What the work involves
You build prosecution problems from the ground up and then judge how models handle them. A typical task starts with a technical disclosure and a set of prior art references, and ends with a scenario package: the fact pattern, a reference answer, and a written rationale explaining why a particular claim structure, amendment, or argument is the defensible one. Some tasks run the other direction — you receive model-drafted independent and dependent claims, or a proposed §103 response, and you grade them for claim scope, antecedent basis, support in the specification, and whether the technical characterization would actually survive an examiner.
The distinctive skill here is catching output that reads like competent patent English but would fail in practice: claims with no written description support, arguments that concede more than they need to, amendments that create indefiniteness or prosecution history problems, means-plus-function language nobody intended. Labs pay for that discrimination because generic legal-writing review does not produce it.
What the screen looks for
- Your USPTO registration number and the technology areas you actually prosecute in — expect follow-ups that go past the résumé line.
- Concrete prosecution history: office actions you have responded to, rejection types you have argued, claim amendments you have made and why.
- Reasoning you can write down. Reference answers are worthless to a lab if the justification is implicit or reads as "experience tells me."
- Calibration on grading — whether you can separate a claim that is merely inelegant from one that is legally defective.
Logistics
- 100% remote, fully async; no scheduled calls or fixed shifts.
- Minimum commitment is 10 hours per week, with hours you set.
- Paid weekly via Stripe. The $90–125/hr band is as observed on the platform and typically varies with technology area and registration type.
- Ongoing project work rather than a fixed-term engagement, with task volume varying by lab demand.