AI is like a marathon, but you have to run it like a sprint: Ganesh Gopalan of Gnani.ai

One of the oldest artificial intelligence (AI) firms of the country, Gnani.ai has recently launched Gnani Artha, an end-to-end sovereign AI stack for Indian enterprises and public institutions, built on two key components -- Gnani Evon v3.3, a 30-billion-parameter open-weights model trained natively across 11 Indian languages -- and Gnani Plexus, the company’s agentic AI platform. Gnani Artha is backed by the government’s India AI Mission.

The company that has the largest deployments of voice AI in the world, in terms of volumes, is now looking at family of models with such models. In an interview with businessline, Ganesh Gopalan, Co-Founder and Chief Executive Officer, Gnani.ai said that as AI is still in the nascent phase in India there are a lot of scope for everyone, but also added that there is no time to look sideways at competition right now and one should focus on doing things they are best at.

Gnani Artha is really our sovereign platform, which will host a bunch of models. It is a 30-billion parameter model. What we are looking to do in the future, there will be a whole family of models. We are planning to launch 100-plus billion parameter models. First is 70-billion parameter, then 100 plus billion parameter and each model has its own uses. For the reason why we have launched this 30 billion parameter model, will be very different from the reasons why we launch a 100- billion parameter model. This parameter model we wanted to do for Indic language capability, and there’s a benchmark that we have called MILU. The tokens consumed here for languages like Gujarati, Malayalam, which is always tougher points for global models, which they struggle, consume at least 40 per cent less tokens than them. We are offering it as a sovereign piece and anybody can take it and launch it within their own servers with complete security.

India being a country of multiple languages and also more dialects, how difficult or challenging is to cover them in your models?

That’s a specific challenge and if you see global languages, global models, they don’t work for this. But, we also have speech text models that work with almost every language of India and we call it Prisma. That’s the most accurate speech to text model in the industry today, and there again, we are adding more languages. Like you said, the biggest challenges are some languages which don’t have a written text. But, we basically started our company as a speech to text company right back in 2017 and we built India’s first speech to text models. And, now we are looking at launching more languages. We are collecting data everyday for that. We have the largest dedicated data set in the world for Indian languages.

As competition is rising, especially with a lot of startups and companies from foreign lands too in this space, and some even launching models for free, how do you plan to beat them?

I think firstly, this is not the time to look sideways at competition. We are in a business that is expanding at an incredible pace and we are actually saying no to a lot of business because we are unable to expand our sales and support teams at the pace that customers are willing to work with us. And, there are going to be opportunities for multiple people. AI is like a marathon, but it’s a marathon that you have to run it like a sprint, and in a sprint the moment you start looking to sideways, you are gone. So, there is a place for everybody, and you should focus on the battles that you are fighting. For instance, a typical model, a lot of people think that it is one metric that defines a model. But, it is not. It’s like Onam Sadhya. You have so many elements in an Onam Sadhya, each tastes different, and each person would have a liking for a different dish when ask them. So that’s what the models are for, and a competitor may be focused on different problems than what we are focused on. I think there is just enough business and market because we are in day zero of AI and we have a lot more to go.

On the policies side, the government is planning for AI regulation now. Do you support the idea and what inputs would you like to give to government for that?

It’s a very tricky thing...do you want to go like the Europe way or the US way. I think India will go somewhere in between. Probably more towards the US way, but it will be somewhere in between because you have to realise that being a democracy of such a large population we should definitely also be very careful of deepfakes, and problems like these, which would cause issues within the country. So, even though I am an AI practitioner, I love my country more and I do feel that there has to be certain amount of regulation for sure. But, it has to be done in such a way that it doesn’t stifle the industry. It’s very important, we don’t have a Silicon Valley in India.

Comments have to be in English, and in full sentences. They cannot be abusive or personal. Please abide by our community guidelines for posting your comments.

We have migrated to a new commenting platform. If you are already a registered user of TheHindu Businessline and logged in, you may continue to engage with our articles. If you do not have an account please register and login to post comments. Users can access their older comments by logging into their accounts on Vuukle.