The reading ability of 2nd grade students of SDN 2 Ketandan is still relatively low, characterized by the habit of spelling letter by letter, the inability to understand the meaning of reading, and the dominance of conventional learning methods that are mechanistic and less student-centered. This condition encourages the need to implement a more innovative learning approach, one of which is the Deep Learning approach which has been echoed since the inauguration of the Minister of Primary and Secondary Education in the 2024-2029 Merah Putih Cabinet. This study examines the implementation of the Deep Learning approach in reading learning in Indonesian language subjects in grade 2 SDN 2 Ketandan for the 2025/2026 Academic Year. The theoretical studies used include the concept of Deep Learning as a pedagogical approach with three main principles of mindful learning, meaningful learning, and joyful learning based on Vygotsky's theory of constructivism, as well as the theory of initial reading ability including indicators of pronunciation, fluency, intonation, and expression. This study uses a descriptive qualitative approach. Data collection was carried out through structured observation, in-depth interviews with 2nd grade educators and six students as representatives of 22 students, poetry reading ability tests, and documentation. The validity of the data was tested using triangulation of sources and techniques, while data analysis referred to the Miles and Huberman model through the stages of data reduction, data presentation, and conclusion drawn. The results show that the implementation of Deep Learning has been carried out systematically through four learning phases: planning, exploration, application, and reflection. The three principles of Deep Learning are integrated in each phase, making the learning process more active, meaningful, and fun. The application of this approach has been proven to be able to improve students' initial reading skills, especially in reading poetry, by encouraging critical thinking involvement, collaboration, and contextual understanding. Thus, the Deep Learning approach is effectively applied to reading learning in elementary elementary grades, although it still requires further development for more optimal outcomes.