Tech Mahindra partners with NVIDIA to launch an education-focused Large Language Model (LLM) as part of its Project Indus initiative. The model, described by the companies as Hindi-first and built with the Indian language context to make it accessible to Hindi speakers and, over time, other languages.
The model, scaled to an 8-billion-parameter architecture, represents a significant expansion from Tech Mahindra’s earlier foundational LLM of about 1.2 billion parameters. According to company statements, the system is designed to deliver AI-powered educational support with an emphasis on India’s linguistic and cultural diversity, starting with Hindi and extending to other Indian languages over time.
Also Read: 5 Reasons HDDs Will Continue to Dominate Enterprise Storage in the AI Era
To reduce data scarcity in Indian languages, the development team generated hundreds of millions of synthetic training tokens. The model was built on NVIDIA’s AI stack, with the NeMo framework managing large-scale training and NIM microservices supporting deployment in production settings. Together, these components provide the computational backbone needed to train, optimise and serve an 8-billion-parameter model reliably at scale.
According to statements from the organisations, the model targets a broad education agenda by providing foundational learning support to students across academic disciplines such as physics and other core subjects. By grounding the model in Hindi and culturally relevant context, the initiative seeks to address limitations posed by global generative AI models that often prioritise English-centric training data. The platform also incorporates Agentic AI functionality, which enables autonomous conversational agents capable of interacting in natural Hindi to support learning scenarios.
Nikhil Malhotra, Chief Innovation Officer & Global Head of AI and Emerging Technologies, Tech Mahindra, said that the collaboration responds to a recognised gap in domain-trained language models suited to local languages and educational contexts. NVIDIA’s representative emphasised the global trend toward “sovereign AI,” with demand for foundational models customised to specific linguistic and regulatory environments.
The effort forms part of wider industry interest in developing AI systems aligned to national priorities, notably in regions with deep language variation. The announcement did not reveal particular deployment timelines or partner institutions beyond the technology providers, it positions Project Indus as a locally grounded alternative to generic large language models for education and related civic services.
