Bharath.club members plug into the global AI evaluation team. Real work, paid by the hour. Help keep AI useful, safe, and grounded in the lives it's meant to serve.
Most of the world's AI is evaluated in a few cities, by a few thousand people, who don't see most of the world's lives. That gap is where AI fails, fluent in the patterns of a research group, illiterate in the texture of a clinic in Indore or a classroom in Tirupur or a panchayat office in Patna.
India fixes that gap by being the gap. We are the largest, most varied population on earth, and we are the population AI is most rapidly being deployed into. The work of evaluating AI well, for the world, has to involve us. So we may as well organise it.
Eval is the cleanest way an Indian doctor, teacher, lawyer or farmer can shape the AI that's about to land in their job. It pays. It's structured. It matters.
You'll be matched with eval projects that fit your domain and your time. A clinician might red-team a medical reasoning model. A teacher might build classroom-ready evals for tutoring agents. A lawyer might score legal-research outputs against ground truth. An engineer might build the harnesses that run all of it.
The work runs through Eval.qa and a network of global frontier labs and applied teams. Members opt in by track, get paid by the hour or by the milestone, and keep their day jobs.
Eval at scale needs trust. Trust needs people who know each other, who hold each other to standards, who can swap a panel of thirty doctors for a different thirty when the project changes. A loose freelancer marketplace can't do that. A community can.
This is why eval is built into Bharath.club, not bolted on. Members vouch for each other. Senior members mentor newer ones into the work. A track record at Bharath.club becomes a track record in the global eval ecosystem.
Eval work is real work. Members are compensated by the hour or by the milestone. No "exposure" pay. No volunteer-only tracks for paid downstream value.
If you're evaluating a medical model, a doctor sets the rubric. If you're evaluating a tutoring agent, a teacher does. Engineers support. Domain leads.
Every eval project trains the next member into the work. Notes are public. Methods are reusable. We're building a school as much as a workforce.
NDAs are real, but defaults matter. Methods, learnings, and aggregate findings get published whenever the contract allows. The community is the better for it.
We don't do every project. Members can flag and refuse work that conflicts with the manifesto. The community's word is its product.
The people closest to where AI lands should be the ones who decide if it's good enough.
Eval is not just an engineering job. The most useful evals come from people who do the work the AI is trying to help with.
Your seat is the first step. Eval tracks open to you the moment you're in.
Enter the community →