The rapid proliferation of large language models (LLMs) has opened new horizons for personalized, AI‑augmented learning. The ability to design new kinds of educational interactions can provide language practice opportunities that previously did not exist. For example, the language of persuasion, AI chatbots offer the unique ability to craft interactions with a high degree of authenticity related to the social purpose of the interaction. Enter Luigi: a chatbot pizza lover who is vehemently opposed to the addition of pineapple on pizza. Your mission, should you choose to accept it, is to convince Luigi to at least try tasting pineapple on pizza (Hawaiian pizza). Students interacting with Luigi are provided with an interlocutory adversary who will not immediately acquiesce to the student's request, but rather will evenly provide opportunities to entire classes of students to use persuasive language. In this presentation, I will demonstrate how to achieve this, hosting LLMs locally using Ollama (no cloud services and free of cost), as well as how to download (pull) models and customize them to create unique learning interactions for educational use. I will also describe the difficulties encountered getting models to produce language that is appropriate for the task, level, and situation.