Coffee Product Attribute Classifier

Oscar Gabriel used Jev to build a hotdog/not-hotdog-style classifier for coffee products. Its Nouveau backend sends Jev typed questions based on each product page, scraped as HTML with Firecrawl.

View on X

oscar gabriel@oscabriel𝕏

I used @typesafeai jev to make hotdog / not hotdog but for coffee except it's a lot more than that. the nouveau backend sends jev one map of typed questions per coffee product page, html scraped with @firecrawl closed facts like roast level are a Choice from a pre-defined list. open facts like farm and altitude are a Choice, too, but the options are the page's own contents, split up into individual lines for the open facts, like tasting notes, some pattern-matching rules propose candidates within those lines, and a Noul gives a yes/no probability for each one. then, some final code cuts th

Sep 24, 2026X postsView on X

Roast level is selected from a predefined list. For open facts such as farm and altitude, the options come from the page’s contents split into lines. For tasting notes, pattern-matching rules propose candidates, and a Noul assigns each a yes/no probability.

Also filed under Research & data

Jev-Powered Research Paper Radar
JEV Paper Radar is a scheduled research-paper radar that asks Jev whether each paper matches configured interests and exclusions, then publishes ranked web and feed output.
Jev Reranks Knowledge Search Results
You.com uses Jev through OpenRouter as a reranking layer for Knowledge in its Web Search API, selecting relevant results between retrieval and synthesis. The setup used 3x fewer tokens and achieved 84% accuracy on the Vertical RTK benchmark.
Simulates Text Attention
The build uses JEV to simulate the attention mechanism by approximating an attention score for each pair of words in a text. It treats texts as bidimensional entities.
Hermes Researches YouTube, X, and LN
Nabendu Biswas uses Jev in a Hermes agent to research YouTube, X, and LN.