Tech Mahindra and NVIDIA officially announced the launch of a new 8-billion (8B) parameter Hindi-first Large Language Model (LLM). Unveiled at the IndiaAI Impact Summit in New Delhi, the model is the latest and most powerful iteration of Tech Mahindra’s Project Indus initiative.
While previous versions of Project Indus focused on broad dialect support, this 8B model is specifically fine-tuned for the education sector, aiming to provide sovereign and scalable AI tutoring for millions of Indian students.
Key Features of the 8B Model
The model represents a major leap from Tech Mahindra’s earlier 1.2-billion parameter foundational model launched in 2024.
- “Education-First” Architecture: Built to provide a deep foundational understanding of STEM subjects, starting with Physics.
- Synthetic Token Training: To overcome the scarcity of high-quality Hindi data, Tech Mahindra used NVIDIA NeMo Data Designer to generate 500 million synthetic tokens, ensuring the model understands complex technical concepts in “natural Hindi.”
- Multilingual Dialect Support: While “Hindi-first,” the model retains Project Indus’s core capability to understand and converse in 37+ Indian dialects (including Bhojpuri, Maithili, and Dogri).
- Agentic AI Capabilities: The model supports Agentic AI, allowing developers to create autonomous AI tutors that can provide personalized, step-by-step guidance to students in their native tongue.
- Infrastructure: The model was developed using NVIDIA NeMo and is deployed through NVIDIA NIM microservices for production-ready scale.
Strategic Impact: The IndiaAI Mission
Tech Mahindra is one of the 12 key players selected under the IndiaAI Mission to build foundational models for the country.
| Feature | Strategic Goal |
| Sovereignty | Ensures student data and educational content remain within India’s digital ecosystem. |
| Accessibility | Designed for low total cost of ownership, making it feasible for budget-strapped schools and institutions. |
| Inclusivity | Bridges the gap for students who prefer studying in regional languages rather than English. |
| Deployment | Initially targeted for Intel AI PCs and eventually cloud-based mobile deployment. |
The “Less is More” Philosophy
Despite being smaller than global giants like GPT-4, Tech Mahindra’s Nikhil Malhotra (Chief Innovation Officer) emphasized that the 8B model follows a “quality over scale” approach. By focusing on rigorous data cleaning and removing over 10 types of bias, the model aims for higher accuracy in the specific cultural and educational context of India.
“Global foundational models are often not designed for countries with deep linguistic diversity like India… Our collaboration with NVIDIA addresses this gap by delivering a Hindi-first, sovereign AI model that enables scalable, relevant, and accessible learning.” — Nikhil Malhotra, Tech Mahindra.
