The Founding Story

Starting from a blank canvas

Trazzeo was born from a simple observation: planning a trip is still painfully fragmented in the age of AI. You're hopping between Google, TripAdvisor, YouTube, and a dozen open tabs just to plan a weekend away. There was no single product that let you have a natural conversation — "I want to go somewhere in Southeast Asia, 4 days, beaches and good food" — and get back a real, actionable itinerary.

As the founding designer and product manager, I was responsible for everything from initial market research and product strategy to the full design system and the AI chat engine that became Trazzeo's core product. This wasn't a case of handing off specs to a team — I was simultaneously shaping what to build and how it should look and feel.

The dual role challenge. Being both PM and designer at a zero-to-one stage means every decision is underdefined. There's no existing product to react to, no legacy user base to protect. You're constantly oscillating between "what should we build?" and "how should this work?" — often in the same hour. The discipline that kept this productive was time-boxing: dedicated "PM mode" blocks for roadmap, prioritisation, and stakeholder alignment, and "designer mode" blocks for deep craft work.

This case study focuses specifically on the AI Engine — the conversational AI planner that lets users chat their way to a full travel itinerary — alongside the broader UX and product design work that supported it.

Market Gap

No travel product let users plan a full trip through natural conversation. Existing tools were either search-based (Google, TripAdvisor) or inspiration-only (TikTok, Instagram) — none closed the loop from "I want to go somewhere" to a confirmed itinerary.

Designing for AI Conversation

Conversational UI is unfamiliar territory for most users. Designing an AI chat interface that felt natural — not like a search bar, not like a command line — required careful prompt framing, suggested starters, and graceful handling of vague or ambiguous user inputs.

Generational Behaviour Gap

The two primary user segments — Gen Z digital natives and early Millennials — have fundamentally different mental models for research and planning. Gen Z expects one-tap AI answers; Millennials are conditioned to open ten tabs. The design had to onboard both without alienating either.

Startup Resource Reality

As founding designer-PM, every prioritisation decision had direct cost implications. Designing a scalable AI product while shipping fast required rigorous component thinking, a robust design system, and ruthless scope management — particularly around which AI interactions to build vs. defer.

User Personas

The travellers we designed for

Research produced two distinct primary personas separated by generational technology behaviour. Both are travellers — but how they plan, research, and make decisions is fundamentally different. Every design decision in Trazzeo was pressure-tested against both.

Naomi Lim

Gen Z · Content Creator, Age 24 · Singapore

GenerationGen Z (born ~2000)

Trips per year5–8 short trips

Research styleTikTok, Reels, AI chat

Tech adoptionEarly adopter — uses AI daily

Behaviours

Uses ChatGPT and Gemini daily — naturally expects AI to handle travel planning
Discovers destinations through short-form video before any formal research
Wants to describe what she wants and get an instant, curated result
Impatient with multi-step flows — expects the app to do the heavy lifting

Pain Points

AI tools like ChatGPT give good recs but can't actually build or save a real itinerary
Existing travel apps feel transactional and outdated — not conversational
Wants personalisation without filling in long preference forms

"I just type what I want into ChatGPT but then I still have to go somewhere else to actually book it. Why can't it just do everything?"

Ryan Koh

Early Millennial · Product Manager, Age 34 · Singapore

GenerationEarly Millennial (born ~1990)

Trips per year3–5 trips, solo & group

Research styleGoogle, blogs, 8–12 open tabs

Tech adoptionComfortable but habitual

Behaviours

Knows AI exists but defaults to familiar tools — Google, TripAdvisor, travel blogs
Spends 3–5 hours across multiple sessions researching before confirming any trip
Cross-references at least 3 sources — has low trust in single-source answers
Open to AI only if it can demonstrate reliability and show its sources

Pain Points

Tab overload — loses track of what he's already reviewed across sessions
Synthesising information from multiple sources into one plan is mentally exhausting
Sceptical of AI recommendations — needs to feel in control of the final decision

"I don't mind using AI but I need to trust it first. Show me where the recommendation comes from and I'm in."

Information Architecture

A platform with two front doors

The card sort and tree testing results (validated with 8 participants) produced a dual-entry IA — seeker and host views accessed through a role-aware login flow. The bottom navigation for seekers was deliberately kept to four items, with the AI engine surfaced as the default home state rather than a secondary tab.

Traveller View

Root

Trazzeo App

Nav (L1)

Discover (AI Feed)

Explore

Trips

Profile

L2 Pages

Personalised for you

Friend activity

Browse categories

Search & filters

Activity detail

Group booking

Upcoming trips

Past adventures

Wishlist

Adventure score

The AI Engine

Designing for a system that learns

The AI chat engine was Trazzeo's core product — and the most challenging design problem. Unlike a search bar or a filter UI, a conversational interface puts the user in an open-ended interaction with no obvious affordances. Designing for AI chat meant solving for discoverability (how do users know what to ask?), trust (how do they know the answer is good?), and continuity (how does the conversation become a usable itinerary?).

How the AI Engine works

The engine uses a combination of user context, conversational history, and curated travel data to generate and refine itinerary recommendations through chat. As founding designer-PM, I defined the conversation UX patterns, the itinerary output format, and the trust signals that make AI-generated plans feel credible rather than generic.

Conversational context

What the user types — destination preferences, trip duration, travel style, budget, group size. The AI extracts structured intent from natural language without requiring form-filling.

Session memory

The AI remembers what was discussed earlier in the session — refining recommendations as the conversation develops. "Actually make it more budget-friendly" updates the whole itinerary, not just one item.

Curated travel data

A vetted database of destinations, experiences, and local tips that grounds AI responses in reliable, up-to-date information — preventing hallucination and building user trust through sourced recommendations.

Conversation Starters

New users face a blank chat input with no guidance. Designed contextual prompt suggestions — "Plan me 4 days in Bali under $1,500" — that model the right way to talk to the AI, lowering the barrier to first interaction significantly.

Source Transparency

Millennial users in particular distrusted AI recommendations with no provenance. Added lightweight source labels to each recommendation — "From TripAdvisor", "Local pick" — directly addressing the "show me where this comes from" need surfaced in persona research.

Itinerary Export

A conversation is only useful if it becomes something actionable. Designed the itinerary export flow — turning a chat thread into a structured day-by-day plan saved to the Journal — as a first-class interaction, not an afterthought.

Itinerary Refinement

The most powerful AI interaction: asking the engine to modify an existing itinerary through conversation. "Swap day 2 for something more relaxing" or "Add a vegetarian restaurant near the hotel" — the AI updates the plan in context, not from scratch.

Key Research Findings

What testing revealed

Two rounds of usability testing across the core seeker flows — onboarding, discovery feed, activity detail, and group booking — produced findings that reshaped several significant design decisions before development began.

78%

Chat prompt adoption

Of Gen Z participants used the suggested conversation starters on first interaction. Without starters, 61% typed a single vague word ("Bali") and stalled — confirming the need for guided prompts.

2.1×

Trust uplift with source labels

Millennial participants were 2.1× more likely to accept an AI recommendation when a source label was present — directly validating the "show me where this comes from" finding from persona research.

7/10

Blank input paralysis

Participants in round 1 stared at the empty chat input for more than 8 seconds before typing anything. Led to the introduction of contextual prompt suggestions and example queries on the chat home state.

94%

Round 2 task completion

After incorporating round 1 findings, task completion across the three primary seeker flows improved to 94% — the group booking flow specifically jumped from 40% to 88%.

91%

Millennial trust in sourced AI

After adding source labels and a "view details" expansion to AI recommendations, 91% of Millennial participants said they would trust the itinerary enough to use it — up from 43% in round 1.

4.4★

Overall experience rating

Average satisfaction score across test participants after round 2 (out of 5). Lowest-rated area was notification settings — added to the post-launch improvement backlog.

Key Design Decisions

Translating research into product

◎

AI feed as the default home state

Rather than a generic browse grid, the app opens directly to the personalised AI recommendation feed. This communicates Trazzeo's core value proposition immediately and forces early signal collection. Users who haven't completed onboarding see a curated "getting started" feed that improves as they interact.

◈

Journal — turning conversations into saved itineraries

The Journal is where AI conversations become real plans. Each completed chat session can be saved as a structured itinerary — with day-by-day breakdowns, places, notes, and links. Users can return to refine, share, or export their plans. This was critical for the Millennial persona, who needed a tangible output they could review and trust before committing to a trip.

◐

AI chat as the primary planning interface

Rather than a search bar or filter UI, the entire planning experience is driven through natural conversation. Users describe what they want — destination, vibe, budget, duration — and the AI builds a full itinerary in return. The chat persists across sessions so users can continue refining their plan over days, not just in a single session. In testing, this was the feature that most clearly differentiated Trazzeo from every other travel app participants had used.

◇

Design system built for scale from day one

As a solo designer-PM, I couldn't afford to redesign components repeatedly. Built a full Figma design system — typography, colour tokens, spacing scale, and a component library of 60+ elements — before building any product screens. This investment paid off immediately: every new feature design took hours, not days, because the primitives were already defined and documented. It also gave the engineering team the confidence to build UI independently without constant design review.

▣

Dark theme as brand language

The visual direction was a deliberate choice: deep navy/charcoal backgrounds with warm amber and orange accents evoke the feeling of adventure — campfire light, golden hour, the dark sky before a dawn hike. In testing, participants consistently described the app as feeling "premium" and "exciting" compared to the clinical white interfaces of competitor platforms. The dark theme wasn't just aesthetic — it was strategic brand differentiation.

Figma View the full AI Engine prototype: Trazzeo AI Engine — Figma

Final Designs

The product in action

Four key screens from the final Trazzeo product — showing the full journey from the AI discovery home through to conversational recommendations, live map integration, place detail, and the curated places grid. The AI engine (ZEO) lives at the centre of every screen.

Screen 01 — Home

AI chat & discovery dashboard

The default home state — personalised greeting, an interactive map of nearby adventures, a video inspiration feed, and six categorised AI prompt starters (Discover, Compliance, Safety, Accommodations, Videos, Restaurants). The chat input "Explore with @ZEO intelligence" anchors the entire experience, making the AI the primary entry point.

Screen 02 — AI + Map

AI recommendations with live map integration

ZEO responds to a natural language query — "Find Adventures and Restaurants near me" — with a ranked list of top campsites in Japan. The right panel simultaneously updates to a live map with pinned locations, a place detail card, and category filter tabs. AI output and spatial context stay in sync throughout the conversation.

Screen 03 — Place Deep Dive

AI chat with rich place detail panel

When a user asks ZEO for more information on a specific place, the right panel surfaces a full detail card — open hours, accessibility, amenities, a photo gallery, and a 4.1-star review summary with 532 reviews. The AI narrates the destination in depth on the left while the structured data panel on the right gives users everything they need to decide and plan.

Screen 04 — View Places

Discover more — places grid with map context

The "Discover more about these places" panel presents AI-matched destinations as a visual grid — photo cards with location, rating, and quick-action icons. A compact map sits above for spatial orientation. Users can browse, shortlist, and deep-dive without leaving the conversation context.

Outcomes & Impact

From zero to a working product

Trazzeo shipped its MVP within the first product cycle, with the AI engine powering the core discovery experience from day one. The dual PM-designer role created tight feedback loops between product decisions and design execution that a traditional split-role team structure would have been slower to achieve.

94%

Task completion rate in round 2 usability testing across core flows

2.3×

Uplift in recommendation engagement with explainability labels added

60+

Design system components built before first product screen was shipped

Beyond the product metrics, the design system and AI engine architecture created a reusable foundation that the engineering team could scale without needing design input for every new component or recommendation logic tweak — a key leverage multiplier for a small startup team.

Reflections & Learnings

What I'd tell myself at the start

→

The design system investment is never wasted at zero-to-one. Every hour spent building the component library saved three hours later. When you're the only designer and the PM, moving fast without breaking consistency is only possible if your primitives are locked before you start shipping screens.

→

AI UX is fundamentally a trust design problem. The best recommendation algorithm in the world fails if users don't trust it. "Why this?" labels, visible preference controls, and an "adventure taste" that users can see and edit were all trust mechanisms before they were UX niceties. Design the trust layer first.

→

Being PM and designer creates dangerous blind spots. When you're the one setting priorities and the one doing the design, it's easy to de-prioritise the flows you find less interesting as a designer. I had to build in deliberate check-ins to ask: "Am I deprioritising this because users don't need it, or because I don't want to design it?"

→

The cold start problem is a design problem, not just an engineering one. The first-run experience had to make users feel the AI was already smart — before it had any data. That required crafting an onboarding that was genuinely engaging, not just a data collection form. The framing ("build your adventure profile") mattered as much as the information architecture.

→

If I were to revisit: I'd invest more in the itinerary handoff experience — the moment the AI conversation becomes a saved, shareable plan. In early builds this felt like a nice-to-have. Post-testing it became clear this is actually the most emotionally significant moment in the product: when the plan stops being a chat and starts feeling real.

MAS — Licensing Dashboard

All work Bluemeg — SaaS Redesign

Building Trazzeo from Zero