My wife and came in 1st place in the HACK category at HackathonCLT 2015. It was an overnight, 24-hour competition and the challenge was to find an innovative way to deliver groceries to customers, without them coming into the store. We competed against 88 other people, analyzing 140 million transactions from the grocery store Harris Teeter. The data spanned across 6 months of activity, all the way down to the customer, product, and transaction level. There were three categories to compete in, and only the winner took home a prize:
- HACK: use data analytics to derive a solution, with a prize of $4,000
- CODE: the traditional hackathon category where you build a web or mobile application to address the solution, with a prize of $2,500
- freeSTYLE: present an idea, no code required, with a prize of $1,000
Rather than focusing on the physical delivery of groceries, we focused on what should be delivered and to who. We tried to reinvent the ice cream truck, but for adults with modern technology. The idea was that you would be at home and get a pop up message on your phone asking whether you wanted a certain product delivered in the next 10 minutes. It could be anything from diapers, bananas, pimento cheese, or OMB Copper. We made it a targetted suggestion to the customer by looking back at their individual transaction history to see how often they buy something and whether they are over-due on buying that item. If you can find a large enough grouping of people that all are out-of-stock on the same product, then there is a good chance that some of them will want it. So the idea is to send out a driver with those out of stock items and see how bites. The actual product we created was an analytic web app that would help Harris Teeter identify which customers to target, and which products to offer them.
The data was sitting in hadoop, and we used hive to access it. My wife gathered data for us, and wrote all of the hive code to access the data. The local data crunching and presentation was written in R. We used dplyr for data transformations, shiny for web dashboards, and either rCharts (NVD3) or ggplot for data visualization. The code is posted on github.com/benporter/hackathonclt2015.
Here's our pitch. If you're having a hard time following my logic, don't worry, it's not you. Trying to explain something technical after running on about 20 mins of sleep from the night before is tough.