The existing system wasn’t scalable and the absence of standard processes to access client resources like data sources, cloud machines, cluster environments, etc. made it challenging to create a production-ready system.
Created a parallel set-up in the cloud and used Google Composer to manage workflow orchestration, ensuring the system has minimum manual intervention and better fault tolerance.
Our robust and scalable solution automated pipeline for model training and increased coverage for new countries, products and categories. This ensured the system has minimum manual Intervention and is easy to debug, extend and modularize, resulting in low maintenance and better fault tolerance.