Inspiration
Travel booking today is fragmented — users need to jump between apps for flights, hotels, cabs, and guides. We wanted to reimagine this process with an agent-to-agent (A2A) protocol that enables seamless collaboration between multiple specialized agents, while ensuring security, personalization, and continuity in the user’s journey.
What it does
Provides a multi-agent system where each agent handles a specific task (flight booking, hotels, taxis, itinerary, guides). Creates a unique session key for each user session to ensure secure communication between agents. Stores user authentication and profile details (name, email, address, age, location). Lets users select travel options (flight, hotel, taxi, metro, etc.) and generates a complete tour plan. Integrates Uber, Ola, buses, and metro APIs to suggest best transport options with real-time fares. Provides travel guides, suggested locations, YouTube videos, and even Google Earth views. Supports chat-based interactions where the user can ask queries anytime, and agents collaborate to respond. Implements a security mechanism where any suspicious/fake link or unauthorized agent resets the session key with user notification.
How we built it
Backend: FastAPI with REST/GraphQL APIs for multi-agent communication. Frontend: Flutter Web app → React.js Streamlit prototype for quick testing AI/LLM Layer: Ollama, LangChain, and Selenium for dynamic queries from web, Wikipedia, DuckDuckGo. Database: Firebase/Firestore and PostgreSQL for storing user data, plans, and session keys. Integration: APIs for flights, hotels (Nuitée LiteAPI demo), Uber, Ola, YouTube, Google Earth.
Challenges we ran into
Designing a multi-agent protocol where each agent has its own card and session logic. Ensuring session security with dynamic key bursting when malicious activity is detected. Managing context memory so that past conversations influence future responses. Integrating multiple APIs smoothly (transport, hotels, YouTube, maps) while keeping the UX simple. Building prototypes across different platforms (Streamlit, Flutter, React).
Accomplishments that we're proud of
Created a working multi-agent A2A protocol demo. Built a Streamlit prototype and completed 50% of the Flutter app. Designed a secure session management system with key refresh. Integrated real-time transport options and hotel booking APIs into one unified flow. Showcased how LLM + LangChain can enhance dynamic queries for travel.
What we learned
How to design a secure and scalable multi-agent system. Effective use of LLMs with external tools (LangChain + Selenium) for live data retrieval. Balancing UX and backend complexity when building across multiple platforms. The importance of API orchestration in creating real-world AI-driven products.
What's next for A2A Travel Concierge & Planner
Fully integrate the Flutter app with FastAPI backend. Add voice assistant support for hands-free planning. Improve recommendation engine using AI personalization. Extend the security model with blockchain-based session tracking. Launch as a cross-platform solution for travelers and agencies.
Log in or sign up for Devpost to join the conversation.