Di AI Engine Wey Dey Run Inside Every Uber Ride

Na 5:00 PM for Friday evening for Sydney CBD.

Office dem dey empty within minutes. Restaurant dem full. 40,000 people open dia Uber app almost di same time. Rain just start to fall.

Di question wey Uber must answer right now simple well well. How many ride request go come inside di next 15 minutes?

If e wrong, di wahala na immediate. Long wait. Cancellation. Customer dem dey switch go competitor. For Uber level, small mistake for prediction no be rounding matter. Na revenue matter wey dem dey measure for millions.

Most people think say Uber dey work anyhow e come. Rider go open app and di system go begin rush find driver wey dey near. No be so e dey work.

Uber AI system dem dey run demand forecast non-stop, 24 hours every day. By di time you open di app on Friday evening, Uber don already know say you dey come. Di driver wey go carry you fit don reposition 20 minutes before because of one forecast wey predict demand for dat exact place for dat exact time.

Dat kain ability no dey built on top guesswork. E dey built on top mathematics.

1 percent better for forecast accuracy for Uber volume na hundreds of millions of dollars extra money every year. No be from one big change, but from thousands of small inefficiency wey dem remove di same time, wey dey add up every hour for every city.

Dis na Part 1 of 5-part series on how Uber dey predict di future before e happen.

Part 2 cover di three-model ensemble architecture. Part 3 waka through every formula with real numbers from one Friday evening for Sydney. Part 4 build di full implementation from scratch with Python. Part 5 show di production PyTorch blueprint wey dey run for big scale.

Every calculation don verify. Every line of code dey run.

Follow dey go. Di mathematics easy pass wetin most people dey expect.

#MachineLearning #ArtificialIntelligence #AI #DataScience #Mathematics #Uber #TimeSeriesForecasting #DemandForecasting #AIEngineering #TechLeadership #TheMagicOfHiddenMathematicsInAI #PredictiveAnalytics #DeepLearning #MLOps #Python

Originally published on LinkedIn by Keshavan Thiagaraja. Read the original on LinkedIn.