Travel Agent GPT

Overview

This project is a GenAI-powered travel planning assistant built on Kore.ai. It supports both new and repeat users, extracts travel preferences from free-form input, and generates structured destination and itinerary outputs.

Problem

Travel planning typically requires multiple back-and-forth steps to collect preferences, confirm constraints, and create an itinerary. Traditional bots struggle with ambiguous user inputs and frequently break when entity extraction is unreliable.

Solution

A Kore.ai bot architecture that uses Generative AI to:

• Extract preferences from natural language (destination, budget, travel type, dates, interests)
• Personalize outputs using historical context for repeat users
• Generate destinations and itineraries using reusable components
• Handle edge cases with structured JSON prompt contracts and session persistence

Architecture

Core flows

• NewConsumerFlow: end-to-end onboarding and itinerary generation
• RepeatedUserFlow: preference extraction + personalization using past travel context

Reusable components

1. Entities Extractor

   • GenAI prompt node + interpretation + script extraction
   • Stores results in context.extractedEntities and persists via session storage

2. Destination Generator

   • Generates destination options from extracted preferences
   • Injects historical data for repeat users when available

3. Itinerary Formatter

   • Generates a structured TRIP-format itinerary
   • Supports dynamic itinerary refinement (modify a single day without rewriting everything)

Reliability improvements

• Structured JSON-style system context for consistent extraction
• Session persistence across dialog tasks
• Replaced unreliable regex routing with GenAI-based intent extraction

Tech Stack

• Kore.ai XO Platform (dialog tasks, entity extraction, routing)
• Generative AI nodes (prompting + structured output)
• Bot scripts for entity/session persistence
• n8n (workflow ideas for future automation extensions)

Results

• More reliable entity extraction across ambiguous user inputs
• Reusable architecture across new and repeat user flows
• Faster iteration by standardizing prompt contracts and storage patterns

What’s next

• Itinerary Comparison Tool
• Booking support and modification flows
• External integrations: flights/hotels/local experiences