MLOps-enabled bottle manufacturing platform

The changing needs of the market demand upgrading existing systems. The client's existing AI implementation was disconnected from the data ingested from the edge device into the AWS storage bucket; a partial model training pipeline was on the Azure cloud platform and the remaining was on-premise. The client needed to streamline the workflows and perform troubleshooting for edge scenarios. Additionally, a commercial model training platform license had to be attached to the ML solution, which was a manual process. Lastly, the pipeline had to be built as a multi-tenant platform to be used by the client’s customers.

Nagarro partnered with the client to build a custom, MLOps-enabled AI workflow for deployment on AWS. The solution was designed to be serverless, with the computing cost generated on a pay-as-you-go model. Data pre-processing was done on Spark EMR cluster and multiple model training was performed on AWS SageMaker. Currently, deployment on edge device (using SageMaker Neo, Edge Manager and IoT Greengrass) is in progress. In the last phase, data ingestion and deployment will be performed on the cloud.

Nagarro’s serverless MLOps implementation of the client’s AI model has provided the client full control of the various AI steps in the end-to-end pipeline. The custom-tailored pipeline components are configurable for the solution requirements. The defined multi-tenant application comprising of MLOps and DevOps best practices and infra-as-code schemes have allowed the client to evaluate the total cost of ownership of the platform and thereby, define the business strategies for the solution to be shared with end customers.