Recommendation rails
“You may also like.” “Complete the look.” “Recommended for you.” Rails that personalize from a shopper’s first visit and update after their next product view.
Machine learning engineer · Coframe · San Francisco
Jonathan Gu
I build the machine learning that decides what you see next.
Recommendations, search and feeds at Coframe, built end to end: from raw shopper events to models serving live storefronts. Father of two. I think interfaces should be alive.
This shelf is a recommender. Open anything and it re-ranks around you.Every toy links to its story below.
- 01Recommendation railsRails that rank themselves.
02SearchWords in. The right products out.
03Feeds & gridsEvery grid, ordered for one person.
04The ML factoryRaw events in. Live models out.
05Head to headLive shoppers decide.
06The experiment engineEvery experiment feeds the next.
07“I do it”From “help me” to “I do it.”
08Little boat, fat boatThe little boat is Maggie.
09Anything with wheelsIf it rolls, it gets studied.
10Five books at bedtimeFive is the largest reasonable number.
11Two languagesFive little ducks, in Mandarin.
12Saturday pancakesHe mixes. She eats the cheese.
01 / 12
Details
You may also like
Cold start: these are Jonathan’s picks.
01 — The shelf, explained
A real recommender, running in this tab.
Every toy has a position in five topics. When you open one, the shelf updates a posterior over what you seem to like, scores every toy against it, keeps each row from turning into an echo chamber, and saves one slot for exploring. Then it moves the toys. It learns only from what you open. Nothing is stored and nothing leaves your browser.
01 Signals
- Nothing yet. Open something on the shelf.
02 Taste
- Machine learning38%
- Platform20%
- Experiments14%
- Play14%
- Learning14%
Dirichlet posterior with recency decay. Starts from my own prior.
03 Ranking
- Recommendation rails
- Search
- Feeds & grids
- The ML factory
- Head to head
- The experiment engine
- “I do it”
- Little boat, fat boat
- Anything with wheels
- Five books at bedtime
- Two languages
- Saturday pancakes
Affinity, a nudge toward the unseen, row diversity, one exploration slot.
Same loop. Different scale.
| Step | On this shelf | At Coframe |
|---|---|---|
| What gets ranked | Twelve toys | Live product catalogs |
| What it learns from | The toys you open | Real shopper behavior, from the first visit |
| The model | A Dirichlet over five topics | Learned embeddings and rankers |
| How it stays fresh | Updates after every open | Updates after the next product view and retrains on schedule |
| Where it runs | Your browser, in a fraction of a millisecond | A production ML API, fast enough to add no visible wait |
| How we know it works | You watch it move | Every model ships as an A/B test against what was already there |
02 — At Coframe
Recommendations. Search. Feeds.
All machine learning. All live.
Coframe uses AI to optimize and personalize websites. I build the machine learning underneath, end to end: the models that decide what each shopper sees, the platform that trains and serves them for every client, and the experiments that prove they work.
Search
Models that rerank results by what a shopper means, trained on how people actually search and click across many stores.
Feeds & grids
Collection pages and homepages ordered for each visitor. One product view and the grid reorders toward their taste.
The ML factory
Events, features, training, evaluation, publishing and serving in one place. A client’s full ML stack is a routine launch, and models retrain on schedule.
Head to head
Every model ships as an experiment against what was already on the page, including established personalization vendors in their own slots.
The experiment engine
Every test, won or lost, teaches an engine that proposes the next ideas for every client. Agents build them; people review them.
- Shopper events
- Features & embeddings
- Candidates
- Ranking
- Serving
- Experiment
- Retrain
One loop, run for every client, every day. Raw events come in, live models go out, the experiment says whether they won, and the result feeds the next model.
03 — Raising two kids
High expectations. High warmth.
Let kids meet hard things and find out they can learn them, with someone close by. Most of what I believe about learning, I believe more since becoming a father.
Not ranked.
Some things you don’t optimize.
“I do it”
The best sound in parenting is “help me” turning into “I do it.” A balance bike gets there faster than any lecture.
Little boat, fat boat
On the walk home along the water, every boat gets a job in the family. The little boat is Maggie. The fat boat, I am told, is Daddy.
Anything with wheels
Monster trucks, fire trucks, trains, cranes. If it has wheels, it requires investigation, ideally from very close up.
Five books at bedtime
Five books, every night, because five is currently the largest reasonable number in the universe. So far, the books are winning.
Two languages
Our home runs in Mandarin and English. The songs we sang to our kids, they now sing back to us.
Saturday pancakes
He mixes the batter with great seriousness. His little sister supervises and handles the cheese.
A walk home from preschool can take three hours. Every boat, fire truck and unusual machine requires investigation.
Pay attention. Try things. Keep what works. Change your mind when reality disagrees. Leave room for surprise.
The same rule, at work and at home.