ABOUT      WORK      CONTACT 







Roam AI
Roam is an AI travel companion for spontaneous, emotionally-driven travelers. It surfaces saved places when you're nearby, adapts plans when reality hits, and guides you through the trip in real-time.
My Role

Product Design

My Contributions

Ideation
Experience Design
Visual Design
Prototyping
Vibe Coding Tools Used

Figma
Lovable
Supabase
OpenAI API
Claude



The Problem

Travel is inherently dynamic, yet travel tools are static. Travelers save dozens of places they never remember, create itineraries that break at the first weather change, and use tools that can plan trips but can't adapt to the moment.

Travel tools currently exist in two modes: planning before the trip, and searching during the trip. There's no middle ground for proactive, contextual guidance.

The Solution

An AI powered companion that really gets you.

Roam is an agentic AI travel companion that knows what you’ve saved, sees when you're nearby, adapts to weather changes, and learns your patterns to provide contextual guidance during your trip.

Research & Discovery

To start, I spoke to a small group of 6 frequent travelers to gain a better understanding of their behavior patterns, pain points, and how other tools fit into their travel routine.

Their travel planning styles ranged from rigid to spontaneus, yet they all had similar frustrations. The travelers struggled with decision fatigue and information overload.

They said...
  “I saved over 100 pins for my trip to Lisbon. I have no idea what most of them are anymore.”

“When it rains, I'm back to googling. My itinerary becomes useless.”

“I don't need more options. I need to know which option is right for me, right now.”
I found that the problem isn’t a lack of information. It’s the lack of timely information. Travelers need proactive guidance to reduce their frustrations, not another tool that helps them search for even more options.

Challenge #1: Contextual Alerts

So many things can make us change our plans last minute. It could be a weather change, traffic, closures, or even just a change of mood. Sometimes, you just need to be reminded of an activity you were interested in doing when planning the trip.

In these high stress moments, how might we help our users make decisions?


Solution:
  • Use AI to consider distance, schedule fit, location hours, user patterns, and past dismissals to decide when proximity actually matters.
  • Use AI to surface alerts via chat and push notifications when it is at least 80% certain of disruption.
  • Always provide alternatives, never just problems.
  • Let users know what's happening, the impact on their plans, and 2-3 AI generated alternatives.
  • Always allow users to have the final choice, if they want to reschedule, replace, or keep their plan.

Challenge #2: Conversational Intelligence

What’s the right interface for an AI tool that initiates conversations, yet gives users full control over their plans?


Solution:
Two tabs that share the same trip data, but serve different contexts. Timely suggestions appear in chat, while strategic suggestions appear inline in plan. Actions in one tab update the other.

Chat Tab (Tactical)
  • Real-time AI chatbot
  • AI initiates most conversations
  • Quick reply buttons, text or voice input

Plan Tab (Strategic)
  • Visual itinerary with day/time structure
  • Inline AI suggestions
  • Drag-and-drop editing

Challenge #3: Learning & Personalization

Every traveler is different, yet travel tools treat everyone the same. 

How can AI help us learn about our users wants and needs without being intrusive?


Solution:
Give users control over what they share, and progressively learn about their preferences.

Quick Onboarding
  • Learn about the users communication style and interests

Behavioral Learning - Observe patterns
  • Note which suggestions are accepted vs dismissed
  • Learn the users daily time and energy patterns

Transparency - Show what AI knows
  • Remember patterns to provide AI insights on previous trips
  • Users can view, correct, or reset any assumption
  • Provide clear settings and privacy controls

Design & Prototyping

To design an AI-first product, I needed to go beyond static mockups. I used Claude to explore AI behavior patterns, Figma for visual design and component systems, Lovable for rapid prototyping and testing live interactions, and ChatGPT as a thought partner for brainstorming edge cases and copy. This combination allowed me to iterate quickly.

I moved between Figma for visual exploration, and vibecoding in Lovable to better understand interaction patterns. This hybrid approach revealed insights that mockups alone couldn't show, like discovering that conversational chat messages felt less intrusive than card-style notifications. Working prototypes forced me to define exact thresholds and edge case behaviors, turning abstract concepts into concrete product design decisions.




    Let’s connectLinkedIn
    Email
    Brooklyn, NY