Smart Matching: How Shadar Finds Exactly What You Want

You see a pair of shoes on a friend. You snap a photo. Within seconds, Shadar shows you vendors nearby who carry that exact style — or something strikingly similar. It feels like magic. It's actually vector math.

When you submit a request — whether it's a photo, a text description, or both — our system generates an embedding: a 256-dimensional vector that captures the essence of what you're looking for. Color, style, category, texture, context — all encoded into a single mathematical representation.

On the other side, every vendor listing on Shadar has its own embedding, computed when the item is first posted. We use pgvector to store these embeddings directly in our database, enabling lightning-fast similarity searches across thousands of listings using cosine similarity.

But raw similarity isn't enough. A perfect visual match from an unreliable vendor isn't useful. So our matching engine applies a composite ranking that combines three signals: embedding similarity (how close the match is), vendor score (from our Vendor Intelligence system), and delivery feasibility (can this vendor actually get it to you quickly?).

The result is a ranked list of vendors, ordered not just by who has the closest match, but by who can deliver the best overall experience. A slightly less perfect match from a top-rated vendor with fast delivery might rank higher than a pixel-perfect match from someone with a history of delays.

We chose this approach over traditional category-based search for a reason. West African fashion doesn't fit neatly into predefined categories. An Ankara print blazer might be 'menswear,' 'African fashion,' 'formal,' and 'contemporary' all at once. Embeddings capture this nuance naturally, without requiring rigid taxonomy.

As more requests flow through the system, our matching improves. We track which matches lead to accepted offers and which don't, feeding this signal back into the ranking algorithm. The system learns what 'good match' means to real buyers in real markets.