Glossary — T
Back to glossary index · Master INDEX
Tailwind CSS
A utility-first CSS framework. Instead of writing CSS classes like .btn-primary in separate files, you compose styles directly in your HTML/JSX with classes like flex items-center px-4 py-2 bg-blue-500 text-white rounded. Each class does one thing. Fast to write, easy to refactor, no naming collisions. The dominant styling approach in modern React/Next.js.
TCP / UDP
Two foundational network protocols. TCP (Transmission Control Protocol) guarantees ordered, reliable delivery — used for HTTP, email, most web traffic. UDP (User Datagram Protocol) is fire-and-forget, no guarantees — used for video calls, gaming, DNS, where speed matters more than every packet arriving.
See also: TCP vs UDP
Terminal
The text-based interface to your computer. Type commands, get text back. The terminal is the program you type into (e.g. Windows Terminal, iTerm2); the shell is the language you’re typing in (Bash, Zsh, PowerShell). Often used interchangeably, but technically distinct.
See also: Terminal & command line, Bash
Test
Code that verifies other code works. Three main flavors: unit (one function in isolation), integration (a few pieces working together), end-to-end (the whole system, driven like a user). Good tests run fast and fail clearly.
See also: Why test?
Throttle vs debounce
Two ways to limit how often a function runs in response to rapid events.
- Throttle: run at most once every N milliseconds. (Good for scroll handlers.)
- Debounce: run once, only after N ms have passed since the last event. (Good for search-as-you-type — wait until the user stops typing.)
Temperature (LLM)
A setting that controls how predictable vs creative an AI response is. Temperature 0 = always pick the most likely next word (deterministic; same input → same output). Temperature 1 = balanced. Temperature 2 = creative/random. Use low temperature for factual tasks, code, structured output. Use higher for brainstorming, creative writing. Most consumer AI hides this setting; APIs expose it.
See also: Temperature & sampling
Token (LLM)
The unit an LLM actually sees. Roughly one token ≈ 4 English characters or ¾ of a word, though it varies. Models have a maximum context window measured in tokens (e.g. 200K tokens for Claude Opus). Pricing is per million tokens of input and output.
See also: Tokens & context windows, LLM
Transformer
The neural network architecture introduced in the 2017 paper “Attention Is All You Need,” and the foundation of every modern LLM. The key idea is attention — letting the model look at every position in the input when computing each output, weighted by relevance. Made it possible to train enormous models efficiently on GPUs.
See also: How LLMs work
TypeScript
JavaScript with optional static types. Wrote function add(a: number, b: number): number — TypeScript checks at compile time that you call it with two numbers and use the return value as a number. Catches a huge class of bugs before runtime. Compiles to plain JavaScript. The default for serious modern projects.
See also: TypeScript, JavaScript
← S · Glossary index · U →