The new model aims to promote democratisation in the education sector by providing a foundational understanding of complex subjects like physics in natural Hindi
The new education-focused model uses a more advanced architecture with 8 Bn parameters
The company had earlier launched a multimodal LLM for telecom operators back in March 2025
IT services major Tech Mahindra has announced the launch of a new Hindi-first LLM focused on education developed in collaboration with NVIDIA as part of its “Project Indus” initiative.
The new model aims to promote democratisation in the education sector by providing a foundational understanding of complex subjects like physics in natural Hindi. It also supports the development of autonomous AI agents that can speak and respond in fluent Hindi.
While the first version of Project Indus featured 1.2 Bn parameters, the new education-focused model uses a more advanced architecture with 8 Bn parameters. The use of NVIDIA’s advanced AI frameworks and infrastructure support, such as the NeMo framework and NVIDIA NIM microservices, enabled these advancements.
Given the lack of Hindi-language data sets, it has been trained on half a billion synthetic tokens (artificially-created data) generated by NVIDIA’s NeMo Data Designer.
“AI is becoming central to national digital infrastructure and inclusive growth, but global foundational models are often not designed for countries with deep linguistic and cultural diversity like India. A key industry challenge is the lack of domain-trained language models grounded in local languages and learning contexts, particularly in education. Through Project Indus, our collaboration with NVIDIA directly addresses this gap by delivering a Hindi-first, sovereign AI model that enables scalable, relevant, and accessible AI-powered learning and citizen-centric services for India,” Tech Mahindra’s CIO Nikhil Malhotra said.
The company had earlier launched a multimodal LLM for telecom operators back in March 2025. That model is also developed on top of NVIDIA AI Enterprise software as well as AWS cloud infrastructure, the LLM will be based on Meta’s Llama 3.1 8b instruct model.
Back then, Tech Mahindra said that the new model will pave the way for the transformation of traditional telecom networks into fully autonomous networks.
Tech Mahindra was also one of the eight entities selected by the Central Government to develop homegrown foundational LLMs under the IndiaAI Mission. Besides, Fractal Analytics, Avataar AI, IIT Bombay’s BharatGen, among others have also been developing homegrown LLMs under the IndiaAI Mission.
Important to highlight that the India AI Innovation summit has seen multiple sovereign models unveiled over the past few days. These include Sarvam’s 105 Bn and 30 Bn parameter LLMs, BharatGen’s 17 Bn multilingual foundation model and Gnani.ai’s voice cloning/text-to-speech system, all optimised for Indian languages and applications.
The event also featured AI hardware like smart glasses, collaborations to bring AI to feature phones, new data platforms, and innovation awards to promising startups.
Over the past few days, NVIDIA has also deepened its push into India’s AI startup ecosystem with investments and partnerships aimed at growing early-stage AI firms, even as its broader strategic moves in AI hardware and infrastructure continue to attract global attention.
Comments are closed.