Skip to the good stuff
ELI5 Park
🏠 ELI5 Park / Docs / How Pip Works

🐿️ How Pip Works

A question helper that only knows what's on ELI5 Park.

The short version

Pip is a search helper, not an AI chatbot. When you ask a question, Pip looks through every ELI5 Park page, finds the section that best matches, shows you a short quote from it, and links you to the page.

Pip never writes its own answers, so it can't make things up. It only quotes what's already on the site. And it runs entirely in your browser: nothing anyone types is sent to us or anyone else.

βœ… What Pip does

Finds the right ELI5 Park page, quotes the most relevant sentences, and links to it. Says "I couldn't find that" when the site doesn't cover something.

🚫 What Pip doesn't do

Answer from the internet, invent explanations, remember anything after the tab is closed, or send questions to a server.

πŸ”¬ Watch Pip think

Type a question to see exactly what Pip does with it: which words it keeps, which page sections score highest, and whether the best one is good enough to show.

Step by step

  1. Pip reads the whole site ahead of timeWhenever a page is added or changed, a small build script (tools/build.mjs) reads every page, throws away menus, code and pictures, and cuts the text into sections at each heading. Tiny bits like a one-line subtitle get glued onto the next section so every section has real explanation in it. The result is one file, assets/search-index.json: about 200 pages split into roughly 900 sections. That file is Pip's entire brain.
  2. Your question becomes a handful of key words"How does my tummy work?" turns into just tummy. Pip lowercases everything, drops filler words (how, does, my, the, what...), and trims word endings so "vaccines" matches "vaccine" and "learning" matches "learn". A small list of kid-speak swaps adds helpers too: tummy also looks for stomach and digestion, shots looks for vaccine.
  3. Every section gets a scorePip uses a classic search formula called BM25, the same idea behind many search engines. In ELI5 terms: a section scores higher when it uses your words, rare words count more than common ones ("momo" matters more than "food"), and saying a word ten times doesn't count ten times as much. On top of that, words in a page's title count triple and words in a section heading count double, a page's opening section gets a small boost (that's usually the main explanation), and menu pages and policy pages are pushed down so they're never the main answer.
  4. Pip checks the winner is actually about your questionSharing one word isn't enough. "Who won the world cup?" shares the word world with a page about Nepali architecture, but that page isn't about the world cup. So the best section must pass two checks: at least one of your words must be in its page title or section heading, and it must contain most of your words (all of them for short questions). If nothing passes, Pip says it couldn't find it and suggests other questions instead of guessing.
  5. Pip shows a quote and linksFrom the winning section Pip picks the one or two sentences that contain the most of your words, highlights those words, and shows links to up to three pages that scored well.

Keeping kids safe

What Pip is not good at

Because Pip matches words rather than understanding meaning, a few things trip it up:

We think these limits are worth it: a helper for kids should be honest about what it doesn't know rather than confidently making something up.

Why Pip doesn't use AI (yet)

Lots of chat helpers today are built on AI language models, the technology behind tools like ChatGPT and Claude. Pip deliberately isn't. Here's the trade-off we made.

🐿️ Pip today (search)

Only quotes sentences that are already on ELI5 Park, so it can't invent facts. Runs on your device, so questions stay private. Free, works offline, and its safety rules are fixed code that always behave the same way.

πŸ€– An AI language model

Understands meaning, not just words. Can explain things in its own words, compare two topics and answer "but why?" follow-ups. But it can sound sure while being wrong, questions are sent to an AI company, and every question costs money.

Why we started without AI

How AI could be added safely someday

If we add AI, it would build on Pip rather than replace it:

  1. Pip still searches firstThe same search finds the few ELI5 Park sections that best match the question, and the same safety rules run before anything else.
  2. The AI only sees those sectionsThe question and those sections, and nothing else, go to the AI model, with instructions to answer in simple words using only that text, and to say "I don't know" if the text doesn't cover it.
  3. Answers still link back to the pagesSo you can always check where an answer came from.

Before switching that on, we'd update our Privacy Policy to say which AI company receives questions, set limits so it can't be abused, and keep today's search-only Pip as a fallback that can be turned back on at any time.

Found a question Pip gets wrong, or have thoughts on AI for kids? Email digitalcommonground@gmail.com.