Sarvam co-founder Vivek Raghavan says that while the company has the technical ability to build the kind of Large Language Models (LLM) that Google is building, it would require huge amounts of capital. Raghavan tells Poulomi Chatterjee that being a full-stack platform, it should make every Indian’s life better. Excerpts:
Q. Having built a couple of LLMs, what are the challenges ahead of you?
We’re competing on a global scale with Big Tech companies which have many more researchers and funding than we us. We are building sovereign AI which means respecting the multilingual and regional diversity of India while also being data sovereign. This means that our AI models are built from scratch with datasets we built without any dependency on others. The key thing is we’re also full-stack platform so the foundational work we do needs to be such that every Indian can make their life better.
Q. Sarvam claims its models are comparable with those of OpenAI and Google’s AI models…
A. Of course, those AI models are much larger than ours. But for specific tasks that we care about, we believe we can do better than them. I don’t claim that Sarvam’s AI models are frontier models but the tasks we care about, or a specific Indian business cares about, that we can do better.
Q. Do you have plans to build similar massive Large Language Models like theirs?
A. We have the technical chops to do that but the amount of capital required for that is a lot. So, if the opportunity arises eventually, we can do something like that but right now I can’t say.
Q. Where do you go now that you have forayed into hardware with the Kaze smart glasses?
A. We look at the glasses not as a consumer but as a business device. It felt natural because this is a continuum of our view that voice is the most important modality out there. The glasses are just in a different form factor. There are also potential use-cases outdoors where say traffic policemen can use it. Our goal is not to compete with Meta’ RayBan glasses – the features and the price point is very different.
Q. Is the pace of revenue generation picking up?
A. We are making steady progress. There are more than 100 companies which leverage our APIs to build custom solutions and our conversational AI products are doing more than a million minutes in a day of calling. Similarly, we also have some Agentic AI orchestrations which are used in a number of interesting use cases. I believe we are seeing good revenue traction but we’re still very early.