Most AI courses start with "import tensorflow" and skip the part where you understand what you just imported. This course goes the other way: you build everything from scratch first, then appreciate what the libraries do for you.

The course covers three layers. First, the mathematical foundations that every paper assumes you know — linear algebra, calculus, probability, and optimization. Not as a textbook exercise but as the language you need to read research. Second, neural network building blocks — backpropagation, normalization, regularization, residual connections — each one taught as a solution to a specific training problem. Third, representation learning — how models learn to encode meaning, from word embeddings through contrastive learning to sequence models.

By the end of this course, you will have implemented a multi-layer perceptron, a CNN, an LSTM, and a word embedding model from scratch. You will understand why Adam works better than SGD, why batch norm helps training, and why RNNs were replaced by transformers. These are the foundations that make the LLM course (Course 2) possible.