Sarvam aims to shape India's GenAI landscape: Co-founder Raghavan

Focus on efficient models and unique needs of the country mark Sarvam's journey

The year 2023 was all about generative AI (GenAI) and the power it gives to the users, as shown by OpenAI. India saw its share of activity in this space as well. Sarvam, the first startup to announce its intention to create a full stack for GenAI with a focus on India’s unique needs, was founded by Vivek Raghavan and Pratyush Kumar. The firm managed to raise $41 million in a Series A round of funding. In an interaction over a video call with Shivani Shinde, Raghavan talks about how Sarvam is planning to boost India's GenAI space. Edited excerpts below:

Is Sarvam focused on creating a large language model (LLM) like OpenAI or other such systems?

One of the things we believe is that creating a large language model (LLM) may not be the big focus. We are creating small models that are efficient and much more cost-effective than the global models. But at the same time, they are good for making a change in India. That's really where we want to focus.

While we will be focusing on Indian languages, we'll also work on use cases, which may or may not have anything to do with language.

How are you going about it?

We have taken the open-source route and created Open Hathi. When you are talking about large models like OpenAI or Gemini or others, we are not attempting to build models of that size at this point. Because it needs a certain level of financial resources. The second part is data. Collecting data from an Indian context and in Indian languages is a very complex thing. And then you need a significant amount of compute to train these models. This may sound easy, but it’s not. You also need talent that can create this. We, fortunately, have a very talented team. For us, it is also an opportunity to do something in deep tech in India. That is something that motivated us to start this.

Open AI is getting challenged in US courts, especially on copyright. Is there a right answer to this?

The copyright issue and other related issues will play out over the next few years. And different languages and countries are taking different views. For example, Japan has a view on it, whereas India has not taken any so far. Japan is saying that using data to train a model is correct. It is encouraged in Japan. For a country like Japan, which is fairly conservative, has taken this stand.

How significant is voice going to be in India when we talk of systems like OpenAI?

If GenAI is going to be consumed in a large amount here, then I think it would be through the media of voice. Because we have made it very hard to type in Indian languages. In many cases, these interfaces will be voice-based. We are thinking of a voice-first kind of GenAI.

Where will Sarvam be in a year?

The way this field is changing, I can't even tell you what I am doing one month from now. I hope that this technology is good enough and that people’s access to both knowledge and services is dramatically improved.

I hope there are significant applications where people are using uniquely Indian channels. You'll see some big adoption of those things. The use cases are key, and I think the use cases will be different than what you see in how Gen AI is used in the developed world.