Lilly Tech Systems
Designing Recommendation Systems at Scale
Master the architecture, algorithms, and production engineering of recommendation systems. From candidate generation and ranking models to real-time personalization and fairness — everything you need to build recommendation engines that power e-commerce, content platforms, and social media at scale.
Your Learning Path
Follow these lessons in order to build a complete production recommendation system, or jump to any topic you need right now.
1. Recommendation Architecture Overview
Two-stage architecture (candidate generation + ranking), content-based vs collaborative vs hybrid filtering, cold start solutions, and real-world system examples from Netflix, YouTube, and Amazon.
2. Candidate Generation
ANN search with FAISS and ScaNN, two-tower models, item-to-item collaborative filtering, embedding-based retrieval, and multi-source candidate merging with production code.
3. Ranking & Scoring Models
Learning-to-rank approaches (pointwise, pairwise, listwise), feature engineering for ranking, deep ranking models (DCN, DeepFM), and real-time feature injection.
4. Real-Time Personalization
User session modeling, contextual bandits for exploration, online learning, real-time feature updates, personalization API design, and A/B testing recommendation strategies.
5. Diversity, Fairness & Business Rules
Filter bubbles and echo chambers, diversity injection with MMR and DPP, business rule layers, position bias correction, fairness constraints, and multi-objective optimization.
6. Scaling to Millions of Users
Embedding table sharding, distributed inference, caching strategies, pre-computation vs real-time trade-offs, and system design for 100M+ users.
7. Offline & Online Evaluation
Offline metrics (NDCG, MAP, recall@k), online metrics (CTR, engagement, revenue), interleaving experiments, and long-term impact measurement.
8. Best Practices & Checklist
Production deployment checklist, common pitfalls, debugging recommendation quality issues, and a comprehensive FAQ for recommendation engineers.
What You'll Learn
By the end of this course, you will be able to:
Design Recommendation Architectures
Architect end-to-end recommendation pipelines for production — from candidate generation through ranking to real-time personalization with proper evaluation.
Build Production Pipelines
Implement candidate retrieval, ranking models, and real-time serving code using Python, PyTorch, FAISS, and feature stores you can deploy at work tomorrow.
Optimize Quality & Fairness
Measure recommendation quality with NDCG and MAP, inject diversity to avoid filter bubbles, and apply fairness constraints for responsible AI.
Scale to Millions
Handle 100M+ user workloads with embedding sharding, distributed inference, caching strategies, and A/B testing frameworks for continuous improvement.