Building an AI-Powered Travel Assistant

Key Takeaways

1. Automation is the backbone of modern infrastructure. By using the AWS Cloud Development Kit (CDK) with Python, the presenters were able to eliminate repetitive manual tasks and minimize human errors when setting up the necessary infrastructure.

2. The quality of the knowledge base determines the quality of the AI. Providing context-rich and domain-specific information to the AI model results in more accurate and helpful responses.

Sections

Booking the Wrong Vegas

• The presenters, Guillermo Ruiz and Elizabeth Fuentes, shared a humorous story about booking flights to the wrong Las Vegas (Las Vegas, New Mexico instead of Las Vegas, Nevada) when preparing for their first time speaking at re:Invent.
• This experience highlighted the challenges of poor customer support, such as long wait times, disconnected systems, lack of personalization, and limited knowledge of support agents.

Building an AI-Powered Travel Assistant

1. Vector Database:

   • The presenters chose to use Amazon Aurora PostgreSQL with vector support to create a scalable and fully managed solution for their AI tasks.
   • A vector database allows for finding similarities between data based on meaning, rather than just exact keyword matching.

2. Knowledge Base:

   • The knowledge base acts as a library, providing the AI with relevant content to deliver accurate and context-aware responses.
   • The presenters used AWS CDK to automate the deployment of the Serverless V2 cluster, database secrets, and PostgreSQL table with custom resources.

3. AI Agents:

   • The presenters introduced Amazon Bedrock, an AI service that allows building AI applications without worrying about the underlying infrastructure.
   • They used prompt engineering techniques to create a structured prompt for their AI agents, which included the agent's role, context, user input, and desired output.
   • The AI agents were integrated with a passenger database, support ticket database, and the knowledge base to provide personalized and actionable assistance.

4. Messaging Integration:

   • The travel assistant was integrated with WhatsApp, allowing users to interact through their preferred messaging platform.
   • The integration used AWS Lambda functions to handle different types of user input (text, voice) and seamlessly pass them to the AI agents.

Conclusion

The presenters demonstrated how to build an AI-powered travel assistant using a combination of AWS services, including Aurora PostgreSQL, Amazon Bedrock, and AWS CDK. The key highlights of the solution are its serverless and cost-efficient architecture, AI-powered retrieval, seamless integration with messaging platforms, and multilingual support.