Engineering
Notes from the team building Favora: the crawlers, the catalogue, and the models that make fashion discovery personal. Real systems, live numbers.
Live catalogue Latest catalogue activity 19 Nov, 03:30 CET
Language models as fashion curators
How well do language models curate fashion for real shoppers? We kept the agent that curates Discover Daily fixed, ran 8 models through it, and judged every one of their 4,460 picks blind.
Read the piece
The Favora dataset
How we turn the fashion internet into one machine-readable catalogue, why we build it ourselves, and the current scale of the dataset.
Read the piece
The stack
One Python pipeline on Specific, with DBOS for durable jobs. Postgres for the catalogue, Typesense for keyword, faceted and vector search. An LLM labels and embeds every item, gated on a hand-labelled gold set. SvelteKit on the web, Expo in the app.
Working on consumer AI, search, or crawling at scale? We are hiring. For everything else: humans@favora.ai