BharatGen Sovereign AI Models initiative is India’s first government-supported effort building foundational AI models across text, speech and vision, tailored to Indian languages and societal needs, as highlighted by PIB.
BharatGen Sovereign AI Models: India’s Government-Backed Multimodal AI Initiative
BharatGen is the first government-supported national initiative to develop a range of sovereign foundational AI models tailored to Indian languages and societal contexts. It spans multiple modalities, including text (via Large Language Models), speech (Text-to-Speech and Automatic Speech Recognition), and vision-language systems.
Currently, BharatGen’s AI models support 15 Indian languages which include Hindi, Assamese, Bengali, Gujarati, Kannada, Maithili, Malayalam, Marathi, Nepali, Oriya, Punjabi, Sanskrit, Sindhi, Tamil and Telugu. Soon, all 22 scheduled Indian languages will be covered.
BharatGen has released domain-specific fine-tuned models for Ayurveda (Ayur Param), Indian agriculture (Agri Param) and Indian legal domain (Legal Param). In addition, all BharatGen models (text, speech and vision) are useful for applications across healthcare, agriculture, education and governance.
Two Technology Innovation Hubs, namely TIH Foundation for IoT and IoE, IIT Bombay and IITM Pravartak Technologies Foundation, IIT Madras, under the National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS) of Department of Science and Technology (DST) are currently active as part of BharatGen network.
The following institutions are a part of the BharatGen consortium:
Institution Name |
Role in BharatGen |
|---|---|
Indian Institute of Technology, Bombayl |
Lead institution, guiding research and integration across consortium partners |
International Institute of Information Technology, Hyderabad |
Vision-language document modeling |
Indian Institute of Technology, Madras |
Speech foundation model development and evaluation |
Indian Institute of Technology, Kanpur |
Legal AI research, domain-specific datasets, and developing tokenization strategies for multilingual models |
Indian Institute of Technology, Hyderabad |
Advanced tokenization and vocabulary optimization for large multilingual LLMs |
Indian Institute of Technology, Mandi |
Inclusive multilingual model development and research on efficient training strategies for LLMs |
Indian Institute of Management, Indore |
Bharat-centric evaluation and benchmarking of LLMs, multilingual and multimodal data collection |
