This research explores how technology-mediated role plays improve students’ English-speaking skills, addressing the limited interactive speaking practice in their engineering education by optimising Google Gemini to stimulate real-life role-play scenarios, and thus contributing to innovative digital language learning approaches. Considering the dearth of interactive speaking practice, this research was conducted to enhance the speaking skills of engineering students through role-plays, leveraging Gemini 2.5 Flash, given the increasing importance of speaking skills in both academia and industry in engineering. It was an interventional, longitudinal and quantitative classroom study conducted over 21 weeks, following the Pretest-Posttest Control Group Design. In this study, 36 first-year engineering students, all aged 18, were selected as participants using convenience sampling. Then, the researchers administered a pretest, consisting of role plays in a conventional mode, to form the research groups: Interventional Group (IG) and Non-interventional Group (NG), based on the pretest scores using the speaking assessment rubric of the Common European Framework of Reference for Languages (CEFR). After the bifurcation of groups, the IG utilised Gemini 2.5 Flash to practise their role plays with the Artificial Intelligence partners and have their role plays assessed according to the CEFR-aligned assessment, along with their practice of role plays in a conventional mode, on par with the NG, which was constrained only to conventional role-play practice. Thus, the IG practised their speaking in a blended learning framework, unlike the NG. Post-intervention, the posttest findings were statistically evaluated and indicated a significant improvement in the IG, as the Test Statistic (i.e., Calculated t = 24.05) exceeded the tabulated t (1.734) at the 0.05 level of significance with n-1 degrees of freedom, supporting the effectiveness of the intervention.