Founding Research Engineer - Audio Intelligence

2 нед. назад

80k–180k USD / yearUSASeniorRemote

machine learningaudio processingresearchpython

Lead applied research and development in audio intelligence for regional languages, progressing projects from experimentation to production.

Условия

A mission worthy of being your life’s work.
True ownership of work.
Top-tier teammates that are ambitious, smart and fun.
Top percentile equity and pay that truly value your talent and opportunity cost.

Другое

You will set Uplift’s applied research agenda for bringing advanced audio intelligence to regional languages, owning each problem from the first experiment through production.
You will scale Uplift’s R&D capacity by building partnerships with leading university labs and turning promising academic research into production systems.
The problems you tackle will come directly from customers, and success will be measured by real-world impact—not just benchmark scores.
We build voice models and voice applications 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.
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. In fact, 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.
But also, it’s very cool making machine speak soooo well (check out our homepage).
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.
Proof of exceptional talent through past accomplishments. Ask yourself: “What have I done that none of my peers did?”
Demonstrated ability to advance the state of AI, even in a small way. This could be through research, open-source work, or a model or system you built.
Strong research and engineering ability. You can read papers, design rigorous experiments, train models, and make them fast and reliable enough for production.
The ability to move 3× faster than your peers through creative problem-solving.
Hammad - CEO: ex Apple, ex Amazon, Cornell alumnus. Previously developed voice technology at Apple Siri and Amazon Alexa for about 9 years. I created my first website at age 13 to sell wallpapers & software. This was year 2003 before even wordpress existed. It was hosted over dialup from home. Zaid - CTO: Zaid previously designed and launched an entire AWS service: Bedrock Guardrails. Before that, at Amazon, Zaid wrote a paper on how to optimally plan and pack products in containers to minimize transportation and destination distribution cost in Amazon’s global import supply chain. This work led to $100m/year saving.
Below is one possible plan. We expect you to improve it once you understand the problems.
First 30 days: Study real customer calls and identify the most valuable audio problems that existing models do not solve. Choose one, build the eval set, and establish the strongest available baseline.
First 60 days: Develop a solution that clearly beats the baseline and test it with a real customer. In parallel, launch a research collaboration with a university lab on another important problem.
First 90 days: Ship your first new audio capability into production and demonstrate that it improves a real customer outcome—not merely a benchmark. Have at least one university lab actively expanding UpliftAI’s research capacity.