Swiftype was one of the sleepers. They'd been growing really fast but were known mostly to their users rather than to the people who talk about startups. There are quite a few of these.
Interesting. I wonder if Swiftype uses Apache Lucene/Solr/Nutch as backend.
Facebook is listed as Client. They have a contract with Microsoft (Bing) and have at least two internal search engines. One of them is the type-ahead one as explained there:
I assume FB uses Swiftype not for the main FB website, maybe the developer documentation or some PR sites. (They use(d) even some SharePoint sites for business pages).
I swear Hacker News is inside my head. I was JUST investigating server-less (from my side) full text search engines like this over the last week. I checked out lunr.js, google custom search engine, sqlite, and a few other options. This looks great and I think this might be what I implement. Thanks!
I kinda anticipated this response. My only defence is that I find it difficult to maintain a 'domain -> crappy-annoying-behaviour' map in my head, but it's a fair point ;)
"while basic site search is simple (think adding a Google GOOG -1.13% site search toolbar to the top of your blog), good, effective site search is very, very hard"
What does the above statement mean? AND Why doesn't swiftype come up in techcrunch's autocomplete search?
There are many factors that make high quality search difficult, but as an example, take relevance that evolves over time. Static text-based rankings are a fine start, but improving search relevance continuously based on usage takes a lot of data. Even when you have the data, leveraging it requires careful tuning to prevent issues like over-training. This is just one of the things we do for people out of the box.
ok, so you guys track the clickstream & optimize for clicks that don't help the user. At a per site level i can see where the bottlenecks come in, but business must be good if those are your problems. Congrats!
