TripPilot

Trip assistant
Recommended places with reasons
Recommended places with details (suggested time, opening hours, ticket, tips)
Ordered routes with Gemini3 power
Ordered routes with reasons and suggestions
Change departure time according to suggestions
Recalculated routes that can be realized
Estimated cost
Share function
Share to Google Calendar with routes details
Community function
Share to Google Maps to navigate
Recommendation function, and estimated hours function

Inspiration

Travel planning is full of "looks fine on a map" traps: a route might be short, but opening hours, realistic dwell time, and travel duration make the day impossible. Traditional planners give you a list of places—but when do you actually arrive? Will it be open? How long should you stay?

We wanted something that feels like talking to an assistant, but behaves like a planner—with explicit constraints, concrete places, and plans you can actually execute.

What it does

TripPilot transforms rough travel intent into a feasible, time-aware itinerary:

Multimodal Input: Speak or type your travel ideas naturally ("2 days in Rome, I love museums and want to see night views")
AI-Powered Itinerary Generation: Gemini converts free-form input into structured day-by-day plans with specific attractions
Real-World Validation: The app fetches actual opening hours and calculates travel time between stops
Constraint Detection: Gemini analyzes your itinerary and identifies conflicts (arriving after closing, insufficient time, etc.)
Smart Suggestions: When issues arise, Gemini proposes actionable fixes—reorder stops, reduce stay time, or start earlier
One-Click Export: Export directly to Google Maps navigation or Google Calendar

How we built it

Frontend: Vite + React + TypeScript + Tailwind CSS
AI Engine: Gemini 3 API with structured JSON output constraints
Maps Integration: Google Maps Platform (Places API, Routes API, Maps JavaScript API)
Cloud Backend: Firebase Authentication + Firestore for trip persistence
Voice Processing: Gemini multimodal API for speech-to-text
Deployment: Vercel with server-side Gemini proxy (API key protection)

Key Architecture Decision: We force Gemini to output strict JSON schemas, not free-form text. This means every AI suggestion is directly executable—"reorder: placeId1,placeId2,placeId3" or "reduce_stay: placeId, 30 minutes"—not vague advice.

Challenges we ran into

Parsing Complex Opening Hours: Google Places returns nested arrays of time periods; handling edge cases (24-hour venues, holidays, multiple time slots) required careful normalization
Constraint Reasoning: Getting Gemini to reason about multiple constraints simultaneously (travel time + opening hours + recommended stay) required a 3-step pipeline: Normalize → Detect → Decide
Voice UX on Mobile: Implementing hold-to-talk with proper pointer capture, auto-stop after 25 seconds, and graceful error handling across browsers

Accomplishments that we're proud of

Closed-Loop Planning: AI suggestion → apply change → recompute routes → compare outcomes—users see the improvement immediately
8+ Language Support: Full localization with locale-aware Gemini prompts (Chinese, Japanese, Korean, French, Spanish, German, etc.)
Natural Voice Input: Hold-to-talk feels like a walkie-talkie, not clunky click-to-record interfaces
Real Feasibility Checking: Not just "this looks good"—actual validation against venue hours and travel times

What we learned

Prompting for structured JSON requires explicit schema constraints and examples
Travel planning has many hidden constraints—opening hours alone are surprisingly complex
Users want AI to propose changes, not silently apply them—transparency builds trust

What's next for TripPilot

Our vision: TripPilot as your AI travel companion—not just a planner, but a guide that stays with you throughout your journey.

Near-term roadmap:

On-trip assistance: Real-time rerouting when plans change (missed bus, attraction closed unexpectedly)
Proactive suggestions: "You have 2 hours before dinner—here's a café nearby with great reviews"
Multi-day optimization: Automatically balance attraction density across days

Longer-term vision:

Live voice guide: Speak to TripPilot while walking—"What's the history of this building?"
Context-aware reminders: "Your museum reservation is in 30 minutes, start walking now"
Post-trip memories: Auto-generate trip summaries with photos and highlights