Flightivity

Inspiration

Sometimes life is boring and you just want to travel somewhere. Our app recommends you the best destinations, based on your friends history.

What it does

As an input we take the bands you follow from Spotify API and get their concerts from BandsInTown API, We also take geotags from photos of you and your friends (VK API). We also take smartly.io flight dataset to train our model. Using factorisation machine we make recommendations on places you likely wanna visit. And we use FinnAir API to show you tickets to these places.

How we built it

• PyTorch, Implicit, LightFm for ML
• IOS app coded in Swift, designed in Sketch.
• Python3 + Django hosted on google Appengine flexible docker containers with Postgres for db.

Challenges we ran into

In the first place, we planned to use Facebook and Instagram data, but we discovered a lot of limitations (mostly sandbox restrictions), making them not applicable during competition.

Finally we ended up with VK API (largest Russian social network) for PoC. Also we were planning to use Songkick API, but failed to get api-key fast, so we ended up using BandsInTown.

Accomplishments that we're proud of

Fully working, nicely designed prototype, based on cutting-edge tech.

What we learned

How to make a pipeline with stack of deep learning neural networks

What's next for Flightivity

Reinforcement learning with pre-trained DQN for best user experience during the trip

Built With

• deep-learning
• django
• google-app-engine
• machine-learning
• matrix-factorization
• python
• rest
• swift