LM-based Personalized Travel Planner
Organizations: Dept. of Computer Science & Engineering, University of Moratuwa, Moratuwa, Sri Lanka · WSO2, Sri Lanka
Abstract
Effective travel planning requires handling multiple information sources, managing constraints, and tailoring recommendations to user preferences. While LLMs have introduced dynamic, context-aware decision-making, single-agent models often struggle with complex multi-step reasoning, constraint adherence, and modular execution. To overcome these limitations, we propose a hierarchical multi-agent framework where a supervising chatbot agent coordinates specialized agents for schedule validation, user query formulation, real-time data retrieval, and constraint-aware itinerary generation. By integrating LLM reasoning, external APIs, and iterative refinement mechanisms based on a deterministic, multi-dimensional scoring framework, our approach ensures adaptability, efficiency, and personalization. Evaluations on the well-known TravelPlanner benchmark demonstrate improvements in Budget Constraint planning by 10%, wider constraints set-based planning by 25%, and personalized planning results by 45%. compared to the ReAct baseline. Our code is available at https://github.com/nimashaklaa/TravelPlanner_Research
Figures & tables
| Metric | Unrefined (Baseline) | Refined |
| Delivery Rate | 100% | 100% |
| Commonsense Constraint Micro Pass Rate | 88.875% | 81.125% |
| Commonsense Constraint Macro Pass Rate | 35% | 35% |
| Hard Constraint Macro Pass Rate | 50% | 60% |
| Final Pass Rate | 25% | 35% |
| Metric | Unrefined (Baseline) | Refined |
| Delivery Rate | 100% | 100% |
| Commonsense Constraint Micro Pass Rate | 88.875% | 91.875% |
| Commonsense Constraint Macro Pass Rate | 35% | 55% |
| Hard Constraint Macro Pass Rate | 50% | 60% |
| Final Pass Rate | 20% | 45% |
| Condition | Personalization | Final Pass | Hard Constraint | Commonsense | Delivery |
| Score (%) | Rate (%) | Pass Rate (%) | Pass Rate (%) | Rate (%) | |
| Full System | 87 | 52 | 68 | 62 | 100 |
| No Evaluator | 79 | 30 | 45 | 49 | 91 |
| No Feedback Agent | 71 | 28 | 42 | 46 | 90 |
| No Eval & Feedback | 65 | 21 | 35 | 40 | 86 |
| TravelPlanner Baseline | 56 | 10 | 28 | 15 | 89 |
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| Tool | Function |
| CitySearch | Search cities within a specified region. |
| FlightSearch | Search flights for a specific date between two cities. |
| DistanceMatrix | Search driving distance, time, and cost between cities. |
| RestaurantSearch | Search restaurants in the given city. |
| AttractionSearch | Search attractions in the given city. |
| AccommodationSearch | Search accommodations in the given city. |
| Total queries | 1,225 (train 45, val 180, test 1,000) |
| Trip lengths | 3, 5, 7 days |
| Difficulty | easy / medium / hard |
| Geographic spread | 192 origins, 127 destinations (test split) |
| Budgets (ours, for analysis) | <1K–3K buckets |
| Environment | Closed-world sandbox with offline tools (no live APIs) |
| Licensing | CC BY 4.0 (dataset), MIT (code) |