About Services Tools Blog Contact us

Style Matched Catalogue Search

Matches a description of a shopper and a chosen style to catalogue products for the part of the body in view, with gender and category filters.

  • Semantic Search & RAG
  • Footwear retail
  • Pipeline
  • Productised
  • Local/private
  • 2026

The problem

A photo of a shopper is not a keyword, so an ordinary product search cannot suggest garments that suit the person and the occasion. If the part of the body in view is ignored, footwear, tops and bottoms also get mixed into one set of results.

What it does

The photo booth sends a one-sentence description of the shopper, a style choice and the part of the body in view. The style adds a short hint such as smart, casual or sport. That text is matched to products by meaning with Ollama's nomic-embed-text, limited by gender and by the categories for that body area, and widened once if nothing comes back. Results are cut to one product per broad type before they are shown.

Outcomes

  • Recommendations follow the person and the occasion and do not rely on a keyword.
  • Footwear, tops and bottoms do not appear in the same set of results.
  • A gender toggle hides the other department.
  • The shopper's selection is remembered by product number.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Style match

Theme
Department

Hint added before the embedding tailored, low contrast, gallery or office

Categories for this region blouses, jackets, knitwear

Search
Ready. Feet, tops and bottoms will not share a result.

Session S-1842

Women · upper · smart
Ids held
2
Department
Women
Region
Upper
Selection
HL-2041Silk blouse£128
HL-2188Tailored jacket£210

The booth stores product ids only. The next step reads this list, not the description.

Body regions

Upper

Blouses, shirts, jackets, gilets, knitwear and coats.

Shoes and trousers are excluded.

Lower

Trousers, chinos, skirts and shorts.

A jacket cannot land in this set.

Feet

Shoes, trainers and boots.

Tops stay on the upper search.

Rules

Gender Women hides the men’s department, and the other way round.

Relax once Ankles is not a stocked region. An empty result is searched again as feet, and only once.

One per type Two blouses collapse to the higher cosine score, so the rail stays a blouse, a jacket and a knit.

How it works

  1. Receive the shopper description, the style choice and the visible body area together.
  2. Add a short hint from the style to the text used for matching.
  3. Turn that text into a meaning-based search.
  4. Limit the search by gender and by the categories for the body area.
  5. If nothing matches, relax the limits once.
  6. Cut the list to one product per broad type and keep it for the session.

Pipeline

flowchart LR
  Desc[Shopper description] --> Theme[Add style hint]
  Theme --> Vector[Embed the text]
  Vector --> Filter[Gender and body region]
  Filter --> Relax[Relax once if empty]
  Relax --> Grid[One product per type]
  Grid --> Session[Save the selection]