Madrasati Muse Spark

Madrasati Muse is a voice-first Egyptian Arabic tutor for primary-school learners. It helps a child ask questions naturally and receive a lesson shaped around their grade, teacher preferences, and textbook. The learner’s PDF is processed on-device: the app extracts text, runs Arabic OCR on scanned pages, and builds a local page index. When a learner asks a question, the app retrieves a few relevant passages and sends those excerpts—not the original PDF—to Meta Muse Spark. Muse Spark leads the lesson with an explanation, a worked example, and a check-for-understanding question, then uses recent answers to choose the next teaching step. Muse Voice Transcribe streams the learner’s speech, shows partial transcription, and detects the end of a turn. Google Gemini 3.8 Flash TTS generates the Egyptian Arabic spoken response. It is a separate speech service; Meta models provide live transcription and teaching. Disclosure: Madrasati’s original product concept existed before the hackathon. Madrasati Muse is a separate implementation focused on the Meta models and voice-first lesson flow. The submission should identify the Muse-specific work accurately.