Weaver and spinner · Kehar Singh, c. 1860 · Public domain

Karya

Paid Indian-language data work, done on a phone

India angle

Pays people in villages, found through local non-profits, to record and transcribe speech in their own languages on a smartphone.

Bengaluru · Live

What it does

Karya is an Indian organisation, based in Bengaluru, that sells Indian-language speech and text data, and evaluations of AI models, to organisations that include companies and the Government of India. People in villages make the data on a smartphone app, by reading sentences aloud and transcribing or subtitling, and are paid for the work.

Languages with little recorded or written data get weaker AI tools. Karya's case is that paying the people who speak those languages is a way to fill that gap and put income in their hands. Forbes India reported in 2025 that it pays at least $5 an hour. That figure is the company's own, and we found no independent audit of what workers are paid.

Microsoft says Karya began in 2017 as a project at Microsoft Research in Bengaluru and became independent in 2021. Forbes India says Manu Chopra started it in 2021 with Safiya Husain and Vivek Seshadri. TIME and Microsoft both put its non-profit partners, who help it find workers, at more than 200. Forbes India reported that it works with over 100,000 people. Its website shows the Government of India, Microsoft, Anthropic, Google, the Gates Foundation and OpenAI under "Trusted by".

This listing is compiled from public sources. We have not used the app, and Karya has not reviewed it.

Built by
Manu Chopra, Safiya Husain, Vivek Seshadri
Based in
Bengaluru
Made for
Companies and government bodies that need Indian-language data; workers in villages do the tasks

Compiled by the Takṣa team from public information, last updated 7 October 2026. Sources: Karya's website, TIME on Karya, 27 July 2023, Microsoft Source Asia on Karya, 7 February 2024, Forbes India on Karya, 16 June 2025, NITI Aayog Frontier Tech profile (figures submitted by Karya).