Eurowings Delivers Real-Time Travel Insights with Databricks

Overview

Eurowings, a Lufthansa Group subsidiary, migrated to a unified data platform called Minerva, built on Databricks with a medallion architecture, to replace siloed data systems. It implemented semantic search on eurowings.com using Databricks AI Search and Model Serving, replacing keyword-based search. The change delivered a 109% lift in search relevance, improved recall from 50% to over 75%, and cut search latency from six seconds to under two seconds. MLflow is used to manage the model lifecycle, with Model Serving acting as a single endpoint for testing different embedding models.

The challenge

Before unifying its data strategy, Eurowings faced significant challenges with siloed data across disparate systems. The previous environment lacked the connectivity required to implement advanced AI use cases, leaving the data and AI team unable to provide integrated experiences.

The solution

Eurowings migrated to a new platform, Minerva, built on Databricks, creating a unified data lake using a medallion architecture. To power semantic search, the company implemented AI Search and Model Serving to handle real-time inference, using MLflow to manage the model lifecycle and Model Serving as a single endpoint that connects to its applications, providing flexibility to test different embedding models.

Reported business value

Eurowings achieved a 109% lift in search relevance, improved recall from 50% to over 75%, and reduced search latency from six seconds to under two seconds, cutting response time by more than 60% for live traveler queries.