Scaling machine learning at Careem

Description

Every day, Careem, the Middle East’s first tech unicorn platform, enables millions of users across 120+ cities to get a safe ride from point A to point B as well as getting food and deliveries on time. The company had to solve a lot of challenges to ensure the delivery of optimized and localized experiences across these domains including choosing the right driver for a booking, estimating the arrival time of the booked car or order, demand and supply forecasting over time for different locations, balancing supply and demand of vehicles across time and locations, and up-front price predictions.

Due to the complexity of the business and the fast-paced nature of its dynamics, thousands of production-level ML models have to be developed, served, and continuously refreshed. Ahmed Kamal explains how Careem is building a scalable and flexible ML platform that facilitates development and enables expeditious deployment of scalable ML services in a cost-efficient way, including examples of ML use cases at Careem currently productionized on top of an ML platform; how to balance between the needs of different users (researchers, data scientists, business users, etc.); how to productionize different ML stages, including dataset generation, model selection, training, tuning and testing, model evaluation, tracking and benchmarking, model serving, and performance monitoring and integration; and the role of ML platforms and AutoML in fostering, democratizing, and accelerating the impact of ML across the organization.