My Role
UX Researcher
UX Designer
Skills
Research
UX Design
Timeline
3 months
Industry
SaaS, B2C

Setting the Scene
The Context
Planning a trip today often means juggling a dozen tabs—blogs, booking sites, maps, and weather forecasts—each offering fragments of information. This scattered experience is overwhelming, time-consuming, and leaves travelers unsure they’re making the best decisions.

Project Overview
The Problem
Why It Matters?

Where it all began:
Who are the users?
Primary Users
Busy Professionals & Millennials
Age: 25–40
Tech-savvy, rely on digital tools for planning.
Want quick, hassle-free trip planning without juggling multiple apps.
Frequent Travelers / Vacation Planners
Often plan 1–3 trips per year.
Struggle with research overload, comparing options, and creating itineraries.
Value efficiency, personalization, and recommendations tailored to their interests.
Families & Group Travelers
Looking for coordinated planning for multiple people.
Need tools to manage schedules, accommodations, and activities for everyone.
Value simplicity and trustworthiness to reduce planning stress.
Secondary Users
Budget-conscious travelers – want cost comparisons, deals, and savings insights.
First-time international travelers – need guidance, tips, and confidence in their choices.
Constraints & Business Goals
Every ambitious idea meets reality checks, and our prototype was no exception. Early on, we realized that Jupyter Notebooks didn’t support real-time microphone input, so the seamless “talk-to-your-travel-agent” experience wasn’t possible out of the box. Instead of pausing progress, we recorded voice separately and transcribed audio using Whisper.
We also ran into data challenges. Without live flight or hotel APIs, we relied on dummy itineraries and hardcoded examples. Surprisingly, users didn’t mind at this stage — what they valued most was seeing how an AI could guide them from idea → plan → itinerary.
Finally, we made a conscious decision to prioritize speed over polish. By taking a prototype-first approach, we could validate the flow quickly before scaling to production.
Key constraints
🎙️ Microphone Integration Limitations
Jupyter Notebooks don’t support real-time input → used audio file upload + Whisper transcription.
🛫 No Real-Time APIs
Dummy itineraries and hardcoded data substituted for flight/hotel search.
⚡ Prototype-First Approach
Prioritized feasibility testing and end-to-end flow over UI polish or deployment readiness.

The Solution
To validate the idea, I built and tested the prototype in a Jupyter Notebook, allowing for quick iteration and experimentation. Different types of travelers brought unique planning styles and expectations—some wanted fast, no-fuss recommendations, others needed help balancing budgets and preferences, while many looked for inspiration and hidden gems.
The voice-enabled assistant adapted seamlessly to these needs:
It condensed hours of research into a few curated options.
It simplified complex planning into a guided, stress-free flow.
It uncovered unique destinations and activities aligned with personal interests.
What emerged was more than just a Notebook experiment—it was proof that travel planning could feel less like work and more like a dialogue. By blending personalization, conversation, and accessibility, TripMate started to transform overwhelming decisions into inspiring journeys.
Voice-enabled AI Travel Assistant prototype using a Jupyter Notebook
Speak or type their travel goals.
Get personalized destination suggestions and day-by-day itineraries.
Listen to the AI assistant read the itinerary aloud.



Multi-Modal Experience
Seamless interaction through multiple input & output modes:
Voice Input: Users can speak their preferences and queries.
Visual Output: Dynamic itinerary, maps, and booking confirmations displayed in notebook.
Audio Feedback: AI reads out itineraries and responses via voice.
Benefits:
Enhances accessibility and user convenience
Mirrors natural travel agent interactions
Supports both exploratory and task-based use cases
Here are the features I integrated and tested!
I enhanced the assistant with key features to improve usability and convenience:
💬 AI Chat — for dynamic, conversational planning and real-time responses.
📄 PDF Download — so users can easily save and share their personalized itineraries offline.
🗺️ Google Maps Integration — providing interactive location context and helping users visualize their travel plans seamlessly.
These additions bring us closer to a fully immersive, user-friendly travel planning experience.

AI Chat

PDF Saved

Google Maps
Impact
Potential at Scale
(Projected Metrics)
With real-time travel APIs and a hosted conversational UI:
User reach (Year 1): 10,000+ travelers
Time saved per user: ~40 minutes on average
Cumulative time saved: ~6,667 hours annually
Conversion uplift (personalized planning): +12–15% potential increase in booking engagement


User Testing & Feedback (n-5)
Avg. Task Completion Rate: 91%
Users successfully generated an itinerary using voice-based input and reviewed the plan.
Voice-to-Itinerary Satisfaction:
⭐ 8.7 / 10 average rating
Users appreciated the novelty, usefulness, and natural flow of voice input.
Time Saved on Planning:
⏱️ Estimated 30–45 minutes saved per user versus manual web search and coordination.
Learning Curve Feedback:
📉 3 out of 5 users reported needing support to use Jupyter—prompting UI simplification in future iterations.
Diving In: Research Phase
🔍Research & Insights → 💡 Design Decisions
72% of travelers feel overwhelmed by too many choices
👉 Designed a conversational UI to narrow options naturally, like chatting with a friend.
84% switch between 5+ platforms
👉 Built a multi-modal flow (voice, text, visual cards) where research, planning, and booking happen in one place.
10–30 hours spent planning a 7-day trip
👉 Introduced auto-generated itineraries with drag-and-drop customization to save time.
Users want simplicity & less friction
👉 Created a single unified experience—no app-hopping required.
Trust and transparency are critical
👉 Added explainable AI recommendations (why a destination or hotel is suggested) to build confidence.

Trade-offs and Design decisions

Design / UX Trade-offs
Static conversational UI
❌ No real-time text bubble or chat UI.
✅ Jupyter print statements simulate a conversation.
🛠 Future: Use Streamlit or Gradio for a better conversational front-end.
No voice wake word or interrupt feature
❌ Users can’t interrupt or ask follow-up questions naturally.
✅ Linear dialog only.
No personalization memory
❌ Assistant doesn’t remember user preferences over sessions.
✅ Each prompt starts fresh unless manually engineered.

Scalability & Product Trade-offs
Prototype isn’t connected to live booking engines
❌ Can’t retrieve real-time pricing, availability, or locations.
✅ GPT can still simulate a convincing itinerary.
No user database or authentication
❌ No saved trips or multi-user support.
✅ Keeps prototype lightweight and secure.
Not optimized for mobile or multi-platform use
❌ Jupyter runs only in-browser on desktop/laptop.
✅ Perfect for developer and proof-of-concept audiences.
Final Design
To rapidly prototype and iterate on the UI, I used UX Pilot. This tool enabled quick creation of realistic mockups, helping us visualize user flows and gather early feedback efficiently before full development.

Next Steps: Models & Scaling consideration
To evolve this AI Travel Assistant beyond this prototype, I'm focusing on smarter, more adaptive AI techniques.
🎯 I plan to use supervised learning to better classify user intents—like distinguishing between a “romantic getaway” or a “family-friendly” trip—so the assistant understands goals more precisely.
🧩 Unsupervised learning will help cluster travel preferences from large datasets, enabling personalized recommendations based on similar user profiles.
🎛️ Reinforcement learning will fine-tune dialog flows by learning from real user feedback and engagement, making conversations more natural and helpful over time.
🔗 Finally, integrating Retrieval Augmented Generation (RAG) will allow the assistant to pull live data from travel APIs combined with LLM responses—giving users up-to-date, accurate information seamlessly.
These steps will transform the assistant from a prototype into a scalable, intelligent travel companion.

