Providing Meaningful Search Results Without Own Index?

I have this idea for a search engine that would be similar to "bangs" in DuckDuckGo. Bangs mean that you prepend your search with a predefined shortcut to direct it to a specific site, like "!w" to search Wikipedia. When you prepend !w DuckDuckGo will redirect you to the Wikipedia search page with your query attached. In essence it's exactly like just going to Wikipedia and searching there from the start.

My idea is this: a lot of sites have a search API. That means my search engine could send your query to multiple different sites and collate the replies into a set for you to peruse.

But how accurate would a search like that be, and how should the results be weighted and presented? A model for each API called would have to be created and maintained, because Wikipedia and StackExchange have different APIs. And speaking of accuracy the latter actually say in their documentation that their search function isn't super and that you should use a "proper" search engine to get better results. On their own site.

Could quality results be found this way? Searching for "How to upgrade Ubuntu" will probably only turn up results from a couple of sources. It's not feasible to propagate the search to hundreds of sites; maybe only five or ten at most.

Of course SEO would be completely powerless to affect results, so that's pretty cool.

The cool thing about a search engine like this is that one could optionally do all the queries on the client side with JavaScript and thereby both saving resources on the server and keeping the user's search history from it.

Hmmm. Maybe I should just try this.


After looking around more I've concluded that too few pages have an API for this project to be viable. That's too bad.

-- CC0 Björn Wärmedal