MLE-bench

RD-Agent is an open source AI framework designed to automate research and development workflows in data-driven domains. It uses large language models and multiple collaborating agents to simulate the typical cycle of research, experimentation, and improvement that human data scientists follow. It separates the process into two core phases: a research stage that proposes hypotheses and ideas, and a development stage that implements and evaluates them through code execution and experiments. By iterating through these stages, the framework continuously refines models and strategies using feedback from previous results. RD-Agent focuses heavily on automating complex tasks such as feature engineering, model design, and experimentation, which are traditionally time-consuming in machine learning and quantitative research workflows. RD-Agent can analyze data, generate experimental code, run evaluations, and learn from outcomes to improve future iterations.