With the specific objective of optimizing ABInBev’s retail sales, the crew of
four put their minds to work at the 36 hour (spanning two nights) event
held at ABInBev’s Global Analytics Center.
Starting off with a deep-dive into the datasets provided, the team formulated a plan to build a tool to simulate retail sales outcomes given various conditions.
With Abishek and Nagappan performing data discovery and inspection of the provided datasets, Elayabharath and I set off on the design of interfaces and expected visualizations for the dashboard. A few hours later, with a better idea of our hack, each of us then devoted some time (well spent, I should add) preparing our development environments. However, with age finally catching up on us confident of our progress, we decided to call it a night and went back home, embarrassingly early on.
Fresh and early the next day, putting our collective data science skills at
work, much progress was made on modeling the given time-series sales data. Given the limited duration of the hackathon, we decided to roll with
sklearn for our analyses and React for our web-app dashboards.
This hackathon too, was not without its emotional twists and
turns — it was with less than 12 hours remaining that we found
yet-another-bug™. It had misled us to think that our model was over 98% accurate, while it was only somewhere around the 80% mark. A few panicked hours later, all was well again as we had a stable version up, which we promptly froze as a rollback.
Mentions need to be made here about AngelHack and the ABInBev mentors — who were available for consultations throughout. Their clarifications about the datasets and specific pointers on making a successful pitch provided valuable insight on how our hack was to be judged.
By the end of the second night, with less than 8 hours for final submissions, we decided to freeze our releases and begin working on our pitch. Given the format of a time-constrained two-minute (?!) pitch, we decided to go with only one speaker and no slide-deck. Our pitch would just be one continuous speech as we walk through a demo of our dashboard.
After getting some more sleep in, we finalized our submission and walked into the preliminary rounds, confident of our work. A few more pitches and about an hour later, we were called on to the main stage as the first finalist to present before the entire audience.
Looking back, it is still difficult to see where our two minutes went. By that time, we had practiced our pitched numerous times, that most of it was a blur as we fit our pitch to almost exactly those 120 seconds.
All these efforts did pay off finally, as team Sömma managed to clinch first-place and emerge as overall champions for the hackathon.
Aiming to build a useful data-driven product, while delivering a friendly user-experience proved to be the dual linchpins behind our victory.
For those interested in more details, here is a link to our code together with a few screenshots of the dashboard that we ended up building.
With one resounding win now under our belt, the team now looks forward to more hackathons in the future. Ohh, and our NYC trip too ;).