Govt Unveils State-Backed Multimodal LLM For Indian Languages

Union minister Jitendra Singh has launched the state-backed multimodal large language model (LLM) for Indian languages. 

Unlike traditional LLMs that predominantly take text as input, multimodal AI models can process various types of data, including text, images, audio and video. 

Speaking at the BharatGen Summit, Singh described BharatGen as a “national mission to create AI that is ethical, inclusive, multilingual, and deeply rooted in Indian values and ethos.” The platform integrates inputs such as text, speech, and image and offers AI solutions in 22 Indian languages.

“This initiative will empower critical sectors such as healthcare, education, agriculture and governance, delivering region-specific AI solutions that understand and serve every Indian,” said Singh.

The AI model was developed under the aegis of National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS) and implemented via IIT Bombay’s technology innovation hub (TIH) foundation for internet of things (IoT) and internet of everything (IoE). 

The project is also backed by the department of science and technology (DST) as well as a consortium of academic institutions, experts and innovators.

What Is BharatGen?

In October 2024, the Union government announced the BharatGen project, touted as the world’s first government-funded multimodal LLM project. At the time, the Centre said that the platform would focus on creating “efficient and inclusive AIs” in Indian languages and would be able to generate high-quality text and “multimodal content” in various Indian languages once completed.

The BharatGen platform has four key “distinguishing features”: 

• Multilingual and multimodal nature of foundation models
• Indigenously built datasets, which will be leveraged to train the LLMs
• Open-source architecture
• Development of an ecosystem of GenAI research in India

In order to “deeply capture” the nuances of Indian languages and address the paucity of data sets in Indic languages, the government then had said that BharatGen would develop processes for collecting and curating India-centric data that represents the country’s diverse languages, dialects, and cultural contexts.

As per the government, the project will also focus on data-efficient learning, particularly for Indian languages with limited digital presence. Besides, the open architecture of the platform will enable smaller businesses to leverage the platform to build products on top of this tech stack and linguistic datasets. 

The project is part of the Centre’s larger push for the AI ecosystem. Earlier this year, the government selected SarvamAI to build the country’s first indigenous foundational AI model. Last week, three more AI startups, Soket AI Labs and Gan.ai, were also selected to build homegrown LLMs. 

The projects are part of the government’s broader INR 10,372 Cr IndiaAI Mission, which was approved by the Union cabinet last year. The Missions aims to facilitate funding for emerging AI startups and spur innovation in the sector.

At the heart of all this is the growing Indian AI landscape, which as per Inc42 data, is projected to be a $17 Bn market opportunity by 2030.