University speaking courses require learning designs that provide repeated oral production, self-monitoring, and usable feedback. This study developed and preliminarily evaluated an interactive podcast-based learning model for third-semester students in the Indonesian Language Education Study Program at Universitas Bengkulu. A development-oriented mixed-methods design (QUAL → quan) was embedded in the ADDIE stages. Formal needs assessment involved three classroom observations, interviews with 15 students, two lecturers, and one educational technology specialist, and document analysis. One intact class of 30 students participated in five podcast cycles across ten meetings. Expert appraisal increased from 85.0% to 88.5% of the maximum possible score after revision, while students reported high practicality (M = 4.18, SD = 0.46); the two lecturers provided a combined descriptive mean of 4.38. The three-component speaking-performance score increased from 63.33 (SD = 8.10) to 81.00 (SD = 6.70), t(29) = 12.12, p < .001, dz = 2.21. Observable speaking assurance, reported separately from linguistic performance, also increased. Because the study used one intact class without a comparison group and did not systematically code feedback quality or individual change trajectories, the findings indicate feasibility and preliminary improvement rather than causal effectiveness. The study contributes a transferable practice–replay–feedback–revision sequence integrating culturally responsive task criteria and privacy-conscious implementation supports.