Building LLM-Powered Recommendation Systems

Course details

What is this course about?
Get a technically grounded overview of how to start building the next generation of intelligent recommender systems. Moving beyond traditional algorithms, this course shows you how to enhance existing systems by applying AI-powered techniques for embedding generation, semantic reranking, and cold start mitigation. Instructor Rishabh Misra outlines how to design sophisticated GenAI-native architectures that enable dynamic experiences such as conversational search and multimodal recommendations. The course emphasizes robust evaluation, including how to measure quality, fairness, and factual accuracy using approaches like retrieval-augmented generation (RAG). By the end, you’ll be prepared to design, evaluate, and operationalize effective and responsible GenAI recommender systems in a production environment.

Instructor
Who teaches this course?
Rishabh Misra is a Principal ML Engineer at Atlassian, where he leads LLM post-training and GenAI personalization initiatives.

Objectives
What will I be able to do by the end of this course?
Articulate the differences between traditional recommender systems and modern GenAI-powered approaches, including the shift to semantic understanding.
Apply practical GenAI techniques such as embedding generation, chain-of-thought reranking, and retrieval-augmented generation (RAG) to improve performance and trustworthiness.
Design high-level architectures for GenAI-native recommender systems, selecting appropriate models and infrastructure like vector databases.
Develop evaluation strategies using metrics for quality, fairness, and factual accuracy.
Create operational plans for deployment, including latency management, model monitoring, and CI/CD pipeline integration.

Audience
Who is this course for?
Software engineers
Data scientists
AI and ML engineers
Technical product managers

Prerequisites
What do I need to know before taking this course?
Basic understanding of machine learning concepts
Familiarity with AI and ML frameworks and tools
Experience in software engineering or data analysis is beneficial

Concepts

Introduction

Discover the power of generative AI for recommendation systems

High-Impact GenAI Enhancements for Recommenders

Choosing your GenAI tool - LLMs, GANs, and diffusion
Creating quality embeddings with sentence transformers
Foundational follow-up - The core shift From item IDs to semantic embeddings
Summarizing user history for better personalization
Solving item cold-start with cross-modal embeddings
Few-shot prompting for personalized explanations
Data augmentation - Creating hard negatives with LLMs
Foundational follow-up - Augment vs. replace The LLMERS production pattern

Architecting GenAI-Native Recommender Systems

Foundational follow-up - How LLMs understand A primer on the transformer
The generative retrieval architecture
The key component - Semantic item tokenization
Building conversational recommenders with tool use and RAG
Multimodal fusion - How cross-attention works
Architectural challenge - Managing long-term user memory

Evaluating GenAI Recommenders - Quality, Fairness, and Trust

Evaluating recommendation lists - Diversity and novelty
Evaluating generated text - ROUGE, BLEU, and BERTScore
The RAG architecture - A deep dive into factual grounding
Red teaming - Proactively finding failure modes

Operationalizing GenAI Recommender Systems at Scale

Production infrastructure - Vector databases and model serving
Foundational follow-up - The two-tower model
Monitoring for embedding drift and quality degradation
Scaling for inference - Quantization and knowledge distillation

Conclusion

Course summary
The future is agentic - Designing recommenders as autonomous agents

Building LLM-Powered Recommendation Systems

Authors

Rishabh Misra