As large language models (LLMs) expand their impact across various disciplines and industries, they have also found their way into translation. LLMs are now being utilized in multiple facets of translation practice and education, including the optimization of machine translation (MT), translation quality assessment, and translator training. While some view LLMs as potential replacements for traditional MT systems, existing research remains fragmented, predominantly addressing isolated aspects of the translation process. Critically, there is a corresponding gap in pedagogical guidance on how to holistically integrate these technologies into translator training. To address this insufficiency, our work presents a comprehensive approach to a case of game localization that leverages LLMs throughout the entire translation process, and explores its direct implications for translation education. In our work, we re-envision the traditional Computer-aided Translation (CAT) workflow for game localization through several key stages. We employ GPT-4o for automated terminology extraction, use OpenAI embedding models for semantic retrieval of translation memories, and apply GPT-4o for context-aware translation generation via structured prompts. Finally, we utilize COMET for quality assessment, with GPT-4o refining segments that fall below a set threshold. Beyond presenting a technical blueprint, this chapter critically examines the pedagogical value of this workflow. It illustrates how educators can structure project-based learning (PBL) activities around these stages to cultivate essential next-generation skills, such as prompt engineering, semantic resource management, and AI output validation. As a result, this workflow not only demonstrates a practical and holistic application of LLMs in a professional context but also provides a tangible framework for training future translators to thrive in an AI-driven industry.