India’s language database platform, Bhashini is working on collecting datasets in regional languages, which is essential to bridge digital divide and gaps in terms of technology, Amitabh Nag, chief executive officer of Bhashini, told FE.

This assumes significance as currently large-language models (LLMs) are mostly restricted to the English language, to train generative AI models. With Bhashini investing to collect data in regional languages, the government will be able to provide generative AI services to people in their own dialects.

Among the usecases, currently PM Kisan, which is integrated with Bhashini offers chatbot support in various Indian languages. Besides, the National Payment Corporation of India (NPCI) has collaborated with Bhashini to allow users make conversational payments through UPI in Indian languages.

Lately, IIT Madras is also piloting Bhashini platform to provide students with access to courses in regional languages.

What makes Bhashini different from other large-language models is its ability to understand scenic text, handwriting, and printed texts. “Multimodality is still not there in large language models. Bhashini can understand computer text, it can understand voice, it can give you an output on voice, it can give you output on text,” Nag said.

Besides working on collecting datasets in Indian langauges, currently, Bhashini is offering real-time translation from one language to another, machine translation APIs to developers for integration, and speech recognition/synthesis. The government-owned platform is being used by over 20 organisations across industry and government, whereas the impact on people is huge.

When asked about the way forward for Bhashini, Nag said, “the focus currently is to increase its adoption. We would, perhaps, unmute the roadmaps as we see an increase in adoption”.

Bhashini is expected to play a key role in India’s AI mission, datasets platform, and help the government in creating a large repository of datasets, which will help companies and researchers to train their AI models.

On plans to monetise the initiative, Nag said currently there is no such plan but to create an impact for which Bhashini was started. However, going forward, based on demand and adoption, it might get into a model where it will be able to just cover its cost, being a section 8 non-profit organisation.