personalized-recommender-course

Why we included this project

A recommender that works in a notebook is one thing; one that serves live traffic is another. This course uses a concrete H&M fashion dataset to show the full path, from a four-stage recommender with a two-tower retrieval network to feature pipelines backed by a feature store, training and evaluation of retrieval and ranking models, and deployment to Kubernetes with KServe. Each module pairs a written article with runnable notebooks and a reusable Python package, so you can follow along locally, on Colab, or via GitHub Actions without a GPU. The final module layers LLM-based ranking on top of the classic architecture. If you're planning your own recommender, the Feature/Training/Inference separation and deployment patterns are directly transferable, and the live Streamlit demo shows the finished system before you start.