Deep Learning

Deep learning is the engine under every system in this guide. LLMs, embedding models, image generators, speech recognition — all are deep neural networks. You won’t be implementing backprop by hand, but understanding what a network is and how it trains makes the rest of AI engineering legible instead of magical.

How deep to go

You need a working model, not a research-grade one. Aim to be able to explain attention to a colleague and reason about why training is expensive. Skip the calculus.

In this section

Neural Networks Neurons, layers, weights, and activations — what a network is and how a prediction flows through it.

Training Neural Networks Backpropagation, optimizers, batches and epochs, regularization, and why GPUs are mandatory.

Key Architectures MLPs, CNNs, RNNs, and the transformer — what each is for and why transformers took over.

What you’ll be able to do

Explain how a neural network turns input into a prediction, describe what happens during training, and recognize the major architecture families — so the word “transformer” in LLM Engineering means something concrete.

Prerequisites

AI Fundamentals — especially the training loop in How Models Learn.