Celebto is an AI-native travel platform that simplifies the 20-tab planning struggle into a single conversational flow. I joined Celebto as the Founding Designer to turn the raw concept of an AI assistant into a premium product. I focused on a "Human-in-the-Loop" philosophy: the AI handles the logistics, but the user remains the architect. We replaced the overwhelming "wall of text" typical of AI tools with a playable canvas, giving travelers the speed of automation without losing their agency.
The Problem
Most AI travel tools fail because they assume the traveler wants to be a passive recipient rather than an active participant.
The Data Dump
Getting a long paragraph of text makes it difficult to visualize the trip or check if the locations actually make sense geographically.
Zero Control
If you don't like a single restaurant or hotel, you usually have to start the entire prompt over.
The Trust Gap
Travelers struggle to trust a bot with their time when they can't see the "why" behind a suggestion or verify it with real-world data.
Design Philosophy
Designing for an AI-native product meant abandoning the traditional "form-and-result" pattern. Instead, I focused on building a "living" interface that prioritizes user agency. We built a system that doesn't just "talk" to you, but actually builds with you, turning raw LLM data into structured, interactive components that feel like a conversation with a local friend.
Modular Component Architecture
o handle unpredictable AI outputs, I designed a system of "Fluid Containers." Every itinerary item is a modular card rather than a text block, allowing the UI to remain organized whether the AI provides a brief suggestion or a deep-dive local tip.
The Synchronized Canvas
A "living" interface requires real-time feedback. I designed a split-view Map and Timeline where the two are logically linked. As the AI streams data, the map populates instantly, providing immediate spatial context and building user trust through visual proof.
Conversational Control Loops
Instead of a "Back" button, I designed "Refinement Loops." By implementing features like the Swap tool and the Local Buddy, the interface allows users to correct the AI in real-time, treating the LLM as a draft-generator rather than a final decision-maker.
Key features
We built a system that doesn't just "talk" to you, but actually builds with you. We turned raw LLM data into structured, interactive components that feel like a conversation with a local friend.
Key features
A multi-step visual flow that translates user "moods" and energy levels into high-fidelity data for the LLM.
Key features
An interactive refinement tool that allows users to replace specific activities without breaking the trip's overall logistics.
Key features
A dedicated interaction layer where the AI shifts from "planner" to "expert guide," offering nuanced advice that a standard itinerary list would miss.
Key features
Every AI-generated card includes live ratings and booking links, bridging the gap between AI imagination and real-world availability.
