Machine Learning Engineer, Search

Shelfwise runs online storefronts for 1,900 independent bookstores. A customer searching "that novel about the lighthouse keeper's daughter" should find it, even when they've got the title wrong and the shop has only one copy.

You'll own search ranking across all our stores. In the first 90 days you'll set up offline evaluation with real queries, ship a hybrid ranker that combines keyword and embedding search, and add "similar books" recommendations that respect each store's actual stock. We use Python, PyTorch, OpenSearch and a feature store on Postgres.

The team is seven engineers across New York, London and Berlin, overlapping from 14:00 to 17:00 UTC. Model changes come with a written evaluation report, and we hold an optional reading group every other Thursday.