Intigrated MLFlow into existing systems that serve prediction periodically.
The goal was to be able to detect any change in model accuracy, training data.
Integrate developed systems into Airflow Scheduler to ensure timely prediction service.
Migrate single machine Airflow service to multi-machine Airflow cluster that serve all the ML systems across machines with message passing system RabbitMQ.
Develop ML serving system using Nvidia-Triton to serve real-time ML inference and ensure scalability.