Freelance AI Trainer: Clinical EHR Vignette Evaluation (US Physicians)

About the role

Mindrift connects specialists with project-based AI work for leading tech companies, focused on testing, evaluating and improving AI systems. It is looking for US-based, actively practicing physicians (MD/DO) to evaluate clinical EHR vignettes for a medical AI evaluation program, with general and internal medicine as the main focus. The work is project-based and part-time.

Key responsibilities

• Evaluate clinical EHR vignettes paired with a question, a proposed answer and a distractor "trap" answer across diagnosis and treatment tasks covering cardiovascular, nervous, hematologic, respiratory, digestive, urinary, reproductive, musculoskeletal and integumentary systems.
• Score clinical reasoning quality: vignette accuracy and completeness, whether the vignette gives the answer away, answer correctness and gradeability, and trap quality.
• Check whether the reasoning chain reaches the answer from vignette facts alone, with correct reasoning traces, and write short rationales.
• Work within three independent blind reads, followed by physician adjudication.

Required profile

• Medical degree (MD or DO) and an active, unrestricted US medical license verified with the issuing medical board.
• Board certification or completed residency training, with recent direct patient care (attending-level preferred).
• Broad, multi-system diagnostic and treatment experience (internal, family, emergency or hospital medicine).
• Strong clinical reasoning: differential diagnosis, next-step management and application of evidence-based guidelines.
• Prior experience in medical AI evaluation, clinical content or exam-question review, or medical annotation/QA is a plus.
• Strong written English (C1+).
• Not suitable for medical students, unlicensed graduates, NPs, PAs, nurses, pharmacists or other non-clinical healthcare titles.

What we offer

• Payment per accepted task, depending on qualification tier and efficiency, up to the equivalent of $100 per hour.
• Estimated workload of around 10-20 hours per week during active phases, with tasks due by the deadline.