CUSTOMER STORY | MakeMyTrip Achieved Millisecond Personalization at Scale with Databricks RTM

[Tushar_Parekar]

"We wanted one source of truth -- one Spark-based pipeline -- rather than two engines to maintain. Real-Time Mode gave us the performance we needed with the simplicity we wanted."
  – Aditya Kumar, Associate Director of Engineering, MakeMyTrip

MakeMyTrip, India's top online travel agency, bypassed traditional micro-batch bottlenecks by switching to Databricks Real-Time Mode (RTM) on Spark. This unified their streaming setup for super-fast personalization, boosting clicks by 7% and simplifying ops for millions of daily users.

Key highlights:

• Sub-50ms latencies: Dropped P50 latency from 1.23 seconds to 44ms for "last-searched" hotels, serving personalized results instantly.
• 7% CTR uplift: Real-time processing of high-volume searches drove higher user engagement and click-through rates.
• Unified Spark architecture: No need for Flink or dual engines same business logic works for batch and real-time, cutting complexity.
• Cost-effective scaling: Handled 64 Kafka partitions with fewer cores via task multiplexing, keeping infra lean.
• Production-ready reliability: Added stream union, backpressure, and fault tolerance through Databricks collaboration.
• AI-ready future: Feeds real-time context to generative AI agents for smarter decisions.

Want millisecond personalization without fragmented systems?
Learn more about Real-Time Mode and power your real-time apps.

Read the Full Story Here 👈