← All work

01Music Recommendation Engine2026

SonicStream

SonicStream — live product screenshot

Captured live from karthikeya2536.github.io

Overview

SonicStream is an AI-powered music platform built for audiophiles. Its recommendation engine combines Content-Based Filtering with Neural Collaborative Filtering, trained in PyTorch as a hybrid neural network over the user–song–artist graph. A FastAPI backend serves the model, Firestore stores listening data, and the whole system ships fully containerized.

Highlights

  • Hybrid recommendation engine: Content-Based Filtering + Neural Collaborative Filtering
  • PyTorch Hybrid Neural Network (HNN) and Matrix Factorization models over graph embeddings
  • FastAPI backend with Pydantic-validated REST endpoints and Google Firestore
  • React + Vite frontend with Tailwind and Framer Motion, fully Dockerized

Stack

  • PyTorch
  • FastAPI
  • Firestore
  • Docker
  • React
  • TypeScript

Next project

EduPredict