Lecture: Next Generation Open Machine Learning Operations with Practicus AI

Introduction

• Practicus AI provides solutions for modern machine learning operations (MLOps).
• It leverages open source cloud-native technology to avoid vendor lock-in.

Key Concepts

Operational Requirements

• Successful AI systems require diverse operational needs.
• Different teams have distinct expertise and tool requirements.
• Use of multiple data sources across various clouds and on-premises locations.
• Utilization of modern data mesh architectures for unified user experience and federated governance.

Deployment & Consumption

• AI models can be deployed and consumed using Practicus AI:
  • Multiple personas can choose different deployment methods.
  • The infrastructure as code is open, well-documented, and repeatable.

User Experience

• Example: Load data from a data lake and build a model to predict customer churn using AutoML.
• Models can be deployed as APIs for immediate consumption by others.
• Supports exporting AutoML code to Jupyter for modifications or custom models.

Technical Implementation

Model Management

• Practicus AI uses model prefixes, names, and versions as logical elements.
  • Logical elements are tied to Kubernetes deployments.
  • Allows for dynamic service mesh setup.
• Multiple versions can be deployed simultaneously for A/B testing.
• Management of hundreds of models and versions daily is supported.

Demo: Web Admin UI

• Steps for hosting models:
  • Click on model hosting and add a new physical deployment.
  • Define CPU, RAM, and optionally activate auto-scaling.
  • Grant access to groups or individual users.
  • Create model prefixes, which define URLs and production settings.
  • Define models and versions with deployment, stage, and traffic weight settings.

Advanced Scenarios

• Practicus AI in multi-location deployment:
  • Supports different public and private clouds with varied data sources.
  • Single interface for business users to explore systems and data sources.
  • Developers use fine-grained access control tokens.
• Global APIs for high availability with automatic traffic routing during failures.

Open Source Benefits

• System operates continuously even after uninstalling Practicus AI.
• Legacy/proprietary systems can be modernized by wrapping with Practicus AI MLOps.

Summary

• Practicus AI offers a comprehensive solution for MLOps with open-source technology.
• Enables modernization, seamless deployment, and consumption of AI models.

Contact

• Questions and further inquiries are welcome.