Mobile App · Machine Learning · Korea
Nalla
Public bike-share fails in a specific, predictable way: the docks you want are full when you arrive and empty when you leave. Nalla is a Korean bike-share app built around that problem rather than around the map.
Results
- 01Shipped to the App Store and Play Store for the Korean market
- 02TensorFlow model forecasting dock availability minutes to hours ahead
- 03Region search and filtering across the bike-station network
- 04Statistics module for day-to-day operational monitoring
01
Predict the dock
Every bike app shows what is there now, which is the wrong number: by the time a rider arrives it has changed. Our TensorFlow model forecasts availability minutes to hours ahead.
02
The operator's screen
The rider app is half the system. The React and Django dashboard had to answer operational questions, so we added region filtering and statistics a person can act on mid-shift.
03
Redistribution
Knowing which docks will saturate means moving bikes before the problem instead of after the complaints. The forecast is not a screen feature, it is how vans get routed.
04
Built for Korea
Shipped to both stores in Korean, for a Korean operator and Korean commuting patterns. Station density and rhythm are things you calibrate locally, not from a spec.