India’s largest online travel agency, MakeMyTrip, has redefined real-time personalization by achieving millisecond-level response times at massive scale using Databricks’ Real-Time Mode (RTM), a next‑generation execution engine within Apache Spark Structured Streaming.
For a platform that serves millions of users daily, speed is critical especially for features like “last‑searched” hotels, where travelers expect instant, highly personalized results as soon as they interact with the search bar. Even minor delays can disrupt the booking journey and impact conversion rates. MakeMyTrip’s engineering team needed sub‑second latency across both consumer and corporate travel businesses without adding architectural complexity.
Traditional micro‑batch processing in Apache Spark proved insufficient, with median latency hovering around 1.23 seconds despite extensive tuning. While alternative streaming engines could meet latency targets, adopting a second platform risked duplicated business logic, higher operational overhead, and long‑term maintenance challenges.
“Millisecond-level responsiveness is no longer optional in digital travel—it directly shapes user engagement.”
Aditya Kumar, Associate Director of Engineering at MakeMyTrip.
Instead, MakeMyTrip chose Databricks’ Real-Time Mode, enabling continuous data processing on the same Spark stack already used for batch workloads. With RTM, the company reduced P50 latency to just 44 milliseconds and P99 latency to roughly 500 milliseconds, while maintaining a single source of truth across pipelines. The improvement translated into a measurable 7% uplift in click‑through rates.
“Our priority was to avoid maintaining two engines and risking logic drift,” said Aditya Kumar, noting that RTM delivered the “performance we needed with the simplicity we wanted.”
The real‑time architecture unifies B2C and B2B clickstream data, performs stateful enrichment using Aerospike to retrieve recent hotel searches, and serves results instantly via Redis. Importantly, enabling RTM required just a single configuration change, allowing teams to transition from batch to real time without rewriting code.
Beyond immediate gains, MakeMyTrip is now exploring RTM as a foundation for its next wave of innovation. The company plans to feed real‑time context into generative AI agents, enabling smarter recommendations and decision‑making across the travel lifecycle.
By compressing infrastructure complexity and delivering ultra‑low latency personalization, MakeMyTrip’s deployment demonstrates how Apache Spark, powered by Databricks RTM, can function as a true real‑time engine supporting both today’s user expectations and tomorrow’s AI‑driven experiences.