Multilingual Multimodal Indic Resources for AI
MIRA
mira@iitjammu.ac.in

Let us build MIRA

A sovereign model factory for frontier AI systems — building India’s sovereign AI capability from data to deployment.

Sovereign AI

End-to-end ownership of datasets, models, architectures, and deployment pipelines.

Multilingual + Multimodal

Unified systems spanning text, speech, and vision — evolving toward omni-modal intelligence.

Efficient + Scalable

Compute-efficient architectures aligned with real-world constraints and deployment needs.

MIRA Stack
  • Curated multilingual and multimodal datasets
  • Frontier-scale foundation models
  • Reusable libraries and toolkits
  • Scalable training and inference platforms
Dual-Track Execution

Engineering Track

Production systems, deployment excellence, reliability, maintainability, cost-efficient inference.

Applied Research Track

Frontier architectures, efficient training paradigms, low-precision systems, research-to-deployment transition.

Translational Domains

Security and Governance

Agriculture

Commerce

Education

Healthcare

Robotics and Science

MIRA Collaborations
Brochure
Hot From The Press

SOTA in 1-bit quantization

We have developed an ultra-low bit quantization technique that achieves state-of-the-art performance on a variety of tasks. This technique is a major breakthrough in quantization and has the potential to revolutionize how we train and deploy models.

To appear in Conference on Parsimony and Learning, Tubingen, 23-27th March 2026. CPAL'26.

Explore the coset library or Read More about the work