[Exclusive] How NASSCOM is Enabling BharatGPT Adoption in India

Building generative AI models is a humongous task in itself, and then helping organisations adopt them is another challenge. This is why BharatGPT creators Ganesh Ramakrishnan and Vizzhy, and SML founder Vishnu Vardhan approached NASSCOM in September 2023, to help their research take proper shape from the commercial standpoint.

NASSCOM’s head of AI Ankit Bose, in turn, showed tremendous support for the initiative and has been helping the IITs develop the model and make it enterprise ready. AIM got in touch with Bose to understand his and NASSCOM’s vision for the same. “We are specifically helping them get through the compute, personalisation, and connecting them with different bodies around India,” said Bose.

Optimistic about the potential impact of BharatGPT on the Indian AI landscape, Bose acknowledged the significance of the initiative. He drew parallels between the ongoing efforts in India and the historic development of global models like OpenAI’s GPT-3, emphasising that such foundational models require considerable time and effort to reach maturity, and BharatGPT is just getting started. 

Enterprise adoption of Indian models

Bose shed light on the timeline for enterprises to transition from the existing offerings by global tech giants to solutions built in India. He explained the stepwise process that enterprises typically follow, starting with problem identification, moving through closed AI APIs, and eventually transitioning to open-source AI models. 

“The aim is to reach a stage where Indic models become a viable option for enterprise adoption,” said Bose, highlighting the top contenders such as Llama, Mistral, and Phi-2 open source models. “These models have only a little amount of tokens from languages other than English, which makes them difficult to use,” Bose pointed out. 

The problem that Bose highlighted is the collection of Indic data and the gap in the number of people working on it. “BHASHINI is a very good start and the government has done a phenomenal job at this,” Bose said, adding that some OCR models are also coming up to convert physical data into digitised format for training models. “It is one of the most important tasks, as we can increase the whole country’s productivity, including people who don’t speak English.”

Bose believes that small language models specialise in different verticals and use cases is the right way to go as they give the ease of deployment. He gave the example of ARTi bot, a conversational chatbot for report generation and assistant for thought leadership at NASSCOM. “These specialised bots help in increasing productivity within the organisation,” he added. 

He also emphasised that companies should experiment with wrappers of AI models and then gradually decide if they should try building models out of open source or not. 

AI education and the rise of AI startups

Bose said that he has spoken to 122 generative AI startups in India, and noted that the number of startups is taking a massive jump. Currently, NASSCOM is working with 26 of them in their GenAI Foundry program helping them secure compute, which they provide with their partnership with CDAC and also helping them sign up with investors. 

“How can we take them global?” Bose added that it is one of the crucial things that NASSCOM has an expertise in while also helping them build their IPs. “Since most of them are technologists and business guys, they need help in taking their product global,” and that is why Bose and his team are taking focused sessions with them on developing business models for generative AI.

Bose says that NASSCOM is also majorly interested in the education sector of generative AI and has been partnering with the Ministry of Education, NCVET, and AICT to integrate generative AI courses into their curriculums for enhancing capabilities of the workers in every field. “The emphasis extends beyond technical knowledge to include ethics, responsibility, and safety, ensuring a holistic approach to AI education,” Bose concluded.