Research
Our research is organized into ten major divisions that share methods, datasets, and scholars. Each pillar pursues foundational questions and the systems that translate them into the world.
Theoretical and empirical research on optimization, generalization, and the principles that govern how machines learn from data at scale.
Focus areas
Designing, scaling and understanding neural networks — from attention mechanisms to billion-parameter foundation models trained on heterogeneous corpora.
Focus areas
Vision-language understanding, 3D scene reasoning and medical image analysis for autonomous systems and clinical applications.
Focus areas
Multilingual NLP with a deep focus on low-resource Indian languages — building models that speak, read and reason in 22+ tongues.
Focus areas
Diffusion, autoregressive and multi-modal generative systems aligned with human intent — from text and images to molecules and code.
Focus areas
Clinical decision support, diagnostic imaging and predictive medicine — built and validated with AIIMS and three partner hospitals.
Focus areas
Methods for interpreting, auditing and certifying machine decisions in high-stakes settings — law, medicine, and public policy.
Focus areas
Neuro-symbolic reasoning over knowledge graphs — structured knowledge as a substrate for verifiable, multi-hop AI reasoning.
Focus areas
Sparse, quantized and distilled models for constrained hardware — from microcontrollers to mobile NPUs operating at sub-millisecond latency.
Focus areas
Vision-language-action policies, perception-action loops and human-robot collaboration for mobile manipulation in unstructured environments.
Focus areas
Open datasets
Curated, documented datasets released under permissive licenses for the global research community.
Open-source releases
Production-grade libraries, training pipelines and pretrained model weights on our public repository.
Funded by
Government of India research missions and industry consortia with combined funding of ₹8.4 Cr+.