TravelColombia.ai · 2025
AI travel planning for 37,000+ monthly users
An AI travel planner that generates real-time itineraries and cuts trip planning time by 60 percent.
- Role
- AI Core and Frontend Architecture
- Year
- 2025
- Stack
- Next.js, TypeScript, Gemini API, Zustand
TravelColombia.ai turns a rough idea of a trip into a complete, editable itinerary in seconds. I architected the AI core and led the frontend for the product, which now serves more than 37,000 monthly active users and reduces the time people spend planning a trip by roughly 60 percent.
The interesting engineering lives in the seam between the model and the interface. Itineraries stream token by token from the LLM, and the UI parses that stream into structured cards for days, activities, and places while it is still arriving. Users can regenerate any single day without losing the rest of the plan, which meant designing the state model around partial, replaceable fragments rather than one monolithic response.
The frontend is a heavily stateful Next.js application. Zustand holds the itinerary graph and the streaming buffers, and the rendering layer is built to stay responsive while long generations are in flight.
> key_details
- 37,000+ monthly active users on a consumer AI product
- Token-by-token streaming parsed into structured itinerary UI in real time
- Per-day regeneration without discarding the rest of the plan
- Complex client state managed with Zustand around partial LLM output