Language
One model, multiple languages
Most Indians speak more than one language, often in the same sentence. We are researching models that move between Hindi, Tamil, Bengali and English the way people do, without translating through English first.
Anuvaya Labs is an AI research lab in New Delhi. We build AI models using sovereign data to excel at solving India-specific problems.
1947
Start. A new country.
1963
Hydropower and irrigation for the north.
1966
High-yield wheat doubles output.
1970
Cold chains link village dairies to cities.
1980
SLV-3 puts Rohini into orbit.
1984
Indian-built digital telephone exchanges.
1995
VSNL opens the public internet.
2010
The first Aadhaar number is issued.
2016
A real-time payments network.
2023
First soft landing near the lunar south pole.
Now
Indigenous AI models that understand Bharat.
We conduct research studies with expert practitioners who understand Indian problems and their nuances.
So our models understand
So our models understand
So our models understand
We use traditional textbooks that capture the cultural, traditional and regional diversity of our country alongside feeding local context that makes our country feel like home.
Medicine
Sushruta Samhita
Palm leaf, Nepal, 12th–13th century
Everyday
The chai stall
A street stall in Varanasi, kettles on every shelf
Mythology
Dashavatara
The ten avatars of Vishnu, Matsya to Kalki. Painted c. 1775
Market
The sabzi mandi
Greens sold by the bunch, prices by the morning
Jyotish
Brihat Jataka of Varahamihira
Palm leaf, Kathmandu, 1399
Healthcare
Primary Health Centre
Sasapasi, Odisha, where a village first sees a doctor
Law
Constitution of India
The Preamble, illuminated by Beohar Rammanohar Sinha, 1950
Linguistics
One sign, three scripts
Odia, Hindi and English at Bhubaneswar station
Pharmacy
The neighbourhood dispensary
Shelves of remedies in Varanasi, fetched by hand
Poetry
Gita Govinda of Jayadeva
Illustrated palm leaf, Odisha, early 18th century
History
Akbarnama
Mughal painting by Manohar, c. 1604
Language
Most Indians speak more than one language, often in the same sentence. We are researching models that move between Hindi, Tamil, Bengali and English the way people do, without translating through English first.
Speech
We feel many Indians will talk to AI before they chat. We study ASR in different accents, noise and code-mixing.
Evaluation
We want to build evaluations for language models in Hindi and other regional languages.
Efficiency
A sentence in Hindi or Tamil can cost several times the tokens of the same sentence in English. We research tokenizers that close the gap.
Reasoning
Doctors, lawyers and jyotishis each work from their own texts and rules. We study how they reach an answer.
Yes. Every participant is compensated, based on their area of expertise and experience. We pay 30% in advance once your participation is confirmed, and the rest when the study is complete.
The study is a remote, two-day conversation. You will answer simple questions from your domain that mimic a consultation, and we will audio record your responses.
Yes. With your consent, we credit you by name for your contribution. Every participant also receives a certificate of contribution from Anuvaya Labs.
Your data will not be used directly to train any AI model. Our research team studies your responses to understand how experts in your field reason and speak with the people they help. Recordings are stored securely, seen only by our research team, and never sold or shared outside Anuvaya Labs.
Practising professionals in medicine, law, Vedic astrology and psychology who regularly consult people in India. Tell us about your practice when you apply.