[ Use case · Research ]
Can Jev filter papers, articles and news for relevance?
Yes. Relevance filtering is a natural Score question: describe what you care about, and Jev rates each abstract or headline against it. One public run classified 1,018 research papers for about 8 cents at a 256 ms median response time.
Is it a fit?
Use Jev when
- You follow a feed (arXiv, RSS, news) that is mostly noise for you.
- You can describe your interests in a paragraph.
- Titles and abstracts are enough to judge relevance.
Skip it when
- You need summaries of what you keep. That's a language model's job.
- Relevance depends on reading full papers; cost rises with input length.
The questions
One request per paper, 3 questions. Jev answers them in parallel.
- Score
How relevant is this to a team building cheap, fast AI routing and classification in production?
Scale: Irrelevant → Tangential → Relevant → Must read
- Choice
What is the main topic?
Options: Routing · Evaluation · Training · Applications · Other
- Noul (yes/no)
Does this include results a practitioner could apply this month?
Options: Yes · No
Getting started
[ 1 · Test it in the playground ]
No code needed. Paste a real paper into TypeSafe's playground and ask: “How relevant is this to a team building cheap, fast AI routing and classification in production?” with the options Irrelevant, Tangential, Relevant, Must read. You'll see the answer and its probability.
Open the playground (opens in a new tab)[ 2 · Use an existing tool ]
We haven't found a ready-made tool for this workload yet. These directories track what's been built on Jev:
Made with Jev (opens in a new tab)Jev.Store (opens in a new tab)JevHunt (opens in a new tab)
[ 3 · Integrate it ]
To run it on every paper automatically, call Jev from the system where your papers live. The developer details below include the full request.
What people have reported
Classifying AI research papers
self-reported1,018 papers for 8¢ (7.9¢ per 1,000), 256 ms median latency
Hassan El Mghari, developer · Made with Jev (opens in a new tab)
Self-reported figures are the author's own; we haven't reproduced them. More on the costs page.
Common questions
- How much does it cost to filter research papers with Jev?
- About 2 cents per 1,000 abstracts at list price. One reported run classified 1,018 papers for about 8 cents.
[ For developers ]Request body, wiring and pitfalls
The request
One request per paper with all 3 questions batched over the same state, so Jev reads the state once. Keys below are the names you'll see in the response.
{
"model": "jev-latest",
"state": "Title: Calibrated Early-Exit Classifiers for Low-Latency Routing\nAbstract: We study confidence calibration for small classifiers that route requests between models, and show that temperature scaling on 2k labeled examples cuts misroutes by 31%...",
"questions": {
"relevance": {
"type": "score",
"instructions": "How relevant is this to a team building cheap, fast AI routing and classification in production?",
"criteria": [
"Irrelevant",
"Tangential",
"Relevant",
"Must read"
]
},
"topic": {
"type": "choice",
"instructions": "What is the main topic?",
"criteria": {
"routing": "Model routing and cascades",
"evaluation": "Benchmarks, evals, calibration",
"training": "Training methods and data",
"applications": "Applied systems and case studies",
"other": "Something else"
}
},
"practical": {
"type": "noul",
"instructions": "Does this include results a practitioner could apply this month?"
}
}
}How to wire it up
- 1.Pull new items from your feeds once a day.
- 2.Send title + abstract (or headline + first paragraph) as state.
- 3.Keep items with relevance ≥ 2; put practical ≥ 0.6 at the top of your digest.
- 4.Rewrite the relevance instructions whenever your focus changes — it's the only tuning you need.
Watch out for
- Very broad interests make everything "relevant". Be specific about what you don't want.
New to writing Jev questions? Read the question design guide.
Related use cases
Can Jev classify keywords by search intent?
Label keyword lists by intent and relevance, at any volume.
Can Jev classify ad copy and creative angles?
Tag ads by angle and hook strength, and pre-check them against policy.
Can Jev power semantic search?
Re-rank search results by how well each one matches the query.