Founding Engineer - Frontier Tuning Platform

2 нед. назад

80k–180k USD / yearUSASeniorRemote

pythonmachine learningdata labelingmodel training

Build and develop a platform for improving audio models for regional languages enabling production-grade models without ML expertise or significant funding.

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 build the platform that lets people around the world improve audio models for their own languages and use cases. Your work will enable anyone who cares about a language to create production-grade models for it — even without ML expertise or significant funding. We think this is the fastest way to build great audio models for every language.
This is a founding role, so your scope will expand and change as UpliftAI grows.
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?”
The ability to move 3× faster than your peers through extreme resourcefulness.
Enough ML depth to turn data collection, labeling, evals, tuning, and deployment into a workflow that non-experts can use—and to debug it when it fails.
Strong product and full-stack engineering skills. You can build a polished React and TypeScript product—and the backend, data pipelines, and training infrastructure behind it.
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 problem.
First 30 days: Ship the first self-service workflow that lets someone contribute audio, train a model, and evaluate the result for one language and use case.
First 60 days: Work closely with one or two real customers as they use the platform to improve their models. Rebuild the product around what you learn.
First 90 days: Enable an external community or customer to create a production-grade model with minimal help from UpliftAI.