Built ground-up for Odia — not translated from Hindi or English. DAKSHINI is Presear's original conversational voice AI agent, engineered from the first principle that India's low-resource languages deserve AI built natively for them, not adapted as an afterthought.
A real conversation, not a scripted demo — DAKSHINI understanding, reasoning, and responding entirely in Odia.
Odia is spoken by over 3.5 crore people — an official language of India and one of only a handful granted Classical Language status. Yet in voice AI, it remains a "low-resource" language: starved of the transcribed speech, parallel text, and annotated conversational data that English and Hindi systems are trained on by default.
Most "regional language" voice assistants aren't built for Odia at all — they silently translate speech to Hindi or English, run the AI there, then translate the response back. Every step loses nuance, mishandles dialects, and breaks on colloquial speech.
DAKSHINI takes the harder path. Speech recognition, language understanding, dialogue reasoning, and voice synthesis are all trained and operated natively in Odia — no translation layer in between.
Every layer of the voice pipeline — listening, understanding, deciding, and speaking — is engineered for Odia from the ground up.
From government helplines to rural clinics, DAKSHINI puts a natural Odia voice in front of services that have long defaulted to Hindi or English.
India has 22 scheduled languages and hundreds of dialects — the overwhelming majority of them ignored by mainstream voice AI, which optimizes for English and, at best, Hindi. Presear's research division exists to close that gap: building foundation voice AI natively for India's low-resource languages instead of retrofitting translation on top of models that were never designed for them.
DAKSHINI is the first agent to come out of this program. It won't be the last.
See DAKSHINI live, or talk to us about deploying original, low-resource-language voice AI for your organization.