple, quick response, you can get away by you can get the response within 300 milliseconds, right?
01:09:30
Speaker
versus you're using a large language model where, you know, one of these 10 trillion parameter ones, right? ah I mean, obviously you shouldn't be using those because they will be expensive. But if you do use, then they we're talking two seconds later, right? So it depends upon how much intelligence you need in the conversation.
01:09:48
Speaker
But I don't see that going down too much. It's just that we would have filler, we would have all those things where humans would not be uncomfortable with the conversation.
01:10:00
Speaker
Okay, got it. ah ah Let me end with the asking you, what are your regrets in this journey of Purple Talk? We're ending either way, we're having fun.
01:10:12
Speaker
This... ah
01:10:17
Speaker
The thing is, there is always the missed opportunities, right? There were so many places. We felt like we were the first come of things, right? We were always there at the right time, right place.
01:10:29
Speaker
Just didn't execute. I like that quick commerce story which you told me. There were various, right? It was like we were the first company to launch an ad network when we started ah Purple Talk.
01:10:43
Speaker
We looked at, oh, hey, let's build games. But then there were so many people building games. We should also have an ad network and then and have analytics. So we launched a platform called AdShare when the App Store launched. We were the first ones.
01:10:56
Speaker
At the same time, there was an American company called Pinch Media which launched it. They were the second one, but we had like about four five hundred developers using us at that time.
01:11:07
Speaker
And, you know, and being in here in Hyderabad, I thought four five hundred was not a lot of developers using it because the news was hundreds of thousands of developers were building games and apps and all that. So four five hundred didn't seem like a big number.
01:11:22
Speaker
So eventually somebody came in and said, hey, this is a great technology. Do you want to sell? And story of our life, we said yes. We made a couple of hundred thousand couple of hundred thousand dollars again. ah And then I bumped into this Pinch Media guy.
01:11:37
Speaker
hey What are they worth today? ah Like a couple of months back and ah after I sold the thing and he was like, hey, what happened? How come you... You know, you shut it down.
01:11:48
Speaker
And I said, these guys bought it from me. And it's like, we we we've always looked at you guys as like this amazing product people and all these customers I wanted, you had. And I asked him how many customers you they had. They had like 40-year customers compared to me. And I thought they were like this big deal and they were doing well.
01:12:06
Speaker
And that company got sold for $400 million. dollars right uh so so basically these these are you know in retrospect you know being in india trying to serve u.s market that was a big problem you if you want to be if you are if you are building for american customers be in u.s so that you can talk to them you can understand what's working right so that something uh was a hard lesson which is we We just weren't talking. We were product builders. We just enjoyed building, but we weren't talking to people. That was a mistake.
01:12:42
Speaker
And that changed, obviously, once we became ah product owners, once we started looking at not just things from engineering point of view, but hey, we got to talk to customers. So that changed. But those are hard lessons, right?
01:12:54
Speaker
There were so many opportunities in last 18 years where we could just double down. We should have stayed longer. we should have, you know, or we should have only talked to people.
01:13:06
Speaker
But that said, you know, we're doing all right. You know, we we have i we we've tried about, like, say, 20 things. so Six, seven of them are doing quite well for us. So so no regrets that way. But, you know.
01:13:21
Speaker
yeah we We could have been billionaires instead of millionaires. theres Right, right. ah do Do you think you get too excited by this any new thing? It just seems to me like you enjoy the 0 to 1, but not the 1 to 10 so much.
01:13:36
Speaker
ah Yeah, maybe. That could be it. So you're you're constantly looking for the next 0 to 1 and the previous 0 to 1, which you got up to 1. Now that 1 to 10 also probably needs some system, either your attention or for you to figure out some way to have that 1 to 10. I think you should be the therapist for entrepreneurs, right? Yeah.
01:14:00
Speaker
It's like and I sit down on a couch and you're like, dude, focus. yeah i mean, you nailed it, which is a lot of my other entrepreneur friends have told me, right? Dude, you know, just focus. Don't do so many things. ah I'm lucky. I have six other co-founders ah who are lot more stable and are focused. So it it that gives me a chance to be the way I am.
01:14:29
Speaker
But... ah You know, you have that self-awareness that you are a zero to one guy. ah You must have developed a management style that lets you be a zero to one guy.
01:14:43
Speaker
ah What is that management style? What are some of those principles that let you be a zero to one guy and still have large outcomes? And I'm sure evolved and maybe those regrets which you spoke about would have driven that decision making. Okay, I need to fix this and you must have fixed it by now. um So what are those fixes?
01:15:03
Speaker
I have a friend whom I'm asking for. He's also a zero to one guy. He's nowhere near your scale, but i constantly tell him that, dude, you got to focus. And he is self-aware that he does not have that ability to focus. But I'm just wondering if I can give him a system.
01:15:20
Speaker
No, so there are two aspects here. One, which is a later realization, which has come in, which is oh you could be excited. You could have spikes of work where you work like three, four days continuously, and then you're like taking it easy for a bit. um Nothing beats consistency, right? Nothing beats, ah you know, ah a person who's probably not as high IQ, but can ask you question on regular basis, right? It's like, hey, what happened? Where is it?
01:15:53
Speaker
ah when When are you going to deliver this? You know, think being consistent ah with your team, ah where they know for sure that if they don't do it, there somebody is going to follow up, right?
01:16:06
Speaker
ah So being operationally strong is something nobody tells you as an entrepreneur. Everybody tells you as an entrepreneur, you got it. this guy was out there thinking of ideas, problems and all that. Yeah, hustle. Yeah, hustle part, right? weren't What is not talked about enough and which is basically, I would say, is as important as all the other aspects. The three aspects which need to go is luck, which is basically the timing. The second is obviously you having the right idea. The third one, which is as important, if not the most important thing, is consistency. being that guy who just doesn't drop the ball, right? Being that guy who just is always like, where is this? Where is this? Where is this? Where are you going to deliver this, right? Because if you are not that guy, ah then yeah, it's hard to succeed.
01:17:02
Speaker
Have you become that guy now? I wish, you know, no, i I just work with good people where I don't have to do that. But sometimes it just feels like, you know, ah early on in my career, if I would have been that guy, that would have been very helpful because then then i i was like, hey, it's not happening. You don't give up and move on to the next interesting thing. It's just that you ask it enough times, it's going to happen.
01:17:27
Speaker
or work with people who will who don't ah make you ask you so many times. right So those those are the aspects. So that's one thing I've learned a little bit late is the operational consistency is going to be super important to succeed, which most entrepreneurs don't do because they they just are excited about problem solving. But operational consistency is as important. So you need to have a great COO or you be that person till you have that great COO.
01:17:55
Speaker
And you are willing to give up equity to get that guy in? like Because you said you have six co-founders. I'm assuming you wouldn't have started with six, but over time, No, no, we we we we all started together. We are all, like, yeah, we all start together. All of us are still together.
01:18:12
Speaker
ah Yeah, so it's just that we all grew, right? At the end of the day, you know, as I said, that when i when I first built my first product way back in 2000, I didn't know I was building a prototype, right? So, and some of these guys were then too. So it's ah it's just... ah ah we We learned we all got better.
01:18:32
Speaker
ah you know That's the good thing, right? if you if you if you beat If you think of yourself as like the person you were 15 years, 20 years back, then obviously, you know, ah we none of us are the same people anymore. So we we we have all, hopefully we have all got gotten better.
01:18:48
Speaker
Amazing. Thank you so much for your time, Sridhar. It was a real pleasure. Thank you.