Case Study

Reinventing Spotify's Discovery

AI Leadership Case Study: How I led the transformation of Spotify's music discovery systems using machine learning.

Use arrow keys to navigate slides. Click fullscreen for the best experience.

Go Deeper

Want to explore the technical foundations behind modern recommendation systems? These articles dive deep into the methods powering music discovery at scale.

Two-Tower Networks & Negative Sampling

Deep dive into the architecture powering candidate generation at scale. Understand how separate query and item towers enable efficient similarity search across millions of items.

Towards Data Science Architecture Overview Google Research Mixed Negative Sampling
Architecture

Modeling Sequences of User Actions

How transformers and self-attention capture temporal patterns in listening behavior. From context-dependent preferences to modern architectures like SASRec and BERT4Rec.

Google Research Transformers in Music Recommendation Shaped.ai From N-grams to BERT4Rec
Sequential Modeling
Spotify Research

Graph Neural Networks for Audio Content

How Spotify uses two-tower heterogeneous graph neural networks (2T-HGNN) to solve cold-start problems and enable cross-format recommendations between podcasts, audiobooks, and music.

Embeddings

Questions?

I'd love to hear from you. Let's connect and discuss AI, machine learning, or potential collaborations.

Get in Touch