IdeaTin

IdeaTin Episode 2: Sociaholic

Today I interview Greg Osuri from Sociaholic.com.

To download alternate versions: (MP3), (MP4) & (MOV-QuickTime).

What is Sociaholic?

A cool way to update your Facebook profile using Twitter.

To develop the video:

• Took one Friday night
• Used: iShowU, Flash CS4 (Trial Version), iMovie
• Stock Music from iStockPhoto – Searched for keyword “happy” for $39.

Total Cost for Video = $70.

Infrastructure:

• Slicehost
• Amazon EC2 (for managing larger than usual message queues)

Biggest Technical Problem:

• Developing a parser to convert the twitter post, to Facebook friendly content (including properly formatting images, videos, links, etc.)

Traffic:

• Initial Launch: ~1 – 5 Signups/Day
• Post-Video: ~20 Signups/Day
• Total Registered To Date: 400

Previous Startup: GridBag

Additional Information

I sent Greg some questions prior to the interview, and his responses can be seen below. It has been lightly edited for grammar, etc.
Why did you do this? i.e. what pain point caused you to develop this project?

I share a lot of links on twitter and used the Twitter application on Facebook to sync my updates. The problem is they seem to come plain and blank, @mentions come as-is which means nothing to a Facebook user and links are usually shortened and encoded (no preview) which destroys the click-thru rate. I wanted some thing that will publish to facebook like facebook does it with interactive experience  (videos, link preview etc.). We found an app (our competitor now) that does it but it took like 12 hours to publish which was terrible.

The click-thru problem may not have much impact of personal profiles but does if you have a Facebook Page that complements your twitter. We developed Sociaholic primarily for businesses that reach out to twitter users and want to expand that to facebook while engaging with them interactively.

What is your background? Are you a programmer, etc.?

I’m primarily a programmer, it took me 2 weeks to write the minimum viable product and launch it. Initially, we weren’t sure if my hypothesis about click-thru rates was valid, so we decided to launch it as soon as possible and see if we are really solving that problem. After 2 weeks, we had around 280 of 330 users actively using the app and most of them have some thing positive say. We launched the product on May 10th and the intro video on May 16 (I did the video myself).

What does the infrastructure of a Facebook App look like, and how does it differ from other apps (iPhone & Web Apps for example)?

There are two primary ways to render a Facebook app, using the Canvas or an IFrame. Writing it for Canvas using Facebook Markup Language (FBML) gives you easier integration. For example, getting the native look and feel and  better performance (caching etc), the down side is you have to test on the facebook platform and are limited on the amount of javascript you could use. On the flip side, iFrame apps gives you a lot of flexibility but may take longer to develop.

Facebook Apps are Web Apps, there is practically no difference except for FBML tags and Facebook Javascript APIs.
iPhone native applications are written in objective-c and run on the device and are capable or delivering rich local experience. We use Ruby on Rails which lets you take advantage of MVC model, i.e., separating your views (UI) from the Controllers and Models (Business Logic). That way you could rapidly build apps and serve content to multiple platforms and devices with little changes to your core business logic code.

What service providers are you using for hosting and why?

We use Slicehost because it was cheap and great to start. We will be moving to a hybrid (dedicated-cloud) infrastructure in a few months.

What interesting problems did you buck up along the way – both from a hacking perspective, and non-hacking?

We don’t use the Facebook API to generate Link previews because it does not work and is very buggy, we’ve developed our own parsers, which is perhaps the most complex piece of the solution, parsing web pages and identifying the optimal preview is very tricky.

When we started, our parsers weren’t that great but as we started observing the links users post in real-time we made several improvements and will continue to do so. Our technology goal its to eventually employ several search engine parsing algorithms and open source it. The solution is working great for the current user base which is very low, we process around 50 tweets/hour and the system seems very stable and we expect it will be for at least 1000 tweets/hour.

Anything miscellaneous that might be interesting to the Hacker News community?

We are also developing a custom load-balancing system that works on custom triggers, for e.g., We get tweets streamed directly from twitter in real-time and that listener does not process these updates but puts them in a message queue which fires up the parser asynchronously to process links etc. and then posts the processed updates to another queue which will be picked up by a facebook publisher.

Since one of the selling points is that tweets gets published to facebook in real time, we want to keep our queue threshold to 10, i.e., if the twitter queue size exceeds 10 tweets that need to be processed, we want to fire up another EC2 instance to process the waiting twitter status and terminate it if less than 9 tweets are waiting. That way we save money. This is also an excellent use case for cloud computing.

And finally, what advice would you give to anyone wanting to build something?

I come from a C background and love optimizing systems, and when we hacked up sociaholic using ruby and sql-lite, I was embarrassed to launch a buggy product, but we came to a stark realization that speed-to-market is extremely important in the early stages. My advice is to launch it as soon as possible, your goal in the initial stages should really be about validating your product and market. The biggest reasons products fail is not because the technology sucks but because people don’t have the same problem that you think they have or don’t really need the solution. Most of the engineers want to build the best product only then launch it, that may or may not but launching it just to validate your hypothesis will clear your doubt and put you in a clear direction.