Find similar science fair projects
Two projects can be about the same thing and share almost no words. Keyword search cannot connect them; searching by meaning can. This is how that works here, and where to use it.
Why keyword search misses the projects you want
A word-match search looks for documents containing your terms. It requires the author to have chosen the same vocabulary you did, which for student research is a bad bet: the same idea gets written up as "low-cost water quality sensor", "colorimetric assay for nitrate" and "Arduino-based turbidity meter" depending on who wrote it and which part they thought was the contribution.
Worse, a descriptive query gets less likely to match the more you describe it, because every extra word is another term the document must also contain. That is why typing a whole sentence into most archive search boxes returns nothing.
What searching by meaning does instead
Each project's text is turned into a vector — a long list of numbers positioning it in a space where things that are about the same subject land near each other, regardless of wording. Your query gets the same treatment, and the search returns whatever is nearest. Match is by subject, not by string.
The practical consequences are worth knowing before you trust a result:
- There is always a nearest project, even when nothing is genuinely close. Distance is shown wherever it matters so you can tell the difference.
- Results are ranked, not filtered. A meaning search returns a fixed number of nearest projects rather than a count of matches, so "12 results" means "the 12 closest", not "12 projects exist".
- It is good at subject and weak at detail. It will find every project about drug delivery; it will not reliably separate the ones that used a particular polymer. Narrow with the category and year filters, which are exact.
Three places to use it
From a search box
Describe the idea in a sentence on Explore. If your words match almost nothing, the page retries by meaning on its own and tells you it did — a search that matched two projects by accident is a worse answer than the closest ten by subject.
From a project you already found
Every project page ends with its nearest neighbours across every fair and year. This is the fastest way to widen a single promising result into the cluster it belongs to, and it reaches work that shares an approach without sharing vocabulary.
From a whole proposal
Paste an abstract into the novelty check and it measures your proposal against the entire corpus at once, reporting how many projects sit within each distance band, how recent they are, and how many of them placed. That is a different question from "show me the top matches": it tells you whether you are entering a crowded field or an empty one.
For tools and agents
The same search, similarity and gap-finding functions are available to language models over MCP, so an assistant can check the archive while you are talking to it rather than guessing from memory. Connect it →
Start searching
20,037 projects, 2014–2026, from 4 competitions.