Idea
On-device itinerary generation platform delivering feasible, personalized travel plans with low latency and high user satisfaction.
Research Paper
Core Innovation
This paper introduces the PLA framework combining heterogeneous lightweight planners, a Bradley-Terry reward model learned from human comparisons, and device-aware local refinement. It uniquely balances combinatorial feasibility and latent desirability on resource-constrained devices, outperforming large language models in feasibility and user preference alignment.
Why It Matters
Travelers face complex itinerary planning balancing strict constraints and subjective preferences, often unmet by existing tools. PLA improves trip completion rates and user satisfaction by ensuring feasible, personalized plans on mobile devices with minimal latency. This scalable approach transforms travel planning workflows for millions of users and travel apps.
Market Size (TAM)
$10–20B TAM for travel planning software; $2–5B SAM from mobile travel apps and online agencies. Driven by rising mobile travel bookings and demand for personalized experiences.
Potential Customers & Pain Points
- Travel app developers – Need personalized feasible itineraries with low latency
- Mobile device users – Require travel plans respecting preferences and constraints
- Tour operators – Seek scalable itinerary customization
- Online travel agencies – Want to increase booking completion rates
Business Model
Licensing PLA as an SDK or API to travel app developers and agencies; subscription or usage-based pricing tied to itinerary generation volume.
Competitive Landscape
- Google Trips
- TripIt
- Sygic Travel
- Rome2rio
Implementation Challenges
- Integrating with diverse travel data sources and APIs
- Maintaining real-time performance on varied mobile hardware
- Capturing evolving and diverse traveler preferences accurately
Validation Strategy
- Deploy PLA in partner travel apps to measure itinerary completion and user satisfaction
- Conduct A/B testing against existing itinerary planners
- Expand human preference data collection across diverse geographies
- Benchmark latency and feasibility on multiple device types
Research Paper Overview
From Feasibility to Desirability: Plan, Learn, Adapt (PLA) Framework for Personalized On-Device Itinerary Generation
Summary
PLA is a three-stage framework that generates personalized trip itineraries on mobile devices by balancing hard feasibility constraints with soft traveler preferences. It combines lightweight planners, a learned reward model capturing emergent schedule qualities, and local refinement within device compute limits. PLA achieves high feasibility and preference alignment, outperforming leading LLMs and improving itinerary completion rates in production.