Mabyduck is a self-serve platform for running subjective quality evaluation studies on AI-generated audio, video, images, and interactive experiences. It is aimed at AI researchers, academics, startups, and enterprises, offering configurable pairwise comparisons, surveys, rubrics, leaderboards, rater pools, and an API for integrating evaluations into existing pipelines. Key capabilities include RLHF data collection, regression tests, embedded experiments, and top-K selection strategies for pairwise experiments.