Pick a subject area, then work through its topics in order. Each one builds on the last.
Classical machine learning: regression, classification, and clustering, the ideas that came before deep learning.
Neural networks from a single neuron up to the mechanisms every modern architecture is built from.
How raw text becomes something a model can compute on: tokenizing it, filtering it, and reducing it to a root form, before it's ever turned into a vector.
Attention, multi-head attention, and the architecture behind every modern large language model.
Turning a raw LLM call into a production system: orchestration frameworks, stateful agents, and observability.