Writing-to-Learn (WTL) is a practice that involves writing assignments specifically formulated to foster conceptual understanding and disciplinary thinking. When properly implemented, WTL has been shown to not only foster conceptual learning, but also support development of scientific reasoning, promote metacognition, and elicit student thinking. While the positive impact of WTL is well-established, systematic barriers inhibit its implementation at a large scale. The most effective WTL assignments are meaning-making tasks with clear writing expectations that include a metacognitive and interactive writing process. At the University of Michigan, we designed and executed a program called MWrite that makes implementing WTL accessible to instructors teaching large-enrollment courses. We provide instructors with support to design effective WTL assignments and further support their implementation of the assignments through specialized undergraduate teaching assistants known as writing fellows and an interactive, automated peer review process for students. For the past 7 years, we supported WTL in 100+ sections of 15+ courses, impacting over 23,000 students submitting over 53,000 assignments and producing a database of over 800,000 peer review comments. The research team has studied the impact of the program in several courses and demonstrated that the assignments support the intended learning goals and that the design elements of the assignments function as intended (e.g., students can provide constructive, content focused feedback to their peers). Furthermore we have analyzed students’ writing and peer review using machine learning and explored the utility of machine learning for providing feedback to both students and instructors. We will discuss how MWrite is structured and the associated research aimed at understanding whether and how the MWrite design is effective and ways in which it could be improved.