I gave a presentation about Kusto Query Language (KQL) within Microsoft for a Learning Day held by my division, and then gave the talk again for an internal presentation series about data-driven methods. I've now published the recording on YouTube so that a wider audience can benefit from it as Kusto keeps on gaining popularity.

You can watch it here:

Kusto is a really fun and fast query language for analyzing and visualizing log data. I hope that people will enjoy it as much as I did.

Update May 2022:

The video has seen a steady flow of about 260 views per month since it was released a year ago. 47% of the view traffic originated from search within YouTube, 14% from suggested videos on YouTube, and 12.5% from external sites (mostly Google Search). 26.5% from other YouTube sources.

I've also recently transcribed the video to add accurate closed captions.

In Google Code Jam this year, there was a problem called "Cheating Detection" that was very different from the usual algorithm-based problems. It was more data-driven: it involved finding the cheater among a set of students, given only the scores of all students in each of the questions in an exam, and the mechanism of cheating that the cheater used (cheating only 50% of the time).

After the round was over, I went to see the solutions of the top contestants, to see what sorts of methods they used to obtain a solution. The solution had to detect the cheater with very high probability, and also needed to be a fast enough program to type that it would be among the first submitted. I found the top solution to be surprisingly simple in its code, so I tried to figure out a mathematical explanation for why it worked. I ended up making a video with this explanation.

You can watch the video here: