AI Framework Generates Personalized, Rule-Compliant Packing Checklists

This paper introduces a novel AI framework designed to generate personalized packing checklists for air travel, effectively balancing strict regulatory requirements with individual user preferences. The challenge lies in creating context-aware checklists that adhere to safety rules, item dependencies, and luggage limits, while also reflecting a user's unique needs. Existing solutions often fall short, being either too generic or unconstrained. The proposed framework operates in three stages. First, a symbolic engine generates an initial checklist that inherently respects regulations and dependencies. Second, a two-stage preference learner analyzes user actions (adding or removing items) to estimate inclusion and priority utilities, mitigating common biases. Finally, a CP-SAT optimizer selects a compact, compliant subset of items, ensuring all hard constraints are met. Evaluated on 604 trip scenarios, the symbolic engine achieved near-perfect recall and rubric validity, significantly outperforming frontier LLMs. The preference learner showed strong predictive performance, and the optimizer guaranteed 100% constraint satisfaction. Deployed in the FlyEnJoy iOS app, this system doubled checklist completions and reduced editing time, demonstrating its practical effectiveness for constrained personalization problems.