Build a Hybrid Recommendation System for an Indie Streaming Catalog
Overview
What this challenge is about.
Use the provided 6-month anonymized event log (around 320M play events, 1.4M unique users in the held-out cohort), audio embeddings (256-d), and track metadata. Implement (1) an ALS (alternating least squares) collaborative filter baseline, (2) a content tower over audio embeddings + metadata, and (3) a hybrid that blends them with context features. Evaluate offline on NDCG@10 (normalized discounted cumulative gain) and catalog coverage. Ship the winning model to the staging A/B harness for a 2-week live test on 10 percent of traffic. Deliver model code, an 8-page eval report, and the A/B readout.
The Brief
What you'll do, and what you'll demonstrate.
Ship a hybrid recommender that beats the production collaborative-filtering baseline on NDCG@10 and improves long-tail catalog coverage in a 2-week A/B test.
Earning criteria — what you'll demonstrate
- Train and evaluate ALS, content-based, and hybrid recommenders end-to-end
- Use audio embeddings as content features for cold-start tracks
- Reason about catalog coverage and diversity alongside ranking metrics
- Read an A/B test result without overclaiming early lift
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Product Manager
Product managers who can read NDCG and coverage trade-offs make the discovery-retention calls without deferring to data science.
This challenge sharpens
- ndcg-evaluation
- ab-testing
- recommendation-systems