Inspiration
Planning a trip can feel scattered across too many tabs: flights, hotels, maps, restaurants, timing, and personal preferences. Reiko was inspired by the idea of a calm travel concierge that can turn a messy travel wish into a clear, realistic plan.
What it does
Reiko is an AI travel curator that helps travelers build a complete trip from one conversation. It collects traveler preferences, suggests a destination flow, creates a day-by-day itinerary, highlights restaurants and places, and prepares booking-style next steps.
How we built it
We built Reiko with a static React frontend and a FastAPI backend. Gemini powers the planning agent, MongoDB Atlas stores trip and user memory, MongoDB MCP supports tool access, and Google Maps/Places grounds itineraries in real locations.
Challenges we ran into
One major challenge was designing the agent architecture. We split the planning flow into multiple specialized agents: a route agent to shape the overall trip path, three focused planning agents to handle itinerary details, and a review agent to check whether the final plan was realistic. The hard part was giving Gemini enough structure to produce a practical itinerary while still leaving room for creative generation.
Another challenge was building the API flow around Google Places. We needed enough real-world place data to make the plan feel grounded, but we also had to avoid unnecessary API calls. We worked on balancing efficiency and quality by deciding when to fetch fresh place details, when to reuse cached or curated data, and how much information each agent actually needed before generating the next step.
Accomplishments that we're proud of
We are proud that Reiko became more than a mockup: it is a deployed full-stack AI travel planner with a public frontend, a Cloud Run backend, Gemini-based planning, MongoDB integration, and Google Maps/Places grounding.
We are also proud of the agent structure. Reiko can turn a simple travel request into a realistic itinerary flow, then review and enrich the plan so it feels useful, not just generated.
What we learned
We learned how to collaborate with AI as a system designer, not just as a prompt writer. A big part of the project was harness engineering: giving autonomous agents enough freedom to generate useful travel plans, while adding guardrails so the output stayed realistic, structured, and safe for users.
We also learned that agentic systems work best when responsibilities are clearly separated. By giving each agent a focused role and adding a review step, we could guide Gemini toward better decisions without removing its ability to create flexible, personalized itineraries.
What's next for Reiko - Curated Trip Planner
Currently, Reiko launches with a curated Japan planning experience, but the system is designed to scale to many more destinations. Next, we want to expand Reiko with destination packs for Europe, the United States, and other popular travel regions.
Log in or sign up for Devpost to join the conversation.