Recommender systems power products used by hundreds of millions of people, and the engineers who build them are among the most sought-after in the industry. This program teaches you to design, build, and deploy scalable, production-grade recommendation systems using enterprise tools and engineering practices.
You will progress through the full recommendation pipeline by building candidate generation systems with two-tower models, graph neural networks, and sequential transformers. You will develop deep learning ranking models such as Residual DCN, DeepFM, and DLRM; implement multi-task learning frameworks including MMoE and PLE; and deploy MLOps pipelines on Azure with real-time feature hydration, drift monitoring, and continuous retraining.
Throughout the program, you will work with industry-aligned architectures inspired by LinkedIn's publicly documented engineering systems, including PYMK, the LiNR neural retrieval index, and the structural goals of the 360Brew generative ranking framework.
By the end of the program, you will be able to design end-to-end recommendation pipelines, optimize deep learning ranking models, implement scalable ANN retrieval, and deploy production MLOps systems that meet enterprise standards for performance, fairness, and long-term utility. This program is intended for advanced machine learning engineers, infrastructure engineers, and data platform architects with experience in Python, deep learning, and core ML evaluation metrics.
Übungsprojekt
Throughout this program, you will complete intensive, hands-on projects mirroring the real challenges faced by LinkedIn and Microsoft engineers. You will design hybrid recommendation engines, build scalable ANN retrieval pipelines, train multi-task ranking models, and construct generative recommendation pipelines using managed cloud APIs and foundation model proxy architectures.
Each production project tracks architectural validation metrics, moving from candidate selection metrics in early courses to long-term utility tracking — including Saves, Dwell Time, and Private Shares — during ranking modules. Projects are executed programmatically through Jupyter Notebook Python SDK workflows on Azure Databricks and Azure ML, and culminate in portfolio-ready engineering documents and deployed cloud architectures.
