This research paper describes an innovative approach for the rapid creation of language proficiency assessments, specifically detailing the development of the Duolingo English Test. The authors utilize machine learning (ML) and natural language processing (NLP) to automatically determine the difficulty of test items, bypassing the traditional, expensive process of pilot testing with human subjects. By aligning item difficulties with the Common European Framework of Reference (CEFR), the developed computer-adaptive test (CAT) is shown to produce valid, reliable, and secure scores that correlate significantly with other high-stakes English assessments like the TOEFL iBT and IELTS. Crucially, this ML/NLP-driven method allows for the creation of an unprecedentedly large item bank, which dramatically enhances test security by minimizing item exposure.