
Voice Language Partner turns language practice into a real conversation instead of a vocabulary quiz. Learners pick a target language, proficiency level, and practice mode — free conversation, real-world scenarios, tutoring, or vocabulary drills — then simply speak out loud. AssemblyAI Voice Agents provide the real-time listening and spoken replies, so there's no typing. Each session is configured on the fly for the learner's language, level, and mode, and every turn is saved as a transcript in PostgreSQL through a FastAPI backend with JWT authentication. When the session ends, Gemini evaluates the full transcript across grammar, vocabulary, fluency, comprehension, and target-language usage. The report shows scores with visual bars, detailed feedback, a mistakes-to-review list with explanations, and a live "Coach says" recommendation generated for that exact conversation. The key idea is a complete learning loop: speak → see what went wrong → get your next step. Instead of generic lessons, feedback adapts to what the learner actually said — so a beginner ordering food in a restaurant scenario gets different coaching than an advanced learner debating hobbies. The dashboard tracks progress per language: average score, best score, streak, full session history, and recent mistakes. It works on desktop and mobile. Built with: FastAPI, PostgreSQL, AssemblyAI Voice Agents API, Gemini API, and a vanilla JS frontend. Deployed live on Render.
30 Sep 2026