Tech Mahindra and NVIDIA are advancing Project Indus with a Hindi-first education-focused LLM designed to enable scalable, sovereign AI tailored to India’s linguistic and educational needs.

Industry leaders emphasize locally grounded models and sovereign AI as critical to inclusive digital infrastructure, with NVIDIA stressing production-ready performance and scale.

Nikhil Malhotra of Tech Mahindra highlights the necessity of language- and context-rich models aligned to India’s diverse languages and educational contexts.

The model leverages NVIDIA technologies, including NeMo and NIM microservices, and uses nearly half a billion synthetic tokens from NVIDIA NeMo Data Designer to address data scarcity across languages.

NVIDIA infrastructure and frameworks ensure production readiness, reliability, and scalability for the education-focused LLM.

The project has scaled the model from 1.2 billion to 8 billion parameters, signaling a substantial upsize, with NVIDIA’s open research platform expected to trim training time and boost educational quality.

Future enhancements under consideration include adopting NVIDIA Nemotron to shorten training cycles and raise instructional quality for education and public-sector domain needs in India.

The initiative contemplates integrating with NVIDIA Nemotron to accelerate sovereign AI development and improve training efficiency for India-specific domains, including the public sector.

Evolving to leverage the Nemotron ecosystem, the project aims to shorten training cycles and enhance instruction quality for India’s education and public-sector requirements.

Project Indus has grown from about 1.2 billion parameters to an 8-billion-parameter architecture for the education-focused rollout.

The Hindi-first model builds on Project Indus, expanding to eight billion parameters to support education with emphasis on Hindi and other Indian languages.

Project Indus aligns with India’s national learning priorities and multilingual, culturally diverse curricula.