Sameh Sharaf

Led the migration of 100+ production ETL pipelines from legacy Oracle and Exasol systems to a modern data lakehouse architecture on AWS and Databricks. Introduced right-sized cluster management — replacing a one-size-fits-all approach — achieving up to 35% faster pipeline execution and direct reductions in cloud infrastructure costs. Implemented comprehensive monitoring, fault handling, and data quality governance throughout.

Proposed and architected a unified cloud infrastructure on Azure and Databricks to consolidate data processing and AI model development across a multinational consumer goods company. The solution was adopted globally, with the company's engineering teams worldwide implementing it across their respective tech departments. Collaborated with data scientists to deploy and productionise demand forecasting models, making them reliable and scalable in production.

Built a near real-time data streaming pipeline processing events emitted by mobile devices, enabling instant monitoring of activity by demographics and region using attributes encoded in each event. The processed stream fed directly into an AI clustering model to detect and track suspicious activity patterns — delivering an operational intelligence capability the client had no visibility into before.

Fully migrated a retail client's data infrastructure from a legacy OLAP data cube model to a modern data lake architecture on Microsoft Azure. The new platform replaced rigid, hard-to-maintain cube structures with a flexible, scalable foundation — enabling faster analytics iteration and unlocking new data use cases for the business.

Led end-to-end data infrastructure handling petabytes of data across real-time streaming and batch pipelines for a high-growth media streaming platform. Designed and administered an AWS Redshift data warehouse with significant query performance improvements, and managed the full Qubole platform — including Hive, Spark, and Airflow clusters — while maintaining data governance and quality standards at scale.

Built an end-to-end subtitle generation pipeline integrating HuggingFace speech-to-text and emotion recognition models to automatically transcribe podcast audio across multiple European languages. Generated emotionally-aware subtitles by mapping recognised emotional tone to text-to-speech output for localised, expressive delivery.

Designed and built an internal data quality service adopted by analysts and integrated across data pipelines. Provided automated summary statistics and configurable validation rules via a publish-subscribe architecture, exposing REST APIs and a Web UI for self-serve data quality monitoring across the organisation.