AWS re:Invent 2025 - Build real-time full-stack Generative AI applications (DEV310)

Building Real-Time Full-Stack Generative AI Applications

Overview

• Presenters: Si, a Senior Developer Advocate at AWS, and Matt, a Senior Principal Engineer at AWS
• Goal: Build a full-stack application from scratch, including a front-end, deployment, and integration of generative AI features
• Motivation: The presenters wanted to create a platform for exchanging and selling games, similar to Goodreads for books

Leveraging Agents and MCP Servers

• Utilizing Kira, an agent-based IDE, to build the application through a combination of VIP coding and Spectre in the open
• Importance of providing additional context to large language models through Model Context Protocol (MCP) servers
• Integrating the Context 7 MCP server to enhance the model's understanding of UI components and React documentation

Front-End Development

• Choosing React and Vite for the front-end project
• Leveraging Kira's documentation integration to bring in relevant information, such as React documentation and Tailwind CSS
• Iterative development process, with Kira automatically fixing errors and adding libraries as needed

Back-End and Deployment

• Creating TypeScript interfaces for user, game, and chat data
• Generating a mock data structure to represent the application's data
• Implementing an authentication flow, including a login page
• Deploying the application using AWS CDK (Cloud Development Kit)

Enhancing the Application with Generative AI

• Integrating a real-time AI chat assistant into the application
• Exploring two approaches for integrating the AI chat: sending a prompt and waiting for a response, or using a streaming approach to display the response word-by-word
• Leveraging the Bedrock MCP server to add AI functionality, including game recommendations based on user preferences

Best Practices for Agent-Driven Development

• Importance of breaking down prompts into smaller, more manageable steps for the agent
• Utilizing "metaprompting" to provide feedback and guidance to the agent on how to improve its responses
• Generating deployment scripts to ensure deterministic and reliable infrastructure deployments
• Integrating with the newly launched AWS MCP server, which provides access to AWS documentation, CLI integration, and curated Standard Operating Procedures (SOPs)

Business Impact and Applications

• Enabling rapid prototyping and development of full-stack applications with integrated generative AI capabilities
• Improving developer productivity and reducing the burden of maintaining up-to-date documentation and infrastructure
• Potential applications in various domains, such as e-commerce, gaming, and content creation, where generative AI can enhance user experiences and automate repetitive tasks

Key Takeaways

• Leveraging agent-based IDEs and MCP servers can significantly accelerate the development of complex, full-stack applications with integrated generative AI features
• Providing clear context and guidance to agents through metaprompting and curated SOPs is crucial for reliable and maintainable agent-driven development
• Integrating agent-driven development with infrastructure as code (IaC) tools, such as AWS CDK, ensures consistent and secure deployments
• The combination of agent-based development and generative AI opens up new possibilities for rapid prototyping and enhancing user experiences across various industries