The first full week back from San Francisco is done. Boy, what a change of pace!
After a hectic time rushing from meeting to meeting, it’s now back to the challenge of coding and trying to re-engage a meeting schedule here back in Australia. The challenge has been switching back to a self motivation mode.
Really our biggest challenge now is focus. It’s not that we don’t know what we need to do, it’s that there is so much of it. I’ve said to a few people that we could literally employ six others full time — not that we have the money to do that yet! The trick is really trying to balance between the longer term things that need focus now to happen in three months, Vs. the shorter term critical things to get us into the hands of end users.
Despite all this, we are closing in rapidly on a product that we can charge money for. We are getting closer to our customers, we are finding out what they need and what they like (and what they’re not so sure of) about Tribalytic. Yet there is still so much more to go to leap the gap.
One of the best thing about this blog is the ability to go back and review history. I was about to say “It’s been so long since we’ve updated the product” yet the current beta was released in Week 21 (http://timbull.com/reflections-on-start-up-life-week-21) which was only four weeks ago! Since then, we’ve spent a week with customers, close to three weeks on the road and now here we are.
In that time, our own experiences and the feedback on Tribalytic have shown several short comings that need to be addressed, so we are back into a round of engineering (a somewhat shorter one this time) to address these. While technically challenging, these are essential for the accuracy and performance of the product. Our next beta (fingers crossed for next week) will support:
- A near real time index of Australian Twitter users and a greatly expanded index.
- A revised stemmer (which means that Bunnings won’t be treated as Bun and related to “Hot, Cross and Easter” (of all bugs this one makes me laugh the most). For those interested, the basic stemming algorithm we were using drops the s, then the ing, then the double n to get to a word “stem” or root. Works great for walking, walked walks, but obvious issues for Bunnings.
- Some level of boolean logic in our search engine (AND, OR and NOT). e.g @kevinruddpm OR rudd
The more astute of you will of noticed that this post is also pretty much two days late now (usually first thing Monday morning). Last week turned into a long week. On a personal note (I can’t comment overly about Alex’s technical struggles although he’s had no shortage of challenges as well) I’ve been challenged with getting our processing pipeline live. This is the thing that collects all the tweets and keeps everything up to date. It would run for a couple of hours, then choke up and die. Finally this morning I nutted it out after working through close to a million lines of log file to locate the issue.
Over 5 days of wasted elapsed time for (I kid you not) one parameter on a command. Not even one line of code, it was 12 characters (including the spaces). (For the technically minded, I hadn’t set a timeout on the socket call to Twitter and on rare occasions the call would just never connect and not fail so it caused the thread to block and never return — adding a timeout resolved the issue).
Finally the pipeline is now up and running and that’s one less issue I need to concern myself with.
Working with Alex continues to be a lot of fun. It’s more challenging now he’s back in Beijing and I’m here in Melbourne, but we continue to be able to work through issues via email and Skype without too many problems. Probably the most challenging thing is that in the last couple of weeks we’ve passed some kind of event horizon where we no longer understand what each other is coding! The complexities of the engine and it’s search are a black box to me, while the processing pipeline and the overall processing architecture is just not worth Alex investing his time in understanding while there’s a search engine to work on.
I’m often amused by our conversations — my favourite of the last week (recorded here for my memory more than anything else) is the detailed discussion on how to weight relevant people around search terms. If two people talk about magnum and icecream 10 times, but one person mentions magnum 7 times and icecream 3 times and the other magnum 3 times and icecream 7 times, who should be listed first?
What if the a user searched for icecream and magnum vs magnum and icecream. What if one persons conversations were all in the first few days and the other persons in the last two? For the record — there are no good answers to these questions, but it does show the rabbit holes you can work yourselves into.
We continue to meet with investors and still have some meetings in the pipeline. There seems to be good consensus that we have a business, the question we now need to address (at least for Venture level investors) is do we have a $100 Million business (seriously… they want to see the possibility).
- Several meetings with some great people.
- Receiving feedback on our reports and the product.
- Getting back to coding — feeling like we are progressing the product towards our customers.
- Now have a Tribalytic blog (http://blog.tribalytic.com) and Twitter account (http://twitter.com/tribalytic) so please follow them!
- Processing pipeline problems wins hands down for this one.
Goal this week
- Get the next beta of the engine ready for release.
- Keep meeting with people.