<![CDATA[In the Merdeka Curriculum framework, vocational English instruction must fundamentally transition from generalized linguistic competence toward highly contextualized English for Specific Purposes (ESP) tailored directly to the dynamics of the modern digital economy. However, empirical observations indicate that students enrolled in vocational beauty programs frequently exhibit low speaking performance, high baseline performance anxiety, and descriptive incoherence regarding product specifications. This study aims to describe the structural operationalization of Deep Learning-based Project-Based Learning (PjBL) utilizing YouTube and to investigate the resulting nature of students’ speaking performance after its execution. This study employed a Classroom Action Research (CAR) design using a mixed-methods approach involving 30 tenth-grade students of Kelas X Kecantikan 2 at SMKN 3 Kediri. Qualitative and quantitative data were gathered through observation checklists, field notes, semi-structured post-cycle interviews, and oral speaking tests evaluated via a contextualized rubric adapted from Brown (2004). The findings revealed that the integrated PjBL framework incorporating scriptwriting as a critical cognitive scaffolding technique effectively mitigated foreign language speaking anxiety and accelerated professional vocabulary acquisition. Quantitative oral scores demonstrated robust upward mobility: the class mean score progressed from 53.2 in the initial cycle to 79.5 in the subsequent cycle, successfully surpassing the established minimum passing criterion (KKM) of 75. The systematic integration of digital pedagogy through YouTube product review vlogs provides an authentic, immersive instructional environment that successfully fosters 21st-century competencies and improves vocational workforce readiness.]]>
