Efficient Architectures
Sparse mixture-of-experts, routing strategies, and model designs that maximise capability per parameter. We build models that do more with less compute.
// RESEARCH
Our research spans efficient architectures, data systems, sustainability, and the societal impact of AI. Everything we build starts here.
// RESEARCH AREAS
Sparse mixture-of-experts, routing strategies, and model designs that maximise capability per parameter. We build models that do more with less compute.
Curation pipelines for high-quality multilingual corpora. From web crawls to synthetic generation, we build the datasets that train frontier models.
Sustainable AI research. We measure and minimise the carbon footprint of training runs, optimise power usage, and design for efficiency.
How AI changes knowledge work. We study human-AI collaboration, skill augmentation, and the economic impact of automation in Indian contexts.
Responsible AI from training to deployment. Bias detection, fairness constraints, privacy preservation, and governance frameworks for high-stakes use.
// RECENT WORK
EACL 2027 Main Conference. A diagnostic study of how well language models execute step-by-step procedures.
EMNLP 2026. Measuring how much instructions really matter for supervision.
A 30-hour Hindi–English code-switching ASR benchmark.
Emotion-aware speech recognition for multilingual India.
Open 1.25B multilingual model for Indian languages.
We're hiring researchers across ML, systems, and applied AI.