INDIC-MMR
The INDIC-MMR project aims to develop the first multimodal large language model capable of culturally grounded and multilingual reasoning for the Indian context. It will introduce a scalable data engine to build two key resources: Indic-Caption for cultural visual understanding and Indic-Instruct for multimodal instruction tuning with chain-of-thought supervision. The model will be trained through a three-stage pipeline involving cultural vision adaptation, vision–language alignment, and reasoning-focused instruction tuning. The system will be evaluated on DRISHTIKON, BharatBench, and a newly proposed INDIC-MMR benchmark. The final outcome will be a deployable prototype that supports educators, researchers, and users in culturally aware AI applications.