AWS re:Invent 2025 - Build an AI-ready data foundation (ANT304)

Building an AI-Ready Data Foundation on AWS

Defining Data Foundation for Data Analytics

• Data foundation for data analytics refers to having access to data tools that enable:
  • Data storage
  • Data processing
  • Data integration
  • Data governance
• These foundational constructs come together to form a typical data pipeline
• The key is being able to leverage this existing data foundation to build AI applications

Evolving Data Needs for AI Applications

• AI applications have evolved from simple rule-based assistants to autonomous multi-agent systems
• This evolution impacts the data pipeline in several ways:
  • Need for additional unstructured data sources
  • Advanced data processing requirements for training, inference, and context management
  • Capturing real-time user feedback and personalization
  • Comprehensive data governance across the entire pipeline

Leveraging the AI-Ready Data Foundation on AWS

AWS MCP Servers

• MCP (Model-Connected Prompts) servers provide a conversational layer between AI models and tools
• Enables true agent autonomy by allowing the model to determine what tools to use
• Replaces the need for custom API integrations
• Examples used in the demo:
  • Redshift MCP server for natural language access to Redshift data
  • S3 Tables MCP server for querying tabular data in S3

Context Management

• LLMs have limited context windows and are stateless, requiring careful context management
• Context includes instruction sets, knowledge bases, and data from tools/MCP servers
• Agentic memory using services like Amazon Bedrock Agent Core can retain context
• Knowledge bases using Amazon OpenSearch as a vector store provide semantic search capabilities

Event-Driven Architectures for Multi-Agent Systems

• Multiple loosely coupled agents working asynchronously can improve resiliency and scalability
• Amazon MSK (Managed Streaming for Apache Kafka) enables event-driven architectures with real-time data exchange
• Allows agents to respond to patterns and scenario changes independently

Ensuring Data Readiness for AI

Metadata and Data Governance

• Metadata is critical for AI agents to understand and utilize data effectively
• SageMaker provides automated metadata generation, data quality scoring, and lineage tracking
• Metadata can be published back to S3 for agents to directly access

SageMaker Unified Studio

• Provides a single interface for data engineers, scientists, and analysts to work with data and AI
• Includes features like one-click onboarding of data sets and a polyglot notebook with built-in data agents
• Data agents can automatically generate code, fix errors, and provide visualizations

Key Takeaways

• Existing data foundation on AWS can be leveraged to build AI-ready applications
• Critical components include MCP servers, context management, event-driven architectures, and data readiness
• SageMaker provides a unified platform to manage data, models, and AI agents
• Automated metadata generation and data governance are crucial for enabling AI agents to effectively utilize data