By Fred Calvez
AI Ukraine – Come for the Talks, Stay for Vareniki!By Andrew Fiorillo
The AI Ukraine conference had its fifth annual gathering at the end of September 2019, and Delivery Hero was out in full force to attract some of the Ukraine’s brightest data scientists. This year Delivery Hero (DH) joined the conference organisers as a sponsoring partner, hosted a booth in the main annex, and brought a pair of speakers from our data science team to share the lessons of scale.
The conference, organised by AltexSoft, has historically been hosted in Kharkov, but for the first time was moved to the capital, down the street from Kiev’s monumental St Andrew’s Cathedral.
The conference attracted over 1000 attendees and offered three presentation streams: Data Science and Machine Learning, AI for Business and Products, and Big Data and Data Analytics. The talks contained a mixture of nitty gritty technical discussions (like highlighting the issues with using generative adversarial deep learning networks, GANs, to generate sounds from scratch) and higher-level business issues (such as the continuing problems that unprepared organisations have bringing ML-based products to market).
One of the technical talks was given by Kateryna Khotkevych, a Software Engineer at DH. The talk was a case study into Delivery Hero’s streaming architecture, and how her team scales to meet exponential growth.On the top floor of the conference center, she laid out the centralised streaming service that DH uses to collect and curate data to empower teams in all parts of our organisation. Built mostly on AWS, the ‘DataFridge’ team has been able to optimise costs and minimise latency while also handling (a continuously increasing) 10 million order-related events on a daily basis.
The organisational aspects of Delivery Hero also got attention in both of the talks. One challenge that comes with the “house of brands” structure is complicated when introducing global services. Simply put, when a new global service is introduced (such as our customer segmentation tooling) it must integrate with the 20+ brands in the DH family — an operational feat. The DataFridge team addresses this problem head-on by standing in the middle between data producers and data consumers, keeping the data clean and fresh. Watch Kateryna’s full talk here.
Yann Landrin-Schweitzer, the Global Director for Data & Machine Learning at Delivery Hero, gave a keynote talk about the not-so-sexy but absolutely vital parts of a data centric organisation. Titled Operationalizing Data & Machine Learning, Yann pointed out that while the position of data scientist is widely regarded as the job of the century, the actual work that goes into it isn’t all deep learning and cutting edge research. With alarming consistency, data scientists report that the majority of their time is spent acquiring and cleaning data. Developing a proper operational perspective for the availability and clarity of data is essential to using it.
According to Yann: “You really need to think of data as an asset. It’s something that is valuable. Just as valuable as your machines, your engineers, your intellectual property … Align everybody on ‘yeah, data matters.’”
Watch Yann’s entire talk below.
On the whole, the team that travelled to Kiev for the conference was pleased with the turnout. Although we don’t yet operate in Ukraine, there was considerable interest from attendees in what we’re doing and in particular, what our data scientists are working on.
As always, if you’re interested in pursuing a career in Data & Machine Learning at Delivery Hero, have a look at some of our open positions: