By Mudit Dube

Copyright newsbytesapp

BharatGen, a government-backed consortium led by IIT Bombay, has been awarded over ₹900 crore under the IndiaAI Mission. The money will be used to create a 1-trillion parameter large language model (LLM) for India. The ambitious project aims to distill this massive AI model into smaller, domain-specific systems for various sectors such as law, agriculture, and finance. Rishi Bal, Executive VP at BharatGen, explained that the trillion-parameter model is not meant for direct consumer use. Instead, it will serve as a base or mother system to create smaller and more efficient models for specialized areas. These could be agricultural advisory tools in regional languages or legal assistants trained on Indian case law. BharatGen is heavily investing in building a sovereign dataset by collating multiple streams of Indian content. The consortium is working with publishers to license their archives and create digital corpora, providing free OCR services to digitize regional texts, and using crowdsourced annotation to capture nuances in Indian languages and culture. These efforts are aimed at ensuring the model reflects Indian contexts rather than relying on foreign data. Training a trillion-parameter model requires thousands of GPUs working in parallel. Bal admitted that hardware availability is a challenge and they have to wait for GPU supply like everyone else. The ₹900 crore funding from the government will be used as a subsidy for availing GPUs. Under the IndiaAI mission, nearly 40,000 GPUs have been made available for different activities including building India’s sovereign LLM models. BharatGen CEO Ganesh Ramakrishnan said their focus is on creating models rooted in Indian data and languages. He emphasized the importance of reliability for real-world applications over raw scale. The company plans to release distilled models to the developer ecosystem, allowing start-ups and enterprises to build applications without having to train massive systems independently. BharatGen operates on a hub-and-spoke model with teams across multiple locations in India. Bal said this structure allows them to bring together engineers, data scientists, and domain experts while keeping operations lean. Ramakrishnan said BharatGen is also exploring public-private partnerships and revenue models like licensing smaller distilled models.