Tutu

How to Build a Recommendation System in TravelTech in 5 Weeks: Kafka and MongoDB Instead of Feature Store

An MLOps engineer from Tutu describes how a recommendation system was built from scratch in five weeks using Kafka, MongoDB, and ClickHouse, deliberately avoiding a Feature Store for speed. The system suggests hotels after a transport ticket purchase, using real-time order data from Kafka and nightly batch features from ClickHouse, all stored in MongoDB for fast inference.

Tutu