Skip to main content

How SniffBot learns

SniffBot is not just a chat box on top of a catalogue. It learns from four places:

  1. The fragrance brain — ScentSell's catalogue and knowledge base.
  2. Your cabinet — what you own, samples included.
  3. Your wear history — what you actually reach for, when, where, and why.
  4. Your feedback — ratings, compliments, thumbs up/down, saved facts.

The fragrance brain​

SniffBot can search:

  • the ScentSell fragrance catalogue
  • fragrance-family and layering docs
  • blind-buy safety notes
  • Australian niche perfumery notes
  • historical fragrance books and knowledge docs
  • ScentSell help and product docs

It does not load every document into every answer. It searches the knowledge base for the most relevant chunks, then answers from those.

Your taste profile​

get_taste_profile is the private read model SniffBot uses to understand you. It looks at:

  • top notes, houses, perfumers, and families in your cabinet
  • gender mix in your cabinet
  • high-rated and low-rated accords
  • what you wear in different weather and times of day
  • what you wore recently, so it does not keep repeating the same bottle

Wear context​

Wear logs are more useful when they include context:

  • time_of_day, derived from your timezone when missing
  • occasion
  • weather and temperature
  • rating, including decimals like 7.75
  • sprays
  • layers
  • notes
  • compliments

That is how SniffBot can answer questions like:

What do I reach for in humid weather?

What gets noticed at work?

What should I wear tonight if I want something clean but noticeable?

Compliments​

Compliments are a strong positive signal. A rating tells SniffBot what you liked. A compliment tells it what other people noticed.

SniffBot uses compliments to understand:

  • your most complimented fragrances
  • what your partner seems to like
  • which scents get noticed in certain contexts
  • what to recommend when you want something noticeable

Thumbs up and thumbs down​

If you thumbs-down a SniffBot answer, ScentSell stores a short-lived suppression so SniffBot avoids repeating the same recommendation direction. It does not permanently rewrite your taste from one tap. It nudges the next few answers.

Saved facts​

You can explicitly tell SniffBot to remember or forget things:

Remember I hate heavy patchouli.

Save that my partner likes clean musks.

Forget what I said about vanilla, I actually like it when it is dry.

These become private memory facts tied to your account.

What SniffBot should do before recommending​

For a broad "what should I wear?" question, SniffBot should check weather, cabinet, wear history, and taste profile first.

If your plan is still unclear, it should ask one short question:

Indoors or outdoors, skin-scent or noticeable, clean or loud today?

For layering, it should ask what the layer should do if you have not said:

Freshen it, sweeten it, soften it, deepen it, make it cleaner, make it sexier, tame sweetness, or add longevity?

Then it uses score_layering_combo to give a score, reasons, warnings, and a spray plan.

Privacy​

Your cabinet, wear history, ratings, compliments, and memory facts are private to your account. MCP clients only get access after you authorise them.

Next steps: