Perit is a YC-backed company that builds realistic, hard tasks used by frontier AI labs to train and test their models. We're looking for experienced data scientists to build and review data science tasks on a paid contract basis, with a path to a full time role.
What you will do
- Review data science tasks end to end: the business brief, messy datasets, the reference solution and the automated grader
- Judge whether a task is realistic, clearly specified and fair to a strong analyst
- Find the traps: leakage, wrong labels, unstated rules, graders that reject valid answers or accept wrong ones
- Run the solution and grader in Docker, test alternative and deliberately wrong approaches, and write up exact fixes
- Over time, design new tasks of your own in forecasting, risk modelling, causal inference, reconciliation and SQL
Must have
- Hands-on experience creating or reviewing tasks for AI training or evaluation, for example RL environments,
Terminal Bench style tasks, SWE or coding tasks, or rubric-based evals for Mercor, Micro1, Turing, Handshake, Outlier, Scale or similar
- 3+ years of real data science work: statistics, validation, calibration, time series or causal methods, plus strong Python and SQL
- Comfortable running code in containers and reading someone else's codebase
- Clear written English. Your write-ups are the product.
Nice to have
- Experience with Harbor, Terminal Bench or similar task frameworks
- Domain depth in finance, insurance, healthcare, utilities or supply chain
Details
- Remote, flexible hours, about 10 to 20 hours a week to start
- Paid per task or hourly, based on experience
- Strong contributors get offered a full-time role
📌 Data Scientist (India)
🏢 Zyka AI
📍 India