Math basics
The few ideas the rest of the path leans on: vectors, matrices, probability, logs, slopes and averages.
Nine parts, one path. Each lesson pairs a short read with a lab you can push on: fit the line, turn the temperature up, watch an agent call a tool.
The few ideas the rest of the path leans on: vectors, matrices, probability, logs, slopes and averages.
Learning as fitting: loss, gradients, and why models overfit.
Text in, next token out: the transformer piece by piece.
The variants and tricks behind modern LLMs: attention, positions, architectures, scaling and multimodal models.
Pretraining at scale, then aligning and serving the model.
How to tell whether a model, a RAG system or an agent is any good.
Prompts, context and retrieval: getting reliable output.
One agent built from LangChain’s parts: a model, tools, the agent loop, middleware and memory.
Graphs of state, nodes and edges: control flow, parallel work, human approval and teams of agents.