Structured catalogue
The product catalogue captured and stored as structured records.
A conversational product search assistant for an industrial supply startup
The challenge
Buyers were filtering a very large electronics catalogue by hand to find matching parts, and each enquiry took too long.
- Keyword and filter search did not match how buyers describe what they need.
- The catalogue was too large to search manually at scale.
- Recommendations needed admin control over which products to promote.
What Winjit built
Vector search
Embeddings for every product and every query, with the ten closest matches retrieved by similarity.
Language model answers
The top matches passed to a language model that returns a precise recommendation.
Metadata filtering
Filters that cut the candidate set by around 99% before search runs.
Admin tagging
Tags that let the team mark preferred products and shape recommendations.
TechnologyPython · Node.js · Angular · MongoDB
Results
1M+
products searchable by conversation
About 99%
smaller search set before retrieval
Top 10
matches ranked for every query