AI Fundamentals

Tokens, embeddings, context windows, RAG — the vocabulary, and the sane path to actually building with AI instead of just prompting it.

0 of 30 lessons complete~350 min total

Start track

Sort

Tokens & TokenizationIntermediateEvery LLM API you'll ever call bills you, limits you, and thinks in tokens, not characters or words — and the gap between those three counts trips up almost everyone the first time.11 min · 0/3 questions correct
Context WindowsBeginnerA model's context window is one shared token budget for the whole request — system prompt, history, retrieved documents, the new message, and the response all have to fit in it together.11 min · 0/3 questions correct
Embeddings & Vector RepresentationsIntermediateAn embedding turns a piece of text into a fixed-length list of numbers such that similar meaning lands nearby in that number space — the mechanism underneath search-by-meaning instead of search-by-keyword.12 min · 0/3 questions correct
Temperature & Sampling ParametersIntermediateTemperature controls how predictable versus varied a model's output feels, by reshaping the probability distribution it samples the next token from — and it's the first knob to reach for when output feels too rigid or too erratic.10 min · 0/3 questions correct
System Prompts vs. User PromptsBeginnerChat APIs structure a conversation into roles — system, user, assistant — and knowing which instructions belong in which role is the difference between a feature that behaves consistently and one that doesn't.11 min · 0/3 questions correct
Prompt Engineering FundamentalsIntermediateA vague prompt gets a vague, inconsistent answer for a specific reason: the model has no signal narrowing what a good response looks like. Prompt engineering is the discipline of giving it that signal on purpose.12 min · 0/3 questions correct
HallucinationIntermediateAn LLM doesn't have a fact database it checks against — it generates the statistically plausible next token. Hallucination isn't a bug in that process, it's what that process looks like when there's no grounding to lean on.11 min · 0/3 questions correct
Streaming ResponsesIntermediateThe same multi-second wait feels broken with a blank screen and feels responsive with tokens trickling in — streaming is a UX fix disguised as a networking detail.11 min · 0/3 questions correct
Function Calling & Tool UseIntermediateThe model never runs code — it asks your code to run something and hands back a request. Everything people call 'the AI checked the weather' or 'the AI queried the database' is this one mechanism underneath.12 min · 0/3 questions correct
Structured Output & JSON ModeIntermediateA model's raw output is free-form text — but your code usually needs a specific, parseable shape it can actually run logic against. Getting from one to the other reliably is its own problem, with solutions that differ a lot in how much you can trust them.11 min · 0/3 questions correct
RAG (Retrieval-Augmented Generation)IntermediateThe model didn't train on your company's internal docs, and it never will — RAG is how you hand it the right answer to look at instead of hoping it already knows one.13 min · 0/3 questions correct
Vector DatabasesIntermediateA vector database isn't storing your documents — it's storing what your documents mean, indexed so you can find the closest meaning in milliseconds instead of hours.11 min · 0/3 questions correct