I am Dheemanth Reddy, co-founder and CEO of Maya Research. We are building the voice interface for the next five billion people, the ones who will come online expecting to talk to technology in their own language and be understood, rather than adapt to keyboards and English menus the way the first billion did.
My reason for this is personal. I grew up in Vempalli, a small town in Andhra Pradesh, in an ordinary middle-class home. My parents are educated, but in their own language, and technology was never really built for them. The picture I carry is my mother sitting beside me while I used the phone, telling me everything she wished she could do herself. To leave my town I taught myself English from scratch, which eventually took me to a master's at New York University, where I studied deep learning under Yann LeCun. That gave me two things: the belief that this problem was tractable, and a lasting bias toward proving things rather than claiming them.
Maya comes straight out of that. Maya 1, our open-weight model, became one of the most downloaded speech models in the world, with more than 330,000 downloads. Maya 2 ranks number one for Hindi on Voice Arena, ahead of the largest labs, and among the top five for English, judged by tens of thousands of blind human votes. We built this with a lean team and modest capital, backed by South Park Commons, by staying close to the problem and doing the groundwork others skip.
What comes next is full-duplex speech-to-speech: voice that is cheaper, more natural, and runs on-device. The test I hold myself to is simple, whether my own mother could use what we build, on her own, in her own language.
Dheemanth Reddy is the co-founder and CEO of Maya Research, building the voice interface for the next five billion people. His open speech models rank among the world's best.
What inspired you to build Maya Research, and what gap did you see in the AI space?
It started at home, long before it was technical. I grew up in Vempalli, a small town in Andhra Pradesh, in an ordinary middle-class family. My parents are educated, but in their own language, and they were never comfortable with technology, not because they could not learn it, but because using it meant leaving their language behind for one built for someone else. The image that stays with me is my mother sitting beside me while I used the phone, telling me all the things she wished she could do herself.
The gap I saw is that the industry optimises for the roughly one billion people who already type in English, and treats everyone else as an afterthought. Voice, the most natural thing a human does, was still robotic, flat and English-first. So I set out to build the one thing that removes the barrier for the rest of the world: speech that sounds genuinely native, carries real emotion, and works across the languages the big labs ignore.
Maya Research is building a voice interface for billions of users. What is the long-term vision behind this mission?
The next five billion people to come online will not adapt to keyboards and English menus the way the first billion did. They will expect to talk to technology, in their own language, and be understood. Maya builds the speech models that make that possible: voice that carries real emotion, sounds genuinely native across many languages, and actually understands what a person means.
I believe voice, not text, becomes the default interface for AI for a simple reason. Speaking is the most natural thing a human does, and it carries emotion and intent in a way typing never will. Text forces you to learn a system, read a screen and type, usually in English. Voice removes that entire layer. Long term, Maya is the voice layer the world builds on, from India to Southeast Asia to the Middle East and beyond, so that talking to a machine finally feels like talking to a person.
What does the recent funding milestone mean for your roadmap and growth plans?
Closing our seed round, backed by South Park Commons, does one important thing: it buys us focus and time to do the hard, unglamorous work well. Practically, it goes into three places, compute, proprietary speech data for languages the internet ignores, and a small, deep team.
It lets us push Maya 2 across more of the world's major languages, bring the cost of running our models down further, and start building the harder frontier, a full-duplex speech-to-speech model that lets people simply talk and be talked to, with no text in between. The milestone is not the money. It is the runway to keep proving that a lean team can build at the frontier.
What has been the biggest challenge you've faced as a founder, and how did you overcome it?
The honest answer is compute. Not talent, not model quality, compute. As a small team, the hardest constraint has been access to affordable GPUs to train and serve frontier models.
We got past it two ways. First, we turned the constraint into a discipline: models purpose-built for speech rather than a general system carrying weight it does not need, proprietary data that lets a smaller model learn more from less, and a team that iterates fast instead of running dozens of expensive experiments in parallel. Second, we worked closely with leading inference providers to bring the cost of running our models down. When you cannot outspend the biggest labs, you are forced to out-think them on efficiency, and over time that becomes an advantage rather than a limitation.
How do you think AI will transform everyday life over the next five years?
Within five years, talking will become a normal way people use technology, not a novelty. The keyboard and the English menu stop being the entry ticket. A farmer, a shopkeeper, a grandparent, anyone, will speak to a device in their own language, mixing languages mid-sentence the way people actually talk, and simply be understood.
AI moves from something you operate to something you converse with. Customer support, healthcare, education, commerce and everyday services will be voice-first, and for most of the world that will be the first time technology truly meets them where they are. The interface effectively disappears. You just speak.
What leadership principle has had the greatest impact on your entrepreneurial journey?
Prove things, do not claim them. I picked that up at NYU and it runs through everything we do. It is why we ship open-weight models and put our work on public leaderboards judged by tens of thousands of blind human votes, rather than polished private demos.
It keeps the team honest, forces the work to actually be better rather than better in a room we control, and it earns trust far faster than any pitch. Conviction plus proof beats noise, every time.
What advice would you give to aspiring founders building AI-first companies?
Pick the hardest, most valuable problem that you have a genuine edge on, and refuse to do anything else. Do not try to be a general-purpose lab. Our edge was the languages, data and voices the rest of the world treats as an afterthought, so that is where we went, and a small, focused team can be the best in the world at exactly that.
Second, do the unglamorous groundwork everyone else skips. For us that was building our own speech data, because that is what is actually defensible. And put your work in the open early, the real world will stress-test it in ways you never could alone, and that feedback compounds. Constraints are not your enemy. They force the good habits that become your moat.
Looking ahead, what legacy do you hope Maya Research will create?
I want Maya to be the reason talking to a machine, in any language, stops feeling like a machine, and I want the people who were left out the longest to be the first ones let in. If we make voice drastically cheaper, genuinely native across the world's languages, and available to every developer building for the next five billion, that is the legacy.
But the test I actually hold myself to is personal. It is whether my own mother could use what we build, on her own, in her own language, with no one sitting beside her. The day that is true for her, and for the hundreds of millions like her, is the day I will feel we truly succeeded. Everything else is in service of that.