Overview
- The expansion scales Project Indus from a 1.2B-parameter model to an 8B-parameter architecture for classroom-aligned explanations in core subjects.
- Training draws on India-relevant datasets supplemented with about 0.5 billion synthetic tokens generated using NVIDIA’s NeMo Data Designer.
- The system is built on NVIDIA’s NeMo platform, NIM microservices and AI Enterprise to enable production deployment and scaling.
- Tech Mahindra says the tool is intended for schools, higher education, skilling centres and digital learning platforms, with pilot details and timelines to follow.
- One report says the model builds on Meta’s Llama 3.1 8B Instruct and uses AWS infrastructure, a detail the companies have not confirmed; the effort sits within the IndiaAI Mission push for locally aligned models.