Laura Maliqi
UX + Product design
3D hero image of women and men collborating on Confluence documentation

Scotland Getaways

• Personal / Self initiated • vibe coding •

A personal luxury deal scout

Scotland Getaways is a personal deal discovery platform built to solve a real, recurring frustration: missing luxury spa breaks and hotel offers that disappear before I notice them. Most travel sites are built for booking. This was built for monitoring.

Design brief (self-set): "Build a personal scout that watches my favourite venues and tells me when something worth booking appears — without me having to remember to look."

Scotland Getaways is a personal deal discovery platform built to solve a real, recurring frustration: missing luxury spa breaks and hotel offers that disappear before I notice them. Most travel sites are built for booking. This was built for monitoring.

My Role

A deliberately experimental approach to product-building. I owned the product direction, UX decisions, and quality — and used AI as the engineering partner throughout.

Me - Product Design
  • Product Manager
  • Product Designer
  • QA Tester
  • Strategic decision maker
  • Source auditor
AI (claude Code) - engineering
  • Technical architect
  • Code implementation
  • Debugging asistant
  • Infrastructure setup

The Problem

I regularly save travel ideas — spa hotels, getaway venues, places I want to visit — but consistently miss the moments when good offers appear. The existing deal landscape had real gaps:

  • Deals appear and disappear within 24–48 hours
  • I don't remember to check websites regularly
  • Notifications from deal sites are generic and poorly targeted
  • Most travel sites optimise for booking, not monitoring
  • No tool watched my specific wishlist venues
The product vision: something closer to a flight alert service or investment tracker than a booking website. Not a place to browse — a system that tells me when to act.

Build Process

1

Basic Deal Aggregation

The first version built a deal database that scraped multiple travel sources, normalised offers into a common structure, scored deals by value, and displayed them in a searchable grid. Data was stored in SQLite. Functionally, it looked like a working product.

2

The Core Promise Was Brokens

Critical finding: During a product audit, I discovered that automatic monitoring did not actually exist. Scans only ran when I manually clicked Refresh. There was no background monitoring, no overnight scanning, no alerts while away from the computer.

The product looked complete but was fundamentally failing its own brief. It wasn't a scout — it was just a searchable list I had to remember to update. This became the immediate highest priority.

Solution — Automatic Monitoring Architecture

Implemented a launchd-based monitoring system that runs automatically, survives Mac restarts, and requires no manual interaction. Four daily scans were scheduled:

Daily scans
User journey showcase

Source Status — live scan log showing all sources checked automatically, deal counts, and timestamps. No browser required

This was the transformation point: Scotland Getaways stopped being a website and became a monitoring system.
3

Notification System — And Why It Wasn't Working

The notification system was built but completely silent. After investigation, the root cause was a date format mismatch between SQLite and JavaScript that meant the query never matched newly inserted deals.

The Problem

SQLite stored: 2026-06-05 16:30:00

JS compared: 2026-06-05T16:30:00Z

Query never matched. Notifications had effectively never fired.

The Fix

Moved all date filtering into SQLite datetime functions

Added permanent deduplication — each deal notifies once only

Three tiers: watchlist · high-value · standard

Preferences page

Preferences — scan schedule, macOS notification status, and adjustable score threshold for what triggers an alert

4

Source Quality Audit

A full audit of all data sources revealed that several were placeholder or mock implementations contributing no real deals. These were removed and replaced with genuine Scottish luxury venue integrations.

Removed (mock / placeholder)

Secret Escapes (mock)

Groupon (placeholder)

VisitScotland (placeholder)

Added (real sources)

Archerfield House

Crieff Hydro

Greywalls

RiverBeds Lodges

SpaSeekers

Result: 60+ real, active Scottish luxury getaway deals — with real images, descriptions, and pricing.
5

Watchlist System

Added venue and location-level tracking so the system could prioritise the places I genuinely care about — boosting their scores, surfacing them first, and triggering priority notifications the moment a matching deal appears.

Watchlist venue

Watchlist Venues — tracks both specific venues (Archerfield, Cameron House, Crieff Hydro) and locations (Glencoe, Isle of Skye). Shows active match counts and score boosts per entry

6

Feed Quality — Expired Deal Problem

Over time, expired deals accumulated permanently. Old Black Friday offers, winter promotions, and seasonal packages remained visible long after they'd gone. A source reconciliation process was added to every scan.

User journey showcase

Each scan now reconciles live source data against the database — only currently active deals remain visible

7

UX Redesign

The initial interface had grown organically and it showed. Filters were oversized, navigation was inconsistent, and the preferences panel still referenced features that had never shipped. A full redesign addressed all of it.

Before

Oversized filter chip rows dominating the page

Poor use of vertical space

Inconsistent navigation patterns

Preferences full of orphaned settings

After

Compact dropdown filters in a unified toolbar

Toolbar: filters · scan status · nav · refresh

Refined wordmark & botanical icon

Preferences rebuilt around real behaviour

Key Product Lessons

Audit your own assumptions

The biggest breakthrough came from stepping back and questioning whether the product was actually doing what I thought it was. It wasn't. Building the habit of auditing before iterating changed how I approached every subsequent phase.

Fake data produces false confidence

Placeholder sources made the database look full and the product feel functional. Replacing them with real venue data immediately exposed what was actually valuable — and what wasn't.

Monitoring beats browsing

Users don't want another place to look. They want the system to do the looking and tell them when it matters. That insight is transferable to almost any product where the user has recurring intent but limited attention.

AI-assisted builds still need product direction

The AI could implement anything I described. But it couldn't tell me what to prioritise, what to cut, or when something was fundamentally broken. That judgment is still the designer's job.

Current State

The platform now runs autonomously. It scans four times daily, survives restarts, removes expired deals automatically, tracks a watchlist of venues and locations, and sends macOS notifications when something worth seeing appears — all without any manual input.

"Don't let me miss a good getaway offer." — Now largely achieved
Image

Future opportunities

  • Screenshot-to-watchlist workflow
  • Personalised scoring based on past saves and behaviour
  • Mobile companion app
  • Better expiry detection for time-sensitive deals
  • Expanded venue imagery

Reflection

This project was genuinely different from client work. There was no brief handed to me, no stakeholder to align with, no pre-existing system to design around. Every decision — what to build, what to cut, what to fix first — was mine.

What surprised me most was how much the work still came back to classic product thinking: understanding what the product actually needed to do, auditing against that goal, prioritising ruthlessly, and redesigning when the interface stopped serving the user. AI-assisted development changed the speed and method of execution — not the thinking that made it worth building.

The vibe-coding approach taught me that designers who can work this close to the product — making real product decisions, testing assumptions, and iterating in real time — have a different kind of leverage. I intend to keep exploring that.