Writing on machine learning, engineering, and data science.
A technical deep dive into how GetYourGuide migrated its activity ranking system from tree-based models to transformer-based deep learning models, including the phased experimentation strategy, architecture insights, and lessons learned.
DDataflow is an open-source tool for end-to-end testing of machine learning pipelines. It addresses the challenge of slow feedback loops in ML development by providing efficient, configuration-based sampling of production data, enabling data science teams to iterate quickly and catch integration bugs early.
A detailed look at how GetYourGuide evolved its MLOps workflows from batch-oriented systems to production-ready real-time model serving, with ownership boundaries, automated CI workflows, and CI/CD practices highlighted.
Describes how the GetYourGuide ML platform was enhanced from batch inference to support real-time model inference, including the motivations, architectural updates, and tools used to serve models efficiently.
Outlines the foundational work behind GetYourGuide’s open-source machine learning platform, focusing on establishing a robust CI/CD pipeline and best engineering practices with tools like MLflow, Drone CI, and Airflow.
Interview from plista about privacy-aware advertising and a new widget format designed to respond to cookie fatigue in ad tech. The piece reflects early product and machine learning work around recommendation surfaces, tracking constraints, and user engagement.
