The essay made the argument: modern AI is the search engine you already know, wearing new vocabulary. This is where you get to play with the parts. Type into them, press play, and watch the algorithms do their quiet, ordinary work.
← // you are in the companion · read the essay first You Already Know AI – You Just Called It Search// 01 · keystroke → ranked resultsThe Query Lifecycle
Start with the whole journey. Follow a single word from the search bar all the way to ranked results – language detection, normalisation, dictionary lookups, and the learning loop that quietly gets better every time. Press play and watch one query travel the whole pipeline, then meet the individual parts below.
// 02 · the ancestor of next-token predictionAutocomplete
Type a prefix and a tree finds every word beneath it. A prefix tree (Trie) walks one character per node; a Ternary Search Tree compares left, equal, or right at each step. An LLM runs this same instinct millions of times a second – given what came before, predict the most likely next piece.
// 03 · counting, not magicN-Grams
How does a model guess the next word? Before transformers, it counted. An n-gram model looks at the last word (or two) and picks the most likely follower from what it has seen before. It is the direct, humble ancestor of an LLM's next-token guess.
// 04 · the typo fixerEdit Distance
How does recieve become receive, and tokyo become Tokyo? Edit distance counts the smallest number of insertions, deletions, and substitutions to turn one string into another – the fuzzy matching that forgives your spelling without you ever noticing.
// the pointSame Problems, New Names
None of this was invented in 2017. Prefix prediction, language detection, fuzzy matching, counting what follows what – search engineers solved these problems for a decade before anyone called them AI. If you understood them then, you already understand more of the machine than you think. The vocabulary changed. The instincts did not.