Collection Explorer
Browse and filter fragrances by brand, perfumer, season, time, mood and scent characteristics.
Live product
Project record 01
Updated 18 Jul 2026
Personal product · Data, recommendation & web systems
Hihie's Scent OS
A personal fragrance operating system that turns a large collection into a usable daily decision tool—combining structured data, weather and location context, mood, discovery and layering ideas.
01 · Overview
The collection was growing. The decision process was not.
The project began with a simple personal friction: owning many fragrances created more choice, but not necessarily better or faster decisions each morning.
Scent OS reframes a fragrance collection as a structured system. Instead of browsing bottles randomly, the product connects fragrance attributes with the surrounding context and returns suggestions that can be understood, explored and improved.
It is also a long-running product laboratory: a place to combine fragrance knowledge, data modeling, recommendation logic, interface design, SEO and automation in one coherent project.
02 · Capability map
The product is more than a digital shelf.
Each capability converts collection data into a different form of practical value.
Context Advisor
Use weather, location and day/night context to narrow the collection toward more suitable choices.
Mood & Ritual Picks
Translate an intended mood or ritual into a recommendation rather than relying only on note families.
Layering System
Suggest two-fragrance combinations and explain the intended contrast, bridge or supporting role.
Fragrance Detail Model
Present notes, perfumer, year, longevity, projection, temperature fit and contextual attributes in one view.
Search & Knowledge Layer
Connect the product to public project pages, structured metadata, sitemap coverage and AI-readable context.
03 · System architecture
A lightweight stack with an editable source of truth.
The architecture stays intentionally simple: structured data can change without requiring a heavy application backend.
Google Sheets
Editable fragrance records, attributes and image references act as the operational content source.
Structured rows
Data Layer
Values are cleaned, mapped and converted into a consistent model for filtering and recommendations.
Usable dataset
Weather & Location
Open-Meteo and location selection add temperature, rain, local time and season context.
Context signals
Recommendation Engine
Rules connect fragrance properties, user intent and context while preserving enough variety for discovery.
Ranked suggestions
Static Web Product
Responsive HTML, CSS and JavaScript deliver the product through GitHub Pages and optimized assets.
Usable interface
Operating principle:
Keep editing simple, keep delivery fast, and add infrastructure only when the product genuinely needs it.
04 · Key decisions
The strongest project knowledge is stored in the trade-offs.
DR-01Accepted
Use Google Sheets as the editable content layer
A personal collection changes frequently and should not require editing large HTML or JSON files manually.
- Why
- Fast editing, familiar workflow and a single visible source of truth.
- Trade-off
- Schema discipline and validation become essential as the sheet grows.
DR-02Accepted
Keep the experience static-first
The product needs speed, reliability and inexpensive hosting more than a complex application framework.
- Why
- Lower maintenance, strong performance and straightforward GitHub deployment.
- Trade-off
- Advanced personalization must be designed carefully on the client side or added later.
DR-03Revised
Move image delivery away from fragile raw URLs
A growing image collection created rate-limit and reliability issues when loaded directly from GitHub raw paths.
- Why
- CDN delivery, WebP assets and lazy loading improve stability and page speed.
- Trade-off
- Image naming and mapping need stronger operational control.
DR-04Removed
Remove Scent Quiz from the main experience
The quiz made the page longer without creating enough value for a personal collection-focused product.
- Why
- Prioritize direct discovery, recommendations and useful collection interactions.
- Lesson
- A feature can be attractive but still dilute the product's real job.
DR-05Expanded
Use selectable global location instead of a fixed city
A fixed weather location becomes inaccurate when traveling and weakens trust in recommendations.
- Why
- Local weather and day/night context should follow the user's actual or selected destination.
- Trade-off
- Location UX and timezone handling require additional care.
DR-06Ongoing
Explain recommendations instead of returning names only
Repeated or surprising suggestions feel arbitrary when the product does not expose its reasoning.
- Why
- Explainability improves trust and makes discovery educational.
- Next
- Strengthen scoring transparency and variety controls.
05 · Product evolution
Each version solved a different layer of the problem.
Collection interface
Turn the fragrance list into a visual, searchable personal library.
Context-aware advisor
Add weather, season, day/night and mood to move from browsing toward decisions.
Editable data system
Shift collection maintenance to Google Sheets and improve image delivery.
Knowledge and discovery
Add perfumer views, richer details, SEO pages and clearer product explanation.
Product refinement
Remove weak features, improve mobile behavior and focus recommendations on trust and variety.
06 · Lessons learned
The product improved when the data model improved.
Recommendation quality begins in the dataset
Rules cannot compensate for inconsistent attributes, vague categories or missing context.
Personalization needs an explanation layer
A suggestion is more useful when the user understands why it fits and how it differs from alternatives.
A spreadsheet can be a serious CMS
For a focused personal product, familiar editing and disciplined structure can outperform unnecessary backend complexity.
Performance is part of product design
Image delivery, lazy loading, mobile layout and dependency choices directly shape the experience.
Removing features is product progress
Eliminating low-value sections can make the core product more coherent than adding another function.
A personal project can still use operating discipline
Decision records, version history and measurable quality prevent experimentation from becoming disorder.
07 · Next development priorities
Move from recommendation features to recommendation quality.
The next phase focuses on stronger data semantics, more transparent scoring and better control over repeated suggestions.
Normalize recommendation attributes
Create clearer scales for freshness, temperature fit, projection, formality, mood and artistic character.
Add diversity and repetition controls
Prevent a few strong candidates from dominating every context while preserving relevance.
Explain the scoring path
Show the most important reasons a fragrance was selected and the factors that reduced its score.
Build a lightweight preference memory
Use recent wear, skips and favorites to adjust suggestions without requiring account infrastructure.
Formalize product analytics
Measure which discovery paths, filters and recommendations create meaningful interaction.
Project record status
First encyclopedia record complete.
Scent OS now has a durable public record covering its purpose, architecture, decisions, evolution, lessons and priorities.