Case study 01 / Full-stack fragrance discovery
ScentConcierge
A personal way to find a signature scent.
A fragrance discovery experience that connects a curated catalogue, budget-aware recommendations and enquiries to an owner-managed backend.

- DiscoverSearch and filter
- CompareNotes and budget
- EnquireValidated request
- ManageOwner-only access
Discovery before a transaction.
The service is a middle-person perfume sourcing concept. The prototype helps visitors explore 50 fragrances, compare scent profiles, build a shortlist and ask about a perfume. It deliberately stops short of promising inventory or accepting payment.
Budgets are optional numeric inputs rather than fixed spending tiers. Recommendation rules respect the entered range and explain their connection to the selected perfume.
A small frontend, backed by a real service.
The storefront uses browser-native JavaScript and a unified responsive stylesheet. Cloudflare Pages serves the assets, Pages Functions handles API requests and D1 stores the catalogue and enquiries.
Cloudflare Access restricts administration to the owner. Server-side validation and authorization protect the management routes; hiding admin controls in the interface is not treated as access control.
Motion that still lets people browse.
Two campaign scenes transition automatically, with touch scene selection and desktop bottle interactions. System reduced-motion settings and a persistent motion control disable movement.
Filters, product details and the shortlist use native dialogs. The same discovery rules carry a chosen budget into related recommendations.
Test the decisions that could mislead a visitor.
Seventeen automated tests cover catalogue integrity, arbitrary and reversed budgets, recommendation constraints, enquiry validation, authorization and cross-origin writes. The Functions build is checked separately.
Browser checks covered five widths from 320 to 1440 pixels, the shortlist and filter flows, scene rotation and reduced motion. These checks are not a full screen-reader or physical-device audit.
A prototype with honest boundaries.
The current recommendation engine compares listed scent characteristics. It does not claim an exact sensory match or replace trying a sample.
Supplier integrations, customer identity, payment processing and fulfilment remain outside scope. The repository is private; the live prototype and this case study provide the public walkthrough.
Source reviewed at revision 26dbf77. The repository is private; the live prototype is available above.