By Fred Calvez
Hacking the DeliveryHackBy Andrew Fiorillo
On Thursday, March 7th we marked the end of the second annual Delivery Hero hackathon: DeliveryHack 2019. With nearly 90 participants divided into over 20 teams, it was our largest internal hackathon to date. To get a better impression of DeliveryHack, have a look at our video.
On the first day of DeliveryHack 2019’s ‘science fair-like’ competition, everyone had the opportunity to pitch their project idea and convince other participants to join their team. The teams then had two and a half days to develop their ideas into a prototype and to prepare a final presentation, all focused around five themes:
- the Customer Experience,
- the Rider Experience,
- the Partner Experience,
- the Employee Experience,
- and the Developer Experience.
We also had special categories for the best application for machine learning, the most promising business idea, and the Heroes’ Choice Award (selected in a company-wide poll).
DeliveryHack participants also had the opportunity to join different learning sessions: Johannes Nicolai from Github talked about “How Github uses Github to develop Github”, Csaba Tamas from AWS gave a session about “Unicorns going serverless” and our very own Karolina Skalska gave insightful presentations about Design Thinking.
One team that shined through the week and took away the award for best business idea was named after Cassandra, the mythological princess cursed to see the future. A common problem for any delivery service is when a provider (e.g. restaurant, grocer) is too busy to handle any new orders; the result is one of the worst experiences for any customer — having an order cancelled after getting a confirmation text. The Cassandra team proposed to solve this problem with a binary classifier, running in realtime and for over 150.000 partners on the platform: too busy for new orders, or ready for more.
Across all Delivery Hero brands we processed over 40 million orders last year. The Cassandra project leveraged our database of order histories to train a predictive ensemble model. The result could be integrated into our distributed data architecture and applied globally. After only three days of development, they were able to build the foundations of a production-ready system.
The winning project for the Heroes’ Choice Award was code-named “Hollywood Dashboard.” The prototype included a real-time streaming, three dimensional, hardware-accelerated visualization of real time order data.
They used the beautiful kepler.gl library to generate a bird’s eye view of Foodpanda Singapore that looked to be ‘straight out of Hollywood.’
We are excited to follow up with these two great ideas as well as with many other promising project results from DeliveryHack.
We’ve already collected feedback from many participants and are planning for our next DeliveryHack week. If you’re interested in joining us and participating, take a look at our available jobs!