Wingman

Inspiration

Aviation is a very complex business with countless variables. Finnair alone averages 300 flights with 30 000 passengers a day flying all over the world. The Operations Control Center (OCC) - the heart of Finnair's operations - is an invaluable part of keeping all moving parts in control. Unfortunately the operating system used to track everything essential requires an immense amount of manual work and expertise, which includes searching relevant information from the web and reviewing data from multiple sources. Wingman is here to help. Everyone needs a good Wingman.

What Wingman does

Wingman helps the operator of Finnair's OCC dashboard by offering several different automated data inputs and machine learning utilizing components. Wingman is tapped onto several sources of data, such as different social media and news platforms, and generates relevant notifications to the OCC-operator. These notifications (we call them 'wings') are then reviewed by the operator and are either accepted to be part of the utilized data flow or are discarded.

Simply put, Wingman enables the operator to focus on the essentials of Finnair's operations controls by making the workflow more efficient.

How is Wingman built

The Wingman is a web-app built on React, which utilizes several data inputs such as FMI-weather data and the Twitter API with its feed from dedicated accounts. This data is processed into relevant event notifications (wings) of categories defined by the OCC-model. During the weekend we focused on two categories: weather and strikes (of employees).

Our weather data is processed through a ML-model to provide notifications about forecasts that might affect operated flights negatively. The model is based on a random-forest-classifier and taught using the data from FMI. The strike-module follows several related twitter accounts, such as the labor unions, scrapes the required information and provides updated notifications of topical events.

The final touch is the sleek UI, which presents the generated wings to the operator to approve. If approved, the wing is listed as an event and the OCC dashboard is updated.

Challenges we ran into

The problem or challenge we identified was relatively simple but rather fundamental. This caused us to rethink our approach several times. Technical problems (if you don't include the "normal" hacking itself) mostly orbited around the insufficient FMI weather data to actually train the ML-model used. We ended up writing some hacky code, but most importantly we made it work.

Accomplishments that we're proud of

We were able to identify a solvable problem, narrow it down to a manageable hack and parse all the components together. Our data and machine learning pipeline actually works and the system architecture is feasible, not just smoke and mirrors.

What we learned

Everyone needs a Wingman!

What's next for Wingman

The first step would be broadening the scope of collected data to other OCC categories deemed worthy. Secondly we would make the different modules and the information they provide more trustworthy by implementing machine learning components. The interesting part is that both the twitter-data-scraper and the used ML-model are reusable and can be used for implementations of other modules with relatively small effort. A third clear step would be tweaking the existing user interface of the OCC and its traffic lights by adding more data visualization such as a view with ongoing flights with updated statuses. The sky is the limit.