GoAgent Travel AI

💡 Inspiration

Planning a vacation today is broken. We spend endless hours juggling between static recommendation lists, complex map pins, and rigid booking sites, trying to piece together a coherent, geographically optimized itinerary. We wanted to build a travel agent that doesn't just "chat" with you, but actually does the heavy lifting—co-piloting your trip interactively, understanding your vibe, and mapping it all out dynamically in real-time.

🛠️ How we built it

We developed GoAgent Travel AI by building a highly responsive, agentic system:

• Frontend: A sleek, interactive dashboard built with React and Tailwind CSS, integrated with Mapbox/Leaflet to render geographic coordinates, routes, and points of interest dynamically.
• Backend: A robust FastAPI (Python) server orchestrating the AI workflows.
• AI & Agentic Core: Powered by the OpenAI Assistant API and GPT-4o. We created a multi-agent workflow where a dedicated "Profile Agent" translates casual user chat and voice commands into structured preferences, while a "Spatial Optimizer Agent" processes geographic locations, clusters activities by proximity to minimize daily travel times, and triggers custom tools to fetch real-time accommodation and routing data.
• Database: Supabase acts as our real-time state manager, keeping user profiles, active chat history, and generated itineraries perfectly synchronized.

🧠 Challenges we faced

Our biggest challenge was maintaining seamless, real-time synchronization between the non-linear chat responses and the physical map visualizer. Translating abstract AI recommendations (e.g., "let's shift the afternoon walking tour to the evening") into precise, filtered database updates that immediately update the UI without losing session state required a carefully designed, event-driven architecture and strict JSON parsing schemas.

🎓 What we learned

We learned the immense power of combining Agentic Workflows with spatial data. Moving past simple text-based chatbots and allowing an LLM to directly interact with spatial tools (like maps and routing APIs) completely redefines how users interact with complex planning tasks. We also learned how to optimize token usage and latency by delegating lighter reasoning steps to faster models, keeping the experience fluid and instantaneous.

Built With

• css
• fastapi
• google-flights
• google-hotels
• google-places
• openai-api
• openweather-api
• pydantic
• python
• react
• serpapi
• sqlite
• supabase
• typescript
• vite