There is a quiet, exhausting reality behind the massive, multi-billion-dollar generative AI boom. To make large language models actually smart, they require millions of hours of human feedback, transcription, and reinforcement learning. But the invisible workforce powering this AI revolution is fundamentally broken. Tech giants rely on a fragmented global gig economy where “ghost workers” are frequently paid pennies to perform grueling data annotation in digital sweatshops. We are building the most advanced technology in human history, yet we are fueling it with an extractive supply chain that guarantees both low-quality data and massive human exploitation.
It takes builders with a profound grasp of both AI infrastructure and human impact to look at this broken supply chain and solve it at the root. That is exactly what Manu Chopra, Vivek Seshadri, and Safiya Husain are doing with Karya.
Founded in 2021 as a spin-out from Microsoft Research, the Bengaluru-based team is building a deeply intelligent, ethical data cooperative that completely rewires how AI models are trained and evaluated across low-resource languages.
Karya deploys an end-to-end data pipeline and evaluation stack — powering massive projects like speech corpora across major Indian languages and multimodal dataset creation at scale. Workers in rural communities use the Karya mobile platform to read sentences, annotate images, and validate AI outputs in their native tongues. But the true paradigm shift is the economic engine: Karya pays workers multiples of the local minimum wage, ensuring dignified digital work, and crucially, gives them ongoing royalty rights for the data they create.
The true moat here is uncompromising data quality driven by an equitable supply chain. Karya recognizes a fundamental truth of the AI era: exploited workers produce garbage data, while invested, fairly-compensated communities produce exceptionally nuanced, high-fidelity intelligence. By completely abstracting away the ethical and qualitative risks of AI training, they deliver unparalleled domain-specific and localized datasets for the world’s most advanced frontier models.
The market? Global AI labs, foundational model builders, and massive enterprises who desperately need high-quality, diverse, and ethically sourced data to train and evaluate the next generation of AI.
Seeing homegrown founders head-down, quietly architecting the infrastructure that ensures the AI economy actually lifts up the people who build it, is profoundly inspiring. They are giving AI models their local context and workers their dignity back.
Let’s celebrate the builders.
✍️ Jay Ingle
#ResponsibleAI #DataLabeling #ProductNation