maya research

We are building AI for India — and we're open-sourcing

So far we've shipped

Veena

Open-source text-to-speech model for Indian languages.

Demo text: "अगर ऑर्डर डिलीवर नहीं होता है, तो आपका पूरा पैसा वापस कर दिया जाएगा…"

0:00 / 0:00

Quality on par with ElevenLabs — and fully open-source.

See full TTS model comparison →

Indic Gemma

13B text-to-text model trained for conversational Indian languages. Check it out →

Indic Gemma preview

IndicVault

High-quality, instruction-tuned dataset featuring question-answer pairs crafted in the contemporary, everyday language spoken across India in 2025.

IndicVault dataset

Download Dataset

Built by Dheemanth Reddy  |  Bharath Kumar