AI Systems · Education · 2026
Primer
Five concepts. Real understanding. A hands-on path through tokens, context windows, temperature, hallucination, and RAG.
Design + build · Live · Next.js · React · TypeScript · Tailwind CSS · Vercel
01 / Overview
Project notes
Problem
Most AI explainers either stay at metaphor level or jump straight to math. Non-technical readers never see what a model actually does with their text.
Build
Primer is a five-lesson path built around one case: a support bot that told a customer refunds take 90 days when the policy says 30. Each lesson asks the reader to predict first, explains the concept, opens a live playground, then checks understanding before marking the module complete.
Result
A reader finishes with a working mental model of why prompts cost what they cost, why a bot forgets, why outputs vary, where hallucination comes from, and how retrieval fixes it.
02 / Features
What it includes
Live Tokenizer
Type any sentence and watch it split into word, subword, space, and punctuation chunks with a running count and estimated cost.
Context Whiteboard
Fill a simulated context window for GPT-3.5, GPT-4, Claude 3, or Gemini Pro and watch old messages fall off the board.
Predict-First Prompts
Every lesson opens with a guess before the explanation, so the answer lands against something the reader already committed to.
Running Case File
One refund-policy incident threads through all five lessons, each one ruling out a different suspect.
Knowledge Checks and Progress
Three-question checks gate each module, an optional baseline gives a before score, and a progress bar tracks the path.
03 / Stack