The Role of Synthetic Data in Multilingual, Multicultural AI Systems: Lessons from Indic Languages
Published in ACL 2026, 2026
Developed UPDESH, a 9.5 million-sample culturally grounded synthetic instruction dataset across 13 Indic languages to narrow performance gaps in multilingual AI. Accepted at ACL 2026 (Main track). Preprint: arXiv:2509.21294.
