Reflections on start-up life: Week Seven

Last week was very productive — I think there is a lesson in there about focus, although it’s also the inevitable issue of everyone else but us being on leave so we didn’t have anything to do BUT product development!

The first release of http://twendly.com was very interesting; interesting enough that it was an itch we’ve had to scratch a bit further. Our major highlight last week was implementing a series of features into Twendly which allowed it to index a lot more data and therefore become more useful. If you’re looking for people, searching a pool of 16,000 which we are now is a lot better than a pool of 35 (our original release).

This has lead to all sorts of technical challenges, not least that when we are now processing almost 20,000,000 rows of data there is a good chance that that a “1 in a million” condition occurs about 20 times every time you process. I’m pleased I’m paranoid about error handling, every single line of code to handle errors that I wrote (usually with a comment like “This is highly theoretical, should never happen”) actually executed (ie. the error did occur).

To quote Terry Pratchett, “One in a million chances occur nine times out of ten.”

It’s been fairly intense and now that http://twendly.com is live and starting to get a little bit of usage (there are a fair few searches going through, even if not everyone searching registers) there is a new feeling of pressure. It’s great to have a showcase, it’s now problematic if that showcase isn’t working as well as it could, the site goes down etc. etc. The data also comes at a price — we can’t just “blow it away” and start over either.

Anyway for those who are wondering just what we’ve been up to and what “people search” is really about, the three attached pictures sum it up pretty well I think. I’ve done three searches using exactly the same terms “Lotus Connections”.

The first is a search for “Lotus Connections” in Twitter — you get the current tweets that mention Lotus Connections, however there is no way to know or understand which of these people talksmost frequently about this — it may be that this was the very first time they ever mentioned it.

The second image shows a search in Google. This takes you to some great information about Lotus Connections, but these are largely all documents — technical or sales information about the Lotus Connections product, useful, but not people you can have a conversation with. Only one link will really lead you to a person (the Lotus Connections Blog link) if that’s what you were after.

Finally we have a search in Twendly. Of these, 8 out of 10 are people and 7 of the Top 10 are actually a great match for this search. There is quite a bit we can do to further refine and improve this (for example I could manually refine the search to exclude web links which I know improves the results even more but would be cheating for this demonstration where I’ve used exactly the same terms for each), but I think this is still very successful in demonstrating the value of the HiveMind engine as a different type of search engine — looking for people and not documents.

Highlights?

• Getting back into it
• Having a nice focus for the week
• Getting a more useful version of Twendly up — excitement that it’s all going so well.
• Seeing things come together in production.

Lowlights?

• Feeling the pressure now with a live site
• Not much really, a very productive week that I was proud of.

Goal this week?

• Not sure — will start with a review of where we think Twendly is at and a decision on wether we put some more time in it