Machine Learning Operations (MLOps) - 4 Weeks PoC

LTI (L&T Infotech)

Model Operations templates to Operationalize Machine Learning models on Azure

LTI AInA, our template-based approach to operationalize the AI/ML models on the Azure platform. AInA leverages pre-built templates to automate the model operationalization process and enable Bring Your Own Model (BYOM) concept when models are developed in other languages and frameworks. Leveraging the Azure cloud solutions, AInA helps deliver time to market and scalability by simplifying the model deployment process. The 4-week proof of concept covering end to end operationalization of one model will help delivery how to build the foundation and then scaling the deployment process using Azure cloud components. Some of the key Azure components included as part of AInA templates are:

Azure DevOps |Azure ML | Azure Kubernetes | Azure Blob Storage | Azure Insights | Azure Monitor The key AI engineering components constituting AInA to help scale the AI/ML projects on Azure data platform are:

Model Deployment: End-to-end model ops template for productionizing ML models using DevOps principles

Infrastructure-as-a-Code: Automate resource provision for training, scoring, etc. leveraging pre-built ARM templates

Model Packaging & Registry: Serialization of models and centralized model registry for AI/ML models

Operational Insights: Setting up of model monitoring to track model drift, data drift, and service health of the models

Model Building is out of scope for this PoC. The scope covers model deployment and operationalization for the models built in Azure or Python or R programming languages with all the best practices.

Our 4-week POC engagement plan includes operationalization of one AI/ML model as explained below: Week 1:

  1. Understanding the current landscape, along with understanding the models, current model operationalization process, and the technology landscape.
  2. Setting up Azure ML and Azure DevOps environment along with fine-tuning the key parameters and selection of the model which will be operationalized.

Week 2 and 3:

  1. Migration of codebase to Azure DevOps Repo of the selected model.
  2. Modularize code and integrate end-to-end machine learning pipelines with Azure ML workspace, using our pre-built AInA templates.

Week 4:

  1. Configure Azure DevOps release pipeline for batch scoring process or model as a service via APIs and setup of automated retraining to showcase the overall model operationalization for the selected model.

The Outcome:

  • Reduction in time to market by accelerating model deployment process.
  • Standardize and streamline the AI/ML lifecycle to provide scalability along with re-usability in operationalizing the models.
  • Focus on automation with DevOps pipeline.

Other consulting services from LTI (L&T Infotech)