Trail runner
$120
Live demo Runs in your browser
Your current
Browse the shelf. Your profile notices how long you look at each item, not only what you click, and the Picked for you slot changes as it learns. This page runs the same decision code as the product, in your browser. Nothing you do here leaves it.
The rest of this site uses the same profile. Each page is tagged with who it is for and what it is about, and the time you spend reading it adds to your vector, under Audience and Topic. The button that suggests where to go next, at the bottom of each page, is picked from it.
The shelf
Hover or focus a card to look at it. Longer looks count for more.
Court classic
$95Canvas low
$60Storm runner
$140Penny loafer
$160Suede driver
$130All-weather loafer
$175Chelsea boot
$190Hiking boot
$210Rain boot
$70Slide sandal
$45Leather sandal
$85
How it works
One profile, one read, one decision
Events carry their dimensions
Each view says what was viewed: style, material, features, color. The profile needs no product catalog, so it works on the day you install it.
The profile keeps a decayed vector
Two scores per value: this session's interest, which fades in minutes, and who you are, which fades over weeks. It is a few hundred bytes, updated in the write the profile already makes.
Decisions weigh expected value
Every candidate gets a probability of success from your vector and a model shared across all visitors, times what success is worth. The decision explores a little and says why it chose. Here the model learns in your browser from what you add to the bag; in the product, a learner that fits each slot's model from logged outcomes is next.
Demo time is compressed: this session's interest halves every 2 minutes and the long-term score every day. In the product they are set per project.
Decide on how people actually browse
A slot can serve by affinity today, from the same profile that holds the rest of the person's history.