Advanced Concepts Overview

What this section is

Foundations, Intermediate, and Advanced are a track: each chapter assumes you finished the one before it, and they end in a capstone. Advanced Concepts isn't that. It's a cookbook, a shelf of standalone, self-contained chapters on topics that don't need to happen in a specific order or lead anywhere in particular. Finish one, and you're done, nothing is left half-built for a later chapter to pick up.

Where you're picking up

Every chapter here assumes you've been through at least Foundations, ideally the whole three-tier curriculum. They lean on things already taught there, tokens, prompting basics, how a model actually generates text, rather than re-teaching them.

What's here

Prompt Engineering — how to write a prompt that says exactly what you mean: cutting the filler that wastes tokens, structuring instructions so nothing is ambiguous, and adding the one kind of constraint that actually reduces hallucinations.

Token & Cost Management — how to think about reducing token usage and LLM cost: trimming what you resend, provider-native prompt caching, routing tasks to the right-sized model, and batching what isn't urgent.

Agent Security — indirect prompt injection: malicious instructions arriving as tool output, like a document or email, instead of user input, and why the fix is constraining what a sensitive tool is allowed to do rather than trying to detect suspicious text.

Human-in-the-Loop — pausing an agent right before a specific, hard-to-reverse tool call, like sending an email or issuing a refund, and requiring a human's explicit approve, edit, reject, or respond decision before the real tool ever runs.

What's next

Pick a chapter title that catches your eye and start there. Nothing here is gated behind anything else in this section.