AWS re:Invent 2025 - Supercharge app intelligence using gen AI with Amazon DocumentDB (DAT313)

Supercharging App Intelligence with Generative AI and Amazon DocumentDB

Introduction to Generative AI

• Generative AI is a new and disruptive technology that is becoming an enterprise-wide initiative
• Mackenzie study estimates Generative AI will add $2.6 to $4.4 trillion to the global economy annually
• The key differentiator for Generative AI applications is the data - your data is what makes these applications unique to your business

Leveraging Existing Data with Amazon DocumentDB

• Customers want to build Generative AI applications on top of their existing data infrastructure, not create new data architectures
• Keeping vector embeddings close to the underlying data store in Amazon DocumentDB simplifies the architecture and improves performance
• Using vector search capabilities in Amazon DocumentDB allows you to leverage your existing knowledge and skills, rather than introducing new database components

Vector Embeddings and Text Representation

• Vector embeddings are a numerical representation of text that allows machines to understand context and meaning, not just keywords
• This enables natural language queries, conversational search, and other advanced use cases beyond simple keyword matching

Access Patterns for Generative AI with Amazon DocumentDB

1. TSQL Plugin for MongoDB Shell:

   • Allows relational database developers to interact with Amazon DocumentDB using familiar TSQL syntax
   • Translates TSQL commands into MongoDB Query Language (MQL) commands
   • Supports a wide range of APIs, operators, and data types

2. Retrieval Augmented Generation (RAG) Architecture:

   • Combines vector similarity search against the data store with fact lookups to provide contextual and accurate responses
   • Example: Chatbot that looks up order details and policy information to handle a return request

3. Model Context Protocol (MCP) Servers:

   • Provides a standardized way to access capabilities and data sources through a client-server architecture
   • Allows developers to explore data, generate code, and get recommendations without deep expertise in the underlying database

Best Practices for Vector Indexes in Amazon DocumentDB

• Choose between IVF Flat and HNSW index types based on your requirements:
  • IVF Flat: Faster index building, smaller index size, but requires pre-populated data
  • HNSW: Slower index building, larger index size, but better performance and flexibility
• Consider query performance requirements, recall rates, and ingestion time when selecting index settings
• Experiment early and commit to using vector embeddings and search - your data is the key differentiator for Generative AI applications

Business Impact and Use Cases

• Customers are building Generative AI proofs-of-concept and rolling out enterprise applications to improve customer experience, employee productivity, content creation, and business operations
• Example use cases include chatbots, virtual assistants, conversational search, and code generation

Conclusion

• Integrating Generative AI capabilities with your existing data in Amazon DocumentDB can unlock new opportunities to enhance your applications and business processes
• The key is leveraging your unique data assets and the vector search capabilities of Amazon DocumentDB to build differentiated Generative AI solutions