FirstPrompt

FirstPrompt teaches how language model text generation actually works, entirely through a transparent simulation in the browser. The Token Table splits your words into the pieces a model reads, with per token cost math you control. The Temperature Dial is the heart of the lesson: a clearly labeled simulated model reshapes its next word odds as you turn the dial or change top k, and fixed seeds make every draw replayable, so you can watch low temperature turn predictable and high temperature turn adventurous. The Context Window panel fills a conversation into a budget with three truncation strategies compared side by side, and the Prompt Builder assembles a structured prompt from role, task, constraints, and format while explaining each part is purpose. No APIs, no accounts, no mystery: everything computes locally and is labeled as a simulation. Forty two unit tests cover the tokenizer, seeded sampler, temperature and top k math, budget accounting, and truncation strategies.

Devpost Learn Skill Pack note: the scope, prd, and spec documents in the repo's devpost/ folder were produced by running the Skill Pack skills against this finished build during the submission period, and each file says so at the top. The app itself was built plan-first from an empty folder on Oct 24, inside the submission period, with 42 passing tests.

Built With

• css
• education
• html
• javascript