Travel Agent

A travel assistant that helps with flights, hotels, cost of living, and more

I built a Travel Assistant that can answer with real-time flight and hotel prices worldwide.
Unlike standard LLMs, it uses tools to query APIs for up-to-date data, giving precise answers on demand.

Github: https://github.com/PietroMelzi/travel-agent

How It Works

The architecture is designed for experimentation. Even though a single agent could handle all tasks, I implemented a multi-agent setup where a manager agent coordinates several specialized agents acting as tools:

  • Flight search agent
  • Hotel search agent
  • Cost-of-living info agent

This setup showcases the potential of tool-equipped LLMs and multi-agent systems.

Trabel Agent Screenshot

Technology Stack

  • Backend / Agent System: Built with the OpenAI Agents library, which simplifies:
    • Assigning tools to agents
    • Handling agent handoff

    An alternative could have been an MCP server to expose tools to multiple clients, but for simplicity, I assigned tools directly to my agents.

  • Frontend: Built with Streamlit for a quick proof-of-concept and easy deployment on Streamlit Cloud.
    This avoided the complexity of building a custom frontend at this stage.

Why This Project Matters

This project is not just about a travel assistant.
It’s a practical exploration of:

  • Tool-equipped LLMs for real-time data
  • Multi-agent orchestration
  • Rapid prototyping and deployment

It demonstrates how LLMs can act as orchestrators of specialized services, rather than just static answer engines.