Design Spotify Song Discovery Engine — System Design Practice | AmanAI Lab

Sheet›Design Spotify Song Discovery Engine

AppleAmazonNetflix

45:00

Hard

The Problem

Problem

Design Spotify's music discovery engine — the system that powers Discover Weekly, Daily Mixes, and Radio — for 400 million users and a catalogue of 100 million tracks.

The system must balance:

•Personalisation (tracks the user will enjoy)
•Discovery (tracks they haven't heard that they'll love)
•Serendipity (pleasant surprises outside their comfort zone)

What you'll be assessed on

Collaborative filtering at scale, audio-based content embeddings, the exploration-exploitation trade-off, and playlist coherence.

Scale & Constraints

▸400M users, 100M tracks
▸Discover Weekly: 30 tracks, refreshed every Monday morning
▸Recommendations generated within 2 hours for all users
▸New track cold start: must appear in relevant playlists within 24h of upload
▸P99 radio recommendation latency ≤ 200ms
▸Diversity: no two adjacent tracks by the same artist

Must Cover0/9

Hints (if stuck)

💡 Matrix factorisation on play counts gives good similarity; audio embeddings handles cold start.

💡 Positive signal: stream >30s, save, add to playlist. Negative: skip <5s, hide.

💡 BPM and key analysis via audio fingerprinting for seamless playlist flow.

💡 Diversity via the inverse frequency boost — boost less-heard artists proportionally.

0 words · auto-saved

Back to sheet

Mock interviewTake quizFlashcards

Problem

Design Spotify's music discovery engine — the system that powers Discover Weekly, Daily Mixes, and Radio — for 400 million users and a catalogue of 100 million tracks.

The system must balance:

•Personalisation (tracks the user will enjoy)

•Discovery (tracks they haven't heard that they'll love)

•Serendipity (pleasant surprises outside their comfort zone)

What you'll be assessed on