Focus Areas

Core Research Pillars

Six interconnected domains where AryavartaAI pushes the boundaries of what AI can understand and do.

🧬

Alignment & Safety

Developing methods to ensure AI systems remain aligned with human values, with special focus on diverse cultural contexts across the Indian subcontinent.

Core Priority
🗣️

Multilingual NLP

Deep research into 22+ Indian languages including Sanskrit, Tamil, Bengali, and Hindi — preserving nuance, grammar, and cultural context at scale.

Active
🔭

Reasoning & Logic

Drawing from Nyāya, Mīmāṃsā and other classical Indian schools of logic to build AI capable of sound, multi-step reasoning and debate.

Active
👁️

Multimodal Intelligence

Unifying vision, language, and audio to enable models that understand Indian art, iconography, classical music, and regional cinema.

Active
🤖

Agentic Systems

Building agents that can plan, use tools, and execute complex workflows — grounded in reliable self-monitoring and human oversight.

Emerging
🏛️

Knowledge Preservation

AI-assisted digitization and interpretation of classical texts, manuscripts, and oral traditions from Bharat's vast intellectual heritage.

New

Publications

Latest Research Papers

Peer-reviewed research, technical reports, and preprints from the AryavartaAI research team.

AryaGPT-2: Scaling Multilingual Reasoning Across Indic Languages

Sharma R., Iyer K., Patel V. et al. — We present AryaGPT-2, a frontier language model trained on 2.4T tokens spanning 22 Indian languages, with novel tokenization for agglutinative scripts.

March 2026
Preprint

Nyāya-Informed Evaluation: A Framework for Logical Consistency in LLMs

Krishnamurti A., Das S. — We propose a novel evaluation suite inspired by classical Nyāya pramāṇa theory to assess the logical coherence of large language model outputs.

Jan 2026
Published · NeurIPS

SanskritBERT: Pre-training Language Models on Classical Sanskrit Corpora

Gupta M., Rao P., Singh H. — A domain-specific pre-trained model for Sanskrit achieving state-of-the-art results on grammatical analysis, meter detection, and sandhi splitting.

Oct 2025
Published · ACL

Cultural Alignment: Measuring Value Alignment Across Diverse Socio-cultural Contexts

Mehta J., Verma T. et al. — A benchmark evaluating AI value alignment within Indian cultural, ethical, and social contexts, revealing systematic gaps in existing RLHF approaches.

Aug 2025
Published · ICML

AryaVision: Understanding Indian Iconography and Classical Art at Scale

Pillai S., Bose A. — A vision-language model trained on 18M annotated images from Indian classical art, temple architecture, and manuscript illustrations.

Jun 2025
Published · CVPR

Team

Research Team

World-class researchers united by a shared vision for sovereign, beneficial AI from Bharat.

👩‍🔬

Dr. Ananya Sharma

Head of Research

👨‍💻

Kiran Iyer

NLP Research Lead

👩‍💻

Dr. Priya Rao

Alignment Research

👨‍🔬

Arjun Mehta

Multimodal Systems

👩‍🏫

Prof. Sunita Das

Sanskrit & Classical AI

👨‍🏫

Dr. Rohan Pillai

Reasoning & Logic