About the role
What we do
We build SOTA voice models for regional languages — Urdu, Bengali, Pashto, Greek, and more — so people around the world can use technology by simply speaking.
We gather our own data, build our own labeling tools, train our own models, and design our own voice infrastructure. This fully integrated approach is what lets us deliver the highest-quality models, and the best developer experience, for regional languages.
Why our work matters
Over 60% of the world's population only speaks a regional language. Today voice models aren’t good enough in their languages.
Large tech companies are unable to give this work tier1 priority because they are busy defending their primary business. Infact, in all big labs regional languages are funded by Marketing/PR budgets.
If UpliftAI doesn’t do its work, a large part of the world may continue to experience significant AI technical lag — further worsening the income divide between countries.
The story of how we started
Hammad left Apple and took a year off to figure out what to do in the next phase of his career. During this time, he started a side project gathering data for Pakistani languages so people could talk to ChatGPT by speaking (this was before they supported voice mode).
To gather this data, he traveled to factories and rural areas in Pakistan. While speaking with business owners there, he learned that many businesses are not using digital tools (like HR software or inventory management) simply because their workforce cannot read or write. (Btw, Pakistan is the 5th most populous country in the world, but 42% of all adults there cannot read).
That is when the idea first emerged: what if everyone could use technology by speaking? Voice interfaces can finally digitizing businesses that currently cannot. This would increase worker productivity — which at scale lifts the entire GDP of developing countries. Globally, 1 billion people cannot read, and voice interfaces will finally give them access to digital services… often for the first time.
