We demo Peachnote Piano Tutor, a prototype that listens to a student's playing, aligns the stream to a reference score in real-time, and delivers teacher‑style feedback. While commercial tutors often focus on note accuracy, our system addresses expressive nuance using a multimodal LLM. By concurrently analyzing the performance as it happens, the tutor responds immediately after the student pauses, combining spoken advice with synchronized visual highlights on the score. Although the prototype accepts both microphone audio and MIDI, the live demo employs a MIDI keyboard to guarantee clean note detection in a noisy exhibition space. The lightweight client‑server‑services stack built on WebRTC transport enables this responsive interaction.