Gen AI music learning platform (internal R&D)

The challenge

Music education has long been plagued by its one-size-fits-all approach. Adult beginners with no prior training find music theory intimidating, and traditional online platforms fail to adapt to individual learning pace or answer nuanced questions without a human tutor in the background. To investigate the future of personalized education, JustSoftLab's Gen AI team launched Lyric Aria as an internal R&D project — aiming to create a fully autonomous music tutor that pushes the boundaries of agent-based AI. The main challenge: building an end-to-end system on Google Cloud Platform that could generate curriculum from scratch, interact intelligently with learners in real time, and adapt difficulty based on performance.

Our solution

We designed and built Lyric Aria entirely on GCP, with two primary flows: automated course generation from uploaded content and interactive consultation with a tool-augmented AI agent. Instructional materials (PDFs, documents) are uploaded into Cloud Storage, processed by Document AI for text extraction, segmented into semantic chunks, and vectorized using Vertex AI Embeddings for storage in Vector Search. Gemini 2.0 Pro assembles structured courses — generating lessons, quizzes, and visual assets via Imagen3. A custom RAG pipeline enables an AI agent to answer music theory questions with cited sources in real time. The agent adapts quiz difficulty based on learner performance, creating a truly personalized learning experience. The entire stack runs serverless on Cloud Run with Firestore for state management.

Key highlights

Tech stack