Transforming Travel: Building Personalized Holiday Packages with Generative AI

Introduction to Generative AI in Travel

Solution Overview

Prerequisites

Deploy the CloudFormation Stack

Data Storage and Management

Amazon Bedrock Configuration

API and Interface Setup

Verify the Setup

Test the Endpoints

Test the Solution

Clean Up

Next Steps

Conclusion

Related Resources

About the Author

Transforming Travel with Generative AI: A Guide to Personalized Holiday Packages

Generative AI is revolutionizing the way businesses create personalized experiences, particularly in the travel and hospitality sectors. Travel agents are now equipped to enhance their services with bespoke holiday packages tailored to meet individual customer preferences. From accessibility needs to dietary restrictions and unique activity interests, this evolution requires sophisticated solutions that marry extensive travel knowledge with real-time pricing and availability data.

In this post, we will explore how to harness Amazon Bedrock to build a generative AI solution that empowers travel agents to craft personalized holiday packages effortlessly. We will delve into how to utilize Amazon Bedrock Knowledge Bases for travel information, Amazon Bedrock Agents for real-time flight details, and Amazon OpenSearch Serverless for seamless package searches and retrieval.

Solution Overview

Travel agencies are facing increasing demands for personalized recommendations while grappling with the challenges of real-time data accuracy and scalability. Imagine a travel agency tasked with offering accessible holiday packages. They must align specific accessibility requirements with live flight and accommodation data, a feat complicated by the slow processing times and outdated information typical of traditional systems.

Our AI-powered solution merges personalization with real-time data integration, allowing agencies to automatically match accessibility criteria with current travel options and provide accurate recommendations in minutes instead of hours.

Three-Layer Architecture

Our architecture comprises three key layers:

1. Frontend Layer: An interface where travel agents input customer requirements and preferences.
2. Orchestration Layer: This layer processes the requests and enriches them with customer data.
3. Recommendation Layer: This final layer incorporates:
   • Travel Data Storage: A searchable repository of travel packages.
   • Real-Time Information Retrieval: Current flight details obtained through API integration.

This layered approach helps travel agents capture customer requirements, enrich them with stored preferences, integrate real-time data, and deliver tailored recommendations that align with customer needs.

AWS Implementation

The AWS implementation encompasses several essential services:

• Amazon API Gateway: Receives and routes requests to AWS Lambda functions for secure API access.
• AWS Lambda: Processes input data, creates enriched prompts, and runs the recommendation workflow.
• Amazon DynamoDB: Stores customer preferences and travel history.
• Amazon Bedrock Knowledge Bases: Helps build a curated database of destinations and travel packages.
• Amazon OpenSearch Serverless: Facilitates simple and efficient vector search capabilities.
• Amazon Simple Storage Service (S3): Stores large datasets, including flight schedules and promotional materials.
• Amazon Bedrock Agents: Integrates real-time information retrieval, ensuring recommendations reflect current availability and pricing.

To simplify the setup process, this solution employs an AWS CloudFormation template that provisions and configures all necessary resources, including service configurations and permissions.

Prerequisites for Deployment

To utilize this solution, you need:

1. A valid AWS account with access to Amazon Bedrock.
2. Permissions to manage the following services:
   • Amazon Bedrock
   • Amazon OpenSearch Serverless
   • AWS Lambda
   • Amazon DynamoDB
   • Amazon S3
   • Amazon API Gateway
3. Access to foundation models in Amazon Bedrock for Amazon Titan Text Embeddings V2 and Anthropic Claude 3 Haiku models.

Deploying the CloudFormation Stack

To deploy this solution, follow these steps:

1. Choose Launch Stack:

   • Click on "Launch Stack" to redirect to the AWS CloudFormation console.

2. Complete Stack Creation:

   • Leave the default settings intact and initiate the stack creation.
   • Monitor the deployment under stack events.

The stack creation typically takes around 10 minutes. Wait for the status to reach CREATE_COMPLETE before proceeding.

Data Storage and Management Setup

The CloudFormation template establishes:

• An S3 bucket with sample datasets (like travel_data.json and promotions.csv), a prompt template, and an API schema.
• DynamoDB tables filled with sample user profiles and travel history.
• An OpenSearch Serverless collection tuned for travel package searches.

Amazon Bedrock Configuration

The CloudFormation template will configure:

• A knowledge base containing ingested travel datasets from Amazon S3 with automatic syncing.
• An Amazon Bedrock agent, including action groups with mock flight data integration.

API and Interface Setup

The template configures resources for:

• API Gateway endpoints.
• Lambda functions demonstrating a mock flight API.
• A web interface for travel agents.

Verifying the Setup

After the stack creation, verify the setup on the Outputs tab of the AWS CloudFormation console, where you’ll find:

• Website URL: Access to the travel agent interface.
• API Endpoint: For programmatic access to the recommendation system.

Testing the Solution

The web interface allows travel agents to input customer details such as:

• Customer ID (e.g., Joe or Will)
• Travel budget
• Preferred travel dates
• Number of travelers
• Travel style

You can also test the API directly. For instance:

curl -X POST \
  -H 'Content-Type: application/json' \
  -d '{
    "userId": "Joe",
    "budget": "3000 GBP",
    "duration": "7 days",
    "travelDate": "2025-07-15",
    "numberOfTravelers": 2
  }'

Example Use Case

Let’s imagine a scenario where an agent helps Joe, a customer who requires wheelchair accessibility and prefers luxury vacations. The agent inputs:

• Customer ID: Joe
• Budget: 4,000 GBP
• Duration: 5 days
• Travel Dates: July 15, 2025
• Number of Travelers: 2
• Travel Style: Luxury

Upon submission, the system initiates a series of actions to query the knowledge base and check real-time flight information, returning personalized recommendations suitable for Joe’s needs.

Clean Up

To avoid incurring unwanted charges, delete the CloudFormation stack. The template ensures that all created resources, including S3 buckets, DynamoDB tables, and OpenSearch collections, are removed efficiently.

Next Steps

To further enhance this solution, you might consider:

• Exploring multi-agent capabilities for specialized travel aspects (hotels, activities).
• Implementing multi-language support with Amazon Bedrock models.
• Integrating with CRM systems for improved customer tracking.

Conclusion

In this post, we demonstrated how to build an AI-driven holiday recommendation system using Amazon Bedrock, providing travel agents with the tools to create personalized experiences. Our implementation showcased the potential of combining Amazon Bedrock Knowledge Bases with Amazon Bedrock Agents, addressing both historical travel data and real-time needs effectively. This innovative approach is essential for travel organizations eager to deliver tailored solutions that integrate real-time pricing, address accessibility requirements, and enhance customer satisfaction.

Related Resources

For additional information, explore the following resources:

• [Documentation]
• [Code Samples]
• [Additional Learning Materials]

About the Author

Vishnu Vardhini is a Solutions Architect at AWS based in Scotland, specializing in SMB customers across industries. With expertise in Security, Cloud Engineering, and DevOps, she architects scalable AWS solutions and is passionate about leveraging Machine Learning and Generative AI for business value.