Long Horizon
மெய்ப்பாட்டு

MEIPPATTU

Foundation motion model for Indian classical dance on NVIDIA MotionBricks. Bharatanatyam first, because the tradition deserves to be encoded properly.

Why this exists

Bharatanatyam is one of the world's most rigorously codified classical art forms.

Every gesture has a precise jati, theermanam, and angular grammar. The Natya Shastra documents the system from the 2nd century BCE. Yet no foundation motion model treats this seriously.

The bottleneck isn't compute or money - it's institutional access. The knowledge lives with teachers, in lineages, in institutions like Kalakshetra. MEIPPATTU is the program that builds the model and the institutional partnerships needed to do this right.

Pipeline
I

BHARATA-ADAVU Dataset.

Motion capture protocol for foundational Bharatanatyam vocabulary. Multi-performer corpus across the canonical adavus, captured with guru oversight to preserve traditional grammar.

II

VQ-VAE Tokenization.

Motion fine-tuning on NVIDIA MotionBricks with custom dance-vocabulary tokens for discrete motion generation aligned to traditional grammar.

III

Music Conditioning.

Custom encoder maps Carnatic music structure - raaga, tala, jati - to motion sequences, generating choreography from audio.

Partnerships

The data problem here isn't volume; it's institutional access.

The knowledge is held by living institutions and lineages - Kalakshetra Foundation in Chennai, Darpana Academy in Ahmedabad, Sri Krishna Gana Sabha in Chennai. MEIPPATTU is framed as preservation and pedagogy, not automation. A serious research partnership, not extraction.

NVIDIA stack
Status
Applications