Machine Learning Platform

Machine learning Platform for Training, Validation, and Serving

XenonStack is a leading machine learning platform building company that builds products based on machine learning/ artificial intelligence. Our client wanted a product where he can automate the process complete process of machine learning from model building to model deployment in production. We were asked to automate and unite the following operation

  • Building a machine learning model
  • Integrating machine learning model with the existing data pipelines
  • Training model over a large data set
  • Versioning of ml/dl models
  • A/B testing of machine learning models
  • Analyzing the performance of various models
  • Serving machine learning models
  • Continuous deployment of machine learning models

Challenges for Building Machine learning Platform

We needed to build a platform that can build, version, validate and serve machine learning models.

A platform where :

  • A data scientist can build a machine learning model, version and train models
  • A data engineer can integrate the model with the existing data pipelines
  • An analyst can visualize the data generated from the machine learning model
  • A testing team can perform A/B testing
  • We can establish a standardized platform that enables cross-company sharing of features data and components
  • We can “Make it easy to do the right thing” (ex: consistent training/streaming/scoring logic)

Other common issues

  • No consistency between ML Workflows
  • New teams struggle to begin using ML
  • Existing ML workflows are slow, fragmented and brittle

Solution Offered to Build ML Platform

Akira-ai a complete platform where we can perform model building, validation, versioning, serving and deployment of machine learning models.Our solution has been inspired by Uber’s Michelangelo and Netflix’s meson project.

Features of Akira AI-

  • Distributed training of machine learning models over big data
  • Model versioning
  • Machine learning model analytics
  • Model validation
  • Model visualization
  • Model impact analysis
  • Model comparison to find out which model is best suited
  • Model serving in production and sandbox environments
Read more