AWS re:Invent 2025 - SageMaker & MLflow: Innovate faster with no infrastructure management (AIM3340)

Summary of AWS re:Invent 2025 Presentation: SageMaker & MLflow: Innovate Faster with No Infrastructure Management

Introduction to MLflow and AI Governance

• The presentation focuses on the importance of AI governance, which is crucial for ensuring traceability, repeatability, and auditability of machine learning (ML) and generative AI (Genai) models.
• MLflow is an open-source tool that addresses key challenges in AI governance, such as lack of standardized infrastructure, security risks, and the need for explainable and reproducible models.
• The presentation discusses the evolution of MLflow, from the initial self-hosted version to the fully managed MLflow instance in Amazon SageMaker, and the recent launch of the serverless version of MLflow.

Challenges with Self-Hosting MLflow

• Self-hosting and managing an MLflow instance leads to operational complexity, such as frequent version upgrades and the need to build integrations with SageMaker security and the Studio IDE.
• The total cost of ownership of self-hosting increases as the team size and use cases grow, often slowing down the pace of innovation.

Fully Managed MLflow in Amazon SageMaker

• The fully managed MLflow instance in Amazon SageMaker provides the flexibility to choose any IDE, while integrating with purpose-built tools and leveraging the security of the AWS infrastructure.
• The managed MLflow instance allows for seamless integration with other SageMaker capabilities, such as the SageMaker Model Registry and deployment to SageMaker Inference or other target environments.

Serverless MLflow: Addressing Infrastructure Management Challenges

• The newly launched serverless version of MLflow eliminates the need for infrastructure management, as it automatically scales up and down based on demand, and is provided at no additional cost.
• Serverless MLflow also offers automatic version upgrades, ensuring that customers always have access to the latest features and bug fixes.
• Serverless MLflow supports centralized experiment tracking, allowing for cross-team collaboration and visibility across multiple environments (e.g., dev, QA, production).

Benefits for Data Scientists

• Serverless MLflow eliminates the need for data scientists to provision and manage an MLflow tracking server, allowing them to focus on the job of building models.
• MLflow is now integrated with other SageMaker capabilities, such as SageMaker Pipelines and Model Customization, automatically logging relevant metrics and information.
• The seamless integration and serverless nature of MLflow enable rapid experimentation and collaboration, accelerating the time to development and deployment.

Customer Success Story: Intuit's Journey with MLflow

• Intuit, a leading financial technology company, adopted MLflow to address the challenges of unorganized experimentation, lack of reproducibility, and limited visibility for leadership.
• Intuit's journey with MLflow can be divided into two acts:
  1. The age of predictive AI, where MLflow helped Intuit democratize AI and improve developer productivity.
  2. The generative AI revolution, which forced Intuit to reimagine their MLOps platform for scale, leading them to migrate to the managed MLflow solution on AWS.
• Intuit's use of managed MLflow enabled them to build confidence in the quality, accuracy, and safety of their AI agents, which power their innovative product experiences for millions of customers.

Key Takeaways

1. Serverless MLflow simplifies infrastructure management, allowing data scientists to focus on model development.
2. Seamless integration with other SageMaker capabilities, such as Pipelines and Model Customization, accelerates the time to development and deployment.
3. Managed MLflow on AWS provides scalability, availability, and security, enabling Intuit to innovate faster and with confidence in their AI-driven product experiences.