TripFlow

AI-powered travel itinerary planning application

Comment

TripFlow

Inspiration

The inspiration for TripFlow came from witnessing the frustration of group travel planning firsthand. As avid travelers, we observed that collaborative trip planning took an average of 20+ hours, with group decisions often leading to conflicts and suboptimal itineraries. The emergence of Google's Agent Development Kit (ADK) in 2025 presented a unique opportunity to reimagine travel planning through specialized AI agents working in harmony. We envisioned a platform where AI wouldn't just provide recommendations, but would understand the nuanced preferences of different travelers and facilitate seamless collaboration.

What it does

TripFlow is an AI-powered collaborative travel planning platform that revolutionizes how people plan trips together. At its core is a sophisticated multi-agent AI system built on Google's ADK, featuring a hierarchical system with five AI agents:

  • Root Concierge Agent: Orchestrates and coordinates all specialized agents, analyzing user intent and delegating tasks
  • Tourism Agent: Leverages Korea Tourism API to curate personalized attraction recommendations
  • Map Agent: Optimizes routes using advanced algorithms considering time windows, preferences, and transportation constraints
  • Booking Agent: Integrates with multiple booking platforms for accommodations and transportation
  • Food Agent: Recommends restaurants based on dietary restrictions, price preferences, and local authenticity

The platform supports real-time collaboration with role-based permissions (OWNER, EDITOR, VIEWER), and localization across 9 languages (EN, KO, JP, ZH_CN, ZH_TW, DE, FR, ES, RU). Built with Flutter for cross-platform mobile deployment and Spring Boot for enterprise-grade backend services.

How we built it

Architecture & Technology Stack

We designed TripFlow using a microservices architecture with Domain-Driven Design (DDD) principles:

Frontend: Flutter 3.0+ + Dart 3.0+ with Provider state management, implementing clean architecture with presentation, domain, and data layers.

Backend: Java 21 + Spring Boot 3.5.0 following Domain-Driven Design patterns:

  • User Context (Firebase JWT authentication, FCM notifications)
  • Trip Context (core domain with collaborative editing)
  • Tourism Context (Korea Tourism API integration with Caffeine caching)
  • AI Context (AI-ADK service integration via Server-Sent Events)

AI Services: Python 3.12+ + Google ADK 1.12.0 implementing a hierarchical multi-agent system.

Infrastructure: PostgreSQL 16 with UUID primary keys and JSONB support, Redis for caching, deployed on Google Cloud Platform using Cloud Run with auto-scaling.

Multi-Agent AI System Implementation

The most challenging aspect was implementing the multi-agent system. We created a Root Concierge Agent that acts as orchestrator, analyzing user intent and delegating tasks to specialized domain agents. Each agent operates independently but coordinates through a standardized communication protocol.

Real-time Collaboration System

We implemented real-time collaboration using WebSocket connections with conflict resolution algorithms. The system maintains operational transforms to handle simultaneous edits and uses optimistic locking for data consistency.

Challenges we ran into

1. Multi-Agent Coordination Complexity

Challenge: Ensuring agents communicate effectively without creating infinite loops or conflicting recommendations.

Solution: Implemented a hierarchical delegation pattern with the Root Agent acting as orchestrator, with robust communication protocols and failure prevention mechanisms.

2. Real-time Collaboration

Challenge: Managing concurrent edits from multiple users while maintaining data consistency.

Solution: Implemented real-time synchronization with conflict resolution and optimized data handling for smooth collaborative editing.

3. Cross-Platform State Management

Challenge: Maintaining complex application state across Flutter's widget lifecycle while ensuring offline functionality.

Solution: Adopted Provider pattern with hierarchical state management (Infrastructure → Service → Feature providers) and implemented persistent storage using SQLite with automatic synchronization.

4. API Rate Limiting & External Service Dependencies

Challenge: Korea Tourism API has strict rate limits (60 req/min), and external service failures could break user experience.

Solution: Implemented multi-layer caching strategy with intelligent TTL management, fallback mechanisms for external service failures, and request queuing with exponential backoff.

5. Performance Optimization for AI Inference

Challenge: AI model inference times needed optimization for better user experience.

Solution: Optimized with parallel agent execution, implemented result caching, and used Gemini 2.5 Flash for optimal speed/quality balance.

Accomplishments that we're proud of

Technical Achievements

  • Time Efficiency: Streamlined trip planning process through AI automation
  • Reliable Performance: Achieved stable service delivery through robust architecture
  • 9-Language Support: Full localization with cultural context adaptation

Innovation in Multi-Agent Systems

Implemented Google ADK with hierarchical agent communication, demonstrating how specialized AI agents can collaborate effectively to provide domain-specific travel recommendations.

Scalable Architecture

Built a system with cloud-native deployment and efficient database design to support user growth.

What we learned

Technical Insights

Multi-Agent Systems: Learned that agent specialization significantly improves output quality compared to general-purpose models. Domain expertise encoded in individual agents produces more nuanced, contextually appropriate recommendations.

Distributed System Design: Learned about building resilient distributed systems, including circuit breakers, bulkhead patterns, and graceful degradation strategies.

Real-time Collaboration: Discovered that operational transforms and conflict resolution algorithms are essential for smooth collaborative editing experiences.

Mobile-First Architecture: Learned to design backend APIs that work seamlessly with mobile constraints (intermittent connectivity, limited battery, varying network quality).

AI/ML Learnings

Prompt Engineering: Developed sophisticated prompting strategies for different agent roles, learning that context-specific instructions significantly improve output quality.

Model Selection: Discovered that Gemini 2.5 Flash provides optimal balance of speed and quality for our use case, while maintaining cost efficiency.

Caching Strategies: Learned that intelligent caching of AI results can substantially reduce costs while improving user experience.

Product Development

User-Centric Design: Learned that complex AI capabilities must be presented through simple, intuitive interfaces. The most sophisticated backend is useless if users can't easily access its power.

Iterative Development: Discovered the importance of MVP validation before building complex features. We refined our multi-agent architecture based on user feedback and performance requirements.

What's next for TripFlow

Short-term Roadmap (Q3-Q4 2025)

Enhanced Personalization: Implementing user preference learning to improve recommendations based on trip feedback and usage patterns.

Android Launch: Completing Android version development and Play Store deployment.

Booking Information Integration: Enhanced booking information aggregation and comparison features.

Medium-term Vision (2025-2026)

Regional Expansion: Extending coverage to Japan and Southeast Asia with localized content and recommendations.

Enhanced User Experience: Improved accessibility features and user interface enhancements.

Smart Analytics: Implementing data-driven insights for travel timing and destination recommendations.

Long-term Innovation (2026+)

AI Evolution: Continuous improvement of AI agents for better personalization and travel condition adaptation.

Algorithm Enhancement: Ongoing optimization of itinerary planning algorithms for better user experience.

Community Features: Developing social sharing and community-based recommendation capabilities.

TripFlow aims to enhance travel planning through intelligent AI assistance that complements human decision-making. We're building a platform that makes travel planning more efficient and collaborative, helping people create better travel experiences together.

TripFlow - Redefining Travel Planning with AI
Built with ❤️ by the TripFlow Team

Built With

Share this project:

Updates