IIT Consortium Accelerates Development of BharatGen AI Model Across Campus Networks

By Subham Sharma, AI Correspondent
Subham Sharma holds a B.Tech in Computer Science from IIT Delhi and has been reporting on India’s AI ecosystem for over three years, with bylines in Science Today and Tech Insights.

India’s homegrown generative AI initiative, BharatGen, is witnessing a significant surge in development activity as teams across multiple IIT campuses ramp up research, data ingestion, and model fine-tuning. Launched on June 3, 2025, by Union Minister Dr. Jitendra Singh under the National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS), BharatGen aims to become the cornerstone of a sovereign Generative AI ecosystem tailored to India’s linguistic and cultural diversity linkedin.comapnipathshala.com.

5 Key Takeaways

1. Consortium Expansion: Twenty-five Technology Innovation Hubs (TIHs), anchored at IIT Bombay, IIT Madras, IIT Kanpur, IIT Hyderabad, and IIT Mandi, are now fully operational in data collection and model training pipelines bharatgen.techen.wikipedia.org.
2. Compute Scaling: The IndiaAI Compute Facility recently added 2,000 NVIDIA H100 GPUs across IIT data centers to accelerate large-scale training runs.
3. Multilingual Focus: BharatGen currently supports generation and understanding in 22 Indian languages, from Hindi and Tamil to Assamese and Gujarati globalindiannewsnetwork.com.
4. Collaborative Hackathon: The ongoing Generative AI Hackathon 2025, hosted at IIT Bombay, has drawn over 400 student teams working on domain-specific BharatGen applications.
5. Ethical Guardrails: Under guidance from the Department of Science and Technology (DST), the consortium has established an AI Ethics Board comprising experts from IIT Delhi, IIIT Hyderabad, and the EkStep Foundation.

Ramp-Up in IIT Networks

IIT Bombay: Technical Backbone

As the lead node, TIH IIT Bombay has orchestrated the integration of cross-campus data lakes and optimized model parallelism across GPU clusters. “Our next milestone is achieving inference latency under 50ms for Hindi text generation,” says Professor Ganesh Ramakrishnan, head of the TIH Foundation for IoT & IoE at IIT Bombay globalindiannewsnetwork.com.

IIT Madras & IIT Kanpur: Domain Adaptation

Teams at IIT Madras focus on healthcare-specific fine-tuning—enabling BharatGen to draft patient education materials in regional languages—while IIT Kanpur researchers optimize the model for legal document summarization across state statutes.

IIT Hyderabad & IIT Mandi: Data Localization

IIT Hyderabad leads efforts on code-switching detection between English and Indian languages, critical for WhatsApp-style conversational AI. Concurrently, IIT Mandi curates vernacular datasets by partnering with local media outlets and community radio stations.

What Happens Next?

• Monthly “Daily Digest” Updates: Starting July 2025, the DST will publish a public progress report highlighting key performance metrics, new language support, and case studies of pilot deployments in government services.
• National Rollout Trials: Pilot implementations in e-Governance portals in Rajasthan and Kerala are slated for Q4 2025, testing BharatGen’s multilingual chatbots in real-world citizen engagement.
• Open-Source Release v1.1: An updated codebase with optimized model weights (up to 40B parameters) is expected to be released under an Apache 2.0 license by December 2025.

Daily Digest: Ongoing Highlights

• June 24, 2025: IIT Bombay team reduced training loss by 12% through mixed-precision techniques.
• June 23, 2025: Over 150 hackathon prototypes submitted, covering agriculture advisories in Marathi and Telugu.
• June 22, 2025: Ethics Board approved guidelines on bias assessment for caste and gender representation.

Fact-Check