Inspiration
Travel planning becomes frustrating when AI tools produce rushed schedules, ignore travel time, and invent vague attractions. I built this project to treat trip planning as a constraint-satisfaction problem rather than pure text generation.
My software engineering experience taught me that reliable systems validate inputs, enforce structure, and handle uncertainty explicitly. Instead of asking open-ended questions, this system collects hard constraints first (dates, duration, mobility, transport limits) before incorporating preferences. I focused on realism, pacing, and transparency so users can trust the plan.
What it does
AI Trip Itinerary Generator creates realistic, preference-aware itineraries while minimizing hallucinations.
It supports two modes:
Option A — Discover a Destination
- For users unsure where to go
- Interprets high-level goals
- Applies distance, budget, and pacing constraints
- Recommends a coherent destination and plan
Option B — Known Destination
- For users with a fixed location
- Builds structured day-by-day schedules
- Respects time limits, rest needs, accessibility, and travel flow
- Avoids overpacked or repetitive days
Key features:
- Structured constraint and preference collection
- Required vs. optional input separation
- Explicit handling of unknown answers
- Logically ordered daily plans
- Validated structured JSON outputs
- Automatic PDF itinerary export
How I built it
The backend is built with FastAPI and it converts structured inputs into validated itineraries.
Core components:
- Constraint-First Question System: collects required constraints before preferences and prevents unrealistic assumptions
- Structured Generation Pipeline: model outputs structured JSON and custom schema validates format and completeness
- Schema Validation: ensures consistent itinerary structure and rejects malformed outputs
- PDF Generation Engine: converts itineraries into formatted PDFs and generates timestamped files automatically
Design Approach:
I treated itinerary creation as a planning problem that balances constraint satisfaction, travel-time realism, activity pacing, preference weighting, and uncertainty handling.
Challenges I ran into
- AI hallucinations: models invent places and overload schedules. I reduced this with schema validation and constraint-aware prompting.
- Flexibility vs. structure: too much structure limits personalization; too little reduces reliability. I balanced required constraints with optional preferences.
- Realistic travel flow: geography, pacing, and fatigue complicate planning. I refined prompts to prevent excessive transitions and day density.
- Output consistency: models produce inconsistent JSON. Schema validation and regeneration resolved this.
Accomplishments that I'm proud of
- Built a full pipeline from user inputs to downloadable PDFs
- Designed a constraint-first system that reduces hallucinations
- Implemented reliable schema validation for structured outputs
- Created two flexible trip-planning modes
- Prioritized realistic pacing over flashy text generation
What I learned
- Structured systems greatly improve AI reliability
- Explicit uncertainty produces more trustworthy outputs
- Planning problems require logic beyond language prompts
- Validation layers are essential for production AI systems
What's next for AI Trip Itinerary Generator
- Complete React frontend with interactive question flow
- Enable browser-based PDF downloads
- Improve constraint modeling for higher accuracy
Log in or sign up for Devpost to join the conversation.