Recommender Systems in the Age of Generative AI

ML Prague 2026 Workshop

Welcome! 🎉

Recommender Systems in the Age of Generative AI

ML Prague 2026 Workshop

May 4, 2026 | 14:00 – 17:30

Workshop Materials 📚

All materials available online:

jankislinger.github.io/recsys-genai

What’s included:

  • All presentation slides
  • Rendered Jupyter notebooks
  • References page with all cited papers

Your Instructors

Jan Kislinger

  • ML Engineer @ Sky Czech Republic
  • PhD Student @ Czech Technical Uni.

Bogdan Bodnár

  • ML Engineer @ Sky Czech Republic

What is Generative AI?

For Text: Transformers

  • GPT, BERT, T5, LLaMA
  • Self-attention mechanisms
  • Generation and understanding tasks

For Images: Diffusion Models

  • Stable Diffusion, DALL-E, Midjourney
  • Iterative denoising process
  • Text-to-image generation

This workshop: Focus on Transformers only

Workshop Schedule

Topic Duration
Introduction & Motivation 20 min
I Foundations 30 min
II Transformers & SASRec 40 min
Break 30 min
III LLMs for Recommendation 50 min
IV Generative Pages 30 min
Wrap-up & Resources 10 min

You’ll Learn…

Part I: Foundations

Classical Foundations

  • MovieLens dataset exploration
  • Collaborative filtering fundamentals
  • EASE - Embarrassingly Shallow Autoencoders
  • Two-tower models
  • Embedding-based retrieval

Why it matters:

These lay foundations for all modern recommendation systems, including generative approaches

Part II: Sequential Models

Sequential Models

  • Transformer architecture deep dive
  • Self-attention mechanisms
  • SASRec - Self-Attentive Sequential Recommendation
  • Temporal patterns in user behavior
  • Beyond static user embeddings

Why it matters:

Transformers revolutionized NLP and beyond

Part III: LLM-Powered Recommendation

LLM-Powered Recommendation

  • Zero-shot recommendation (no training data!)
  • Metadata augmentation with LLMs
  • Cold-start problem solutions
  • LLMs as zero-shot rankers
  • Bridging collaborative filtering and semantic understanding

Why it matters:

LLMs unlock new capabilities impossible with traditional methods

Part IV: Generative Pages

Generative Pages

  • Re-ranking strategies
  • Slate optimization techniques
  • Diversity and coverage constraints
  • Whole page optimization

Why it matters:

Modern platforms optimize entire personalized experiences, not just individual items

What You’ll Build

  • EASE (collaborative filtering)
  • SASRec (transformer-based sequential recommendation)
  • LLM-powered recommendations
  • Conversational recommendation
  • Generative recommendation pages

Tech Stack

Tooling

  • JupyterLab
  • Quarto
  • Ollama

Python Packages

  • Polars
  • PyTorch
  • plotnine
  • NumPy & SciPy

All running locally on your machine!

Prerequisites

Note

Level: Intermediate to Advanced

You should have:

  • Python programming experience
  • Basic ML knowledge
  • Laptop with setup completed

Be active!

Activity is appreciated and rewarded:

  • Participation in discussions
  • Hands-on coding during exercises
  • Asking questions
  • Sharing insights

🎁 Sky merch for active participants!

Let’s Get Started!