Personalized search engine for music lovers

Ever searched for "energetic workout songs" or "rainy day music" and received generic results that don't quite match your vibe? This search engine solves that problem. It gets to know you by building a dynamic musical profile based on your listening habits—what you search for, which songs you click on, and what you play.

Instead of just re-ranking a standard set of results, our engine intelligently blends your musical profile with your search query before the search even begins. This creates a unique, personalized query that reflects both your immediate intent and your long-term preferences. The result is a more relevant and satisfying music discovery experience, surfacing tracks that are not only a good match for your query but also a great match for you.

Key Features 🎵

• Spotify Integration: Log in with your Spotify account to stream music directly through our interface.
• Dynamic User Profiles: Your profile is built from your interactions (searches, clicks, and playbacks) and evolves over time.
• Hybrid Search: We combine the power of semantic search with traditional keyword search to ensure both relevance and precision. This is all fused together using Reciprocal Rank Fusion.
• Deep Personalization: The core of our engine is the fusion of your user profile embedding with your query embedding, creating a search that is truly tailored to your tastes.
• User Analytics Dashboard: Visualize your listening habits, including your top genres, artists, and liked songs over time.

Setup

First, you need to setup the local elastic engine

curl -fsSL https://elastic.co/start-local | sh

cp elastic-start-local/.env .

sh elastic-start-local/start.sh

Endpoints

After running the script:

• Elasticsearch will be running at http://localhost:9200
• Kibana will be running at http://localhost:5601

more info

• initial setup

• elastic python documentation

To run the web app

install uv if you do not have it

brew install uv

use uv as package manager:

uv venv
source .venv/bin/activate
uv add -r requirements.txt
uv run app.py

or simply:

uv run app.py

then you can see the GUI locally on http://127.0.0.1:5000

To get the corpus

mkdir corpus
curl -L "https://zenodo.org/records/5603369/files/wasabi-2-0.tar?download=1" -o corpus/wasabi-2-0.tar
cd corpus
tar -xf wasabi-2-0.tar
cd json
unzip json.zip -d ../
cd ..
rm -rf json rdf wasabi-2-0.tar
cd ..

remove any other files in the corpus folder except for album.json, artist-without-members.json, and song.json

Start indexing

uv run src/indexing.py

To calculate the embeddings

uv run src/embedding.py -i SONG_JSON_DIR -o OUTPUT_DIR

then you need to replace the original song.json with the output json file, and re-index to load these embeddings on Elastic Search you can also specify the batch size with -b BATCH_SIZE and which field for embedding with -f FIELD_NAME