Purpose
The machine learning notes now split four ways. Deep learning covers neural networks from first principles and the main architecture families. NLP covers language representations and textbook modeling ideas. Recommender systems covers retrieval, ranking, and feedback signals in personalization. Serving systems covers the systems side of large-model inference: kernels, memory, batching, and parallelism.
These areas connect in useful ways. Deep-learning architecture choices shape what later NLP and recommender models can express, and serving constraints push back on those architecture choices. Large-model serving leans on the same performance reasoning that shows up in performance engineering and the same hardware constraints that show up in hardware notes.