link

Wednesday, December 23, 2015

Transformy just got better

It’s been awhile since we posted an update, but we wanted to have something real to show you. And now we have.


Over the past few months, we’ve made Transformy better. Gradually, slowly, but definitely better. Today we feel like we owe you some more details, because the improvements are made possible by the continuous feedback and usage statistics you provide us with. So thanks for that!


When we launched Transformy, the tool was more like a prototype than an actual product. Despite the primitive capabilities, it has helped a lot of you every day to get your data munging and wrangling done without too much hassle.


But due to the limited capabilities of the prototype, sometimes you hit a wall. Today we’ve upgraded our algorithm, to deal way better with variations between your source lines.


Let’s explain with an example, because that’s the Transformy way. Suppose we extracted the following dump from the web service of our CRM suite. We get our contact names and their birthday. Of course, we’ve minimized it for the sake of the example, we hope your CRM contains more info.


{"name": "John Ismael", born: "1965-08-24"},
{"name": "Aaron Michell Runebergh", born: "1981-04-16"},
{"name": "Kaylee Richardson", born: "1976-09-02"},


Great, but we need it in a CSV format, because that’s what our visualization tool expects. Also, let’s drop the month and day from the birthday, as they’re not relevant for this report anyway.


So let’s just give the example and be done with it!


John Ismael, 1965


Previous versions of Transformy would output the following result.


John Ismael, 1965
Aaron Michell, born
Kaylee Richardson, 1976


Why is that? The algorithm did not take into account the variation happening in the second line; Aaron Michell Runebergh is composed out of 3 words rather than the 2 words of John Ismael used in the example line. Transformy got confused by this and was unable to properly align the source lines, making it miss the 3rd name and misinterpreting the year. That’s bad.


Luckily, the new algorithm is way better at this. It will take the structure of every line into account to make an educated guess at the alignment. In this case, it will understand that the name is composed out of 3 words and properly align them with the 2 words of the example name, resulting in the desired output:


John Ismael, 1965
Aaron Michell Runebergh, 1981
Kaylee Richardson, 1976


Next to the changes under the hood, we’ve also added minor usability improvements. It’s now possible to copy the result to the clipboard with one button. To help us further improve Transformy, you can also signal whether the result is what you expect, all in one mouse click. So check it out!


These improvements are only the beginning. Expect a lot of improvements over the coming weeks and months. So please, keep providing us with feedback. What’s working, but more importantly, what isn’t? How would you like to use Transformy? What are your most painful data wrangling tasks?

Tuesday, June 23, 2015

Transformy was first on Hackers News and Product Hunt!

A few people asked us about how we got to making transformy and what kind of traffic we received from being first on Hackers News and Product Hunt. So here's the story. 

On the 14th of March, the first version was ready. We showed transformy to some people but did not receive much feedback. To many, it wasn't even clear what the app was supposed to do, while we felt it was too simple at the time to show to a tech-savvy audience such as the one on Hackers News.

We started working on a much more complicated version of the algorithm but it was taking way longer to complete. So after a talk, we decided that for a one-week project this was taking too long and we should get some feedback to see if anyone was actually interested in this thing.

We rushed the launch

By Thursday, April 23rd (yes, way longer than a week), we had reverted to the first version and ironed out most bugs. The next day, I wanted to post transformy on Product Hunt. Apparently you can only log in with a twitter account... and twitter is blocked where I work. No problem, I ask Tomas to post it. Well, apparently you need an invitation before you can actually post something. And none of us had a Product Hunt account yet. So that was disappointing. We thought this could get some traction on Product Hunt. Not so much on Hackers News, people over there seem to be more sceptical, and transformy wasn't that advanced yet.

Around 3PM I submitted the link on Hackers News anyway, not expecting much to happen but always hoping. It took us only about an hour to get on the front page, and a little later we even made it to the first spot! The next day someone had submitted us on Product Hunt as well. Again, after a few hours we got first on Product Hunt as well.

So many visitors

We received huge amounts of traffic the first few days. Transformy was also picked up by some other sites, but nothing as impressive as these two.


Most part of the traffic was obviously from Hackers News but Product Hunt kept sending us visitors for a longer period.


We posted transformy on a Friday and in hindsight that might not have been a great idea.  I think traffic would have been higher if we had posted it earlier in the week, but maybe it would have been more difficult to get to the first spots as well.

More than just traffic

For the long term, the traffic we received is probably the least significant, compared to the other things we gained in the process:
  • Lots of insight on how people use transformy.
  • Feedback about bugs and unexpected behaviour.
  • Some people made great suggestions, and we will be implementing a few ideas soon.
  • Many subscribed to our mailing list and started following us on Twitter.

Problems

As was probably to be expected, there were a few problems when traffic started peaking. First off, we had made a small error where a new request was sent to the server after every letter that was typed. And naturally, our server was having some trouble with that but we were able to fix it quite fast.

Worse was the fact that we hadn't properly tested our contact form and mailinglist signup. We lost about 35 contact requests and 120 mailinglist signups.

Last thing, and this was more of an oversight, we didn't have a twitter account. Assuming the traction would be minimal we never really thought it was necessary. We created one and added a link to it a few hours later and we still managed to gather some followers, but it was definitely a missed opportunity.