Skip to content

Booking.com - Search & Ranking at Scale

Project Summary

Type: Enterprise Role (Senior Engineer)
Focus: Search & Ranking at Scale

Key Features:

  • Search and ranking improvements A/B tested across millions of users
  • Fixed critical data pipeline bug restoring ranking quality for entire platform
  • Migrated Spark workflows from on-premise to cloud infrastructure
  • Worked on core search engine serving millions of daily property searches
  • Improved filtering and relevance for home properties

As a Senior Engineer on Booking.com's Search & Ranking team, I worked on the core search engine that powers property discovery for millions of daily users globally. The challenge: continuously optimize ranking algorithms to improve user experience while maintaining data pipeline reliability at massive scale. The solution: rigorous A/B testing, robust data pipeline engineering, and cloud migration for improved scalability and reliability.

The Problem

Serving millions of daily property searches requires constantly optimizing ranking algorithms while maintaining data pipeline reliability at massive scale:

  • Millions of users depend on accurate, relevant search results every day
  • Ranking quality directly impacts business metrics—poor results mean lost bookings
  • Data pipeline reliability critical—any downtime or data quality issues degrade search quality
  • A/B testing at scale needed to validate improvements without risking user experience
  • Legacy infrastructure limitations require migration to cloud for better scalability
  • Complex data workflows processing massive datasets for feature engineering and model training

A single bug in the data pipeline can cause ranking quality to degrade across the entire platform, affecting millions of users and millions in revenue.

Architecture

flowchart TB
    subgraph search [Search Flow]
        query[User Query<br/>Millions Daily]
        search_engine[Search Engine<br/>Property Discovery]
        ranking_model[Ranking Model<br/>Relevance Scoring]
        ab_testing[A/B Testing Layer<br/>Experiment Framework]
        results[Search Results<br/>Optimized Rankings]
    end

    subgraph data_pipeline [Data Pipeline]
        raw_data[Raw Data<br/>Property Listings<br/>User Behavior]
        spark_processing[Spark Processing<br/>Feature Engineering]
        feature_store[(Feature Store<br/>Model Features)]
        model_training[Model Training<br/>Ranking Algorithms]
    end

    subgraph infrastructure [Infrastructure]
        on_premise[On-Premise<br/>Legacy Systems]
        cloud[Cloud Infrastructure<br/>Scalable & Reliable]
    end

    query --> search_engine
    search_engine --> ranking_model
    ranking_model --> ab_testing
    ab_testing --> results

    raw_data --> spark_processing
    spark_processing --> feature_store
    feature_store --> ranking_model
    feature_store --> model_training
    model_training --> ranking_model

    spark_processing -.->|Migration| cloud
    on_premise -.->|Migrated| cloud

Technical Approach

Search & Ranking Optimization

Worked on improving the core search and ranking algorithms that determine which properties users see:

  • A/B tested search improvements for home properties, improving filtering and relevance
  • Analyzed user behavior data to understand what makes results relevant
  • Iterated on ranking signals to better match user intent with property listings
  • Optimized filtering logic to reduce noise and improve result quality
  • Measured impact through rigorous experimentation and metrics analysis

The A/B testing framework allowed us to validate improvements safely, rolling out changes gradually and measuring impact on key metrics before full deployment.

Data Pipeline Engineering

Identified and fixed a critical Spark workflow bug that was causing an unrated properties surge—restoring quality ranking predictions across the platform:

  • Root cause analysis of data pipeline issues affecting ranking quality
  • Debugged complex Spark workflows processing massive datasets
  • Fixed data quality issues that were causing properties to be incorrectly rated or excluded
  • Restored ranking quality for the entire platform by fixing upstream data problems
  • Improved monitoring to catch similar issues earlier in the future

This bug fix had immediate impact—restoring proper ranking quality meant users saw more relevant results, directly improving user experience and business metrics.

A/B Testing at Scale

Designed and executed A/B tests across millions of users:

  • Experiment design ensuring statistical significance while minimizing risk
  • Traffic allocation strategies to test improvements safely
  • Metrics tracking across user engagement, booking conversion, and revenue
  • Rollout strategies for gradual deployment based on test results
  • Analysis frameworks to understand why changes worked or didn't work

The A/B testing infrastructure enabled rapid iteration on search improvements while maintaining platform stability and user experience.

Cloud Migration

Migrated on-premise Spark workflows to cloud infrastructure:

  • Assessed legacy systems to understand dependencies and migration requirements
  • Designed cloud architecture for improved scalability and reliability
  • Migrated Spark workflows with zero downtime
  • Improved system reliability through cloud-native features (auto-scaling, managed services)
  • Reduced operational overhead by leveraging cloud infrastructure management

The migration improved system reliability and scalability while reducing operational burden, enabling the team to focus on building features rather than managing infrastructure.

Results

Metric Impact
Search improvements A/B tested across millions of users with measurable improvements
Data pipeline bug fix Restored ranking quality for entire platform
Cloud migration Improved reliability and scalability of Spark workflows
Home property search Improved filtering and relevance through targeted optimizations
Platform scale Millions of daily searches served reliably

Working at one of the world's most data-driven companies, every change was measured and validated through rigorous experimentation. The improvements I contributed to directly impacted millions of users and millions in revenue.

Tech Stack

Java Apache Spark Cloud Infrastructure A/B Testing Data Pipelines

Key Learnings

At Booking.com's scale, small improvements to search relevance compound across millions of users and millions in revenue. The most impactful work wasn't just building new features—it was fixing critical bugs that degraded user experience across the entire platform. The data pipeline bug fix taught me that ranking quality is only as good as the data feeding it. Cloud migration showed how infrastructure improvements enable faster feature development and better reliability. The rigorous A/B testing culture ensured that every change was validated with real user data before deployment—a practice that should be standard at any data-driven company.

  • Need help with search, ranking, or data pipelines at scale?

    I help companies build scalable search systems and reliable data pipelines. Let's discuss your challenges.

    Book Free Intro Call