/dq/media/media_files/ryfkg9bCCCZCYrw9Ici2.png)
India’s sovereign AI efforts received a major boost with the government-backed BharatGen initiative, supported by Rs. 988.6 crore from the Ministry of Electronics and Information Technology (MeitY). Among the academic partners in the consortium, the International Institute of Information Technology Hyderabad (IIITH) is playing a central role in advancing Indic Vision-Language Models (VLMs).
BharatGen: A national mission in Generative AI
BharatGen is led by IIT Bombay under the National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS). It brings together institutions including IIIT Hyderabad, IIT Mandi, IIT Kanpur, IIT Hyderabad, IIM Indore, and IIT Madras. The initiative is designed to build India-specific solutions that make AI accessible across multiple Indian languages, addressing the country’s linguistic diversity.
Patram: Indic Vision-Language Foundational Model
IIITH’s team, led by Prof. RaviKiran Sarvadevabhatla, recently introduced Patram, India’s first Vision-Language foundational model for Indic documents. Built with 7 billion parameters, Patram can process scanned or photographed documents and respond to queries about them. Prof. Ravi Kiran explained that the model was trained from scratch, with curated datasets ensuring its effectiveness across varied use cases. He added that Patram can also be fine-tuned for domains such as medicine or law.
eVikrAI: Indic E-Commerce Model
The team also contributed to eVikrAI, a Vision-Language Model for Indic e-commerce. It helps sellers create and translate product descriptions across multiple Indian languages, reducing the reliance on manual cataloguing.
IIITH has been involved in national AI programs, including its contributions to Bhashini, India’s language translation mission. The institute’s Centre for Visual Information Technology (CVIT) has developed research in vision-language technologies for years, with students playing an active role in these projects.
Future Research and Expansion
Prof. Ravi Kiran noted that future work includes improved versions of Patram, capable of handling multipage documents and text highlighting. He also emphasised the importance of partnerships to develop domain-specific models, such as for science. While the models are open source, the training methods are not yet public, though the team plans to share them at a later stage.
IIITH Director Prof. Sandeep Shukla said BharatGen could support sectors such as finance, agriculture, law, and cybersecurity. Telangana IT Minister Duddilla Sridhar Babu highlighted IIITH’s early contributions through Patram and Bhashini, adding that the new government funding will further strengthen India’s AI capabilities.