Clinicians Safety Review for an AI Health Research Assistant

About the tool We're building an AI assistant that runs contextual web searches and synthesises health research so people can learn about their condition or the options available to them. It's an educational tool, it doesn't diagnose, doesn't prescribe, and doesn't tell anyone what to do. What the review is for Reviewers are not asked to validate whether a finding is medically correct. The question is narrower: does the output stay inside the safety rules the generation pipeline is meant to enforce? For example: - Does it avoid giving a specific dose to this user and avoid diagnosing them? - Avoid telling them to start or stop a medication - Avoid asserting one treatment is better than another - Cite its clinical claims, and stay in plain language? Each criterion is written down and worked through against the same review packet the system itself sees. Labels collected here are used to calibrate an automated judge. This is an LLM that will run these same checks at scale once it agrees with reviewers often enough. Consistency across reviewers matters more than throughput. How it works - Log into our labelling app, pick a batch, work through findings one at a time. - Each finding is scored against a short fixed rubric (yes / no / not-applicable, with a short note). - A few minutes per finding. Fully async, pause whenever, progress is saved. - Short onboarding and a calibration set before live labelling begins. Who we're looking for - US based clinical or health-research background: physician, public health, health researcher. - Careful, consistent, comfortable following a written rubric without drifting toward personal preference. - Comfortable reading AI-generated text critically. - English fluency. No coding needed. - Valuable but not required: prior hands-on experience with LLMs (ChatGPT, Claude, Gemini and similar), and especially anyone who has worked at the intersection of health and LLMs: evaluating model outputs, building or reviewing medical AI tools, or thinking about safety and guardrails in clinical contexts. - Google account for authentication only