Legal Expert - AI Training & Evaluation

About the role

Advanced AI models are being pushed into legal work — research, contract review, compliance. Whether they're actually any good at it depends on the quality of the people who train and test them. That's you.

You'll create complex legal tasks that a model should be able to handle but often can't, and you'll evaluate what the model produces — where it's right, where it's subtly wrong, and where it's confidently making things up. The work rewards precision. A vague task or a lazy evaluation is worse than none at all.

This is not paralegal work and it's not data entry. Each task is a piece of real legal thinking — the kind of question a senior associate would be asked and have to get right. If you want templated, repetitive work, this isn't for you. If you like taking a hard legal problem apart and explaining exactly why an answer is wrong, you'll probably enjoy it.

What you'll actually do

On any given task you might be:

• Designing a complex legal problem — a research question, a contract to analyse, a compliance scenario — with a clear, defensible model answer
• Reviewing AI-generated legal responses line by line and grading them on accuracy, reasoning, and completeness
• Catching hallucinated case law, misread clauses, missed exceptions, and reasoning that sounds right but isn't
• Writing clear rationales for your evaluations so the model (and the team) learns from them
• Working across legal research, contract analysis, regulatory compliance, and legal reasoning — not just one niche

Every task is reviewed. You'll be measured on the quality and rigour of what you submit, not on volume.

What we're looking for

• 3+ years of relevant legal experience — in practice, in-house, or at a firm
• Strong legal research and analytical skills. You can trace a question to the right authority and explain why it applies
• Comfort with contracts, compliance, and reasoning across more than one area of law
• Precision in writing. You say exactly what's wrong and why, without padding
• Low tolerance for plausible-sounding nonsense — from a model or anyone else
• Reliable on deadlines. Task-based work only works if the tasks come back on time
• Curiosity about how AI is going to change legal work, and a preference for shaping it over watching it happen

What you get

• ₹20,000–25,000 per task — paid for depth of thinking, not hours logged
• Remote, flexible work you can fit around a practice or a full-time role
• A front-row seat to how frontier AI models are trained and evaluated on legal work, and a hand in making them better
• A YC-backed team that moves fast and keeps things direct
• More tasks and larger projects if your work is consistently strong