• Worked on the YouTube team that classifies abusive content using Deep Learning

• Designed and implemented a data pipeline to achieve a 4x performance improvement for reading features required for classification

• Improved the appeal process of an abusive website by providing different versions of features

• Enabled parallel loading of features by moving to new backend system

• Testing the system extensively using unit tests and integration tests

• Technologies Used: C++, Python, Piper, Blaze, Google Cloud Platform, Protobuffers