NewThinking about AI? Book a free consultation and we'll show you where it can save your team time.Talk to us
AI search demoeCommerce

AI Product Search

Store search that understands what shoppers mean, running live in your browser

AI Product Search on desktop, laptop and phone
48
sample products searched by meaning
384
numbers describe each product’s meaning
23 MB
open-source model that runs on your device
0
searches sent to a server
Live demo

Search a sample store by meaning, not keywords

Ask the way a shopper would. On the left, ordinary keyword search. On the right, a small AI model that runs inside your browser and understands what you mean, so the right products show up even when they use different words.

The AI model (23 MB) downloads once and runs on your device. Nothing you type is sent to us or anyone else.

Keyword search

What most store search boxes do: find products that contain your exact words.

Pick an example or type a question.

AI search

Understands what you mean, even when the product uses different words.

Pick an example or type a question.

A sample store with 48 made-up products. The open-source model (all-MiniLM-L6-v2) downloads once from Hugging Face, about 23 MB, and your searches never leave your device. On a real store we index your full catalogue, add filters, synonyms and sales data, and tune results with your search logs.

01

The challenge

Shoppers describe what they need ("keep my coffee hot on a hike"), but most store search only matches exact words, so a flask described as "holds drinks warm for 12 hours" never shows up. Every search that comes back empty or wrong is a sale lost.

02

Our approach

Every product is turned into an embedding, a list of 384 numbers that captures its meaning, by an open-source sentence model. Each search becomes an embedding too, and products are ranked by how close the meanings are. In this demo everything runs on your device with WebAssembly; on a live store the product embeddings sit in a vector database next to the catalogue and update as products change.

03

What we built

  • Search by meaningFinds the flask for "keep my coffee hot on a hike", though the product never mentions coffee, hot or hiking.
  • Side-by-side proofKeyword results next to AI results for the same question, on the same products.
  • Private by designThe model runs in the browser, so searches never leave the shopper’s device.
  • No per-search feesAn open-source model instead of a paid AI API.
  • Ready for real cataloguesThe same pipeline works for Shopify, WooCommerce and custom stores, with filters, synonyms and sales signals on top.
04

The results

On the same 48-product sample store, keyword search finds nothing useful for most real questions, while AI search puts the right products first in milliseconds, with no server and no per-search fees. The same approach scales to full Shopify, WooCommerce or custom catalogues with a vector database.

Inside the build

Designed for the phone first

Most shoppers arrive on mobile, so every key page was designed and tested at phone size.

AI Product Search: Example questions on mobile
Example questions
AI Product Search: AI vs keyword results on mobile
AI vs keyword results
AI Product Search: More results on mobile
More results

Want results like AI Product Search?

Tell us what you're working on. We'll reply within one business day with ideas and a free quote.