Building the Learning Loop Around an LLM Language Tutor

Instead of giving learners a generic chatbot, Pangea guides them through small conversational tasks, tracks their learning profile, and adapts the session around their target language, level, goals, and context. For the live demo, I’ll show the working React/Django app: onboarding, language/profile setup, starting an AI conversation, mission-style practice inside a session, and the backend architecture that stores users, conversations, messages, languages, and learning state, all context for the personal AI tutoring.

I’ll also walk through the parts that are still messy: prompt design, conversation state, how to make the AI feel useful without over-controlling the learner, and how I’m thinking about evaluating whether a “completed” mission was actually natural and useful for language retention.
What will another builder learn? (Required)
Share the hard-won lesson: what broke, what surprised you, what tradeoff mattered, or what technique another builder can reuse or avoid.