jev-research-pipeline
Daily research monitor: Jev screens new papers against your open questions, code routes keep/review/drop, an LLM writes notes.
Overview
jrp (jev-research-pipeline) is a command-line research monitor for one person following a field. You keep two to four open questions per topic; each morning it takes the next topics in rotation, gathers new papers and repositories, and writes one Markdown note per topic into a folder such as an Obsidian vault, in English, Chinese or Japanese. The author reports about a tenth of the cost of an LLM agent approach.
Plain Python runs the loop. Every screening call goes to Jev via Pydantic AI's typesafe model: Noul gates (on topic, method transferable, evidence compatible) and Score dimensions per paper and question, plus picking the evidence sentence. Code routes each pair to keep, review or drop. An LLM chosen by the user writes the explanation and checks its draft once, and Jev scores the finished section. Ticks in the note steer the next day's run. Offline tests replay recorded cassettes. It is described as an unofficial, experimental pilot.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this paper on topic for this research question? | yes · no |
| Yes / no | Is the method transferable / the evidence compatible? | yes · no |
| Score | How much does this source help answer the question? | — |
How it works
- Fetch new papers and repositories for the day's topics
- Jev answers gates and scores per paper and question
- Code routes each to keep, review or drop
- An LLM writes a question-centric section; Jev scores it
- User ticks steer the next run
Strengths
- Questions steer the search rather than keywords
- Cheaper than an all-LLM agent per the author
- Recorded-cassette offline tests
Caveats
- Pilot status
- Needs a TypeSafe key and an LLM subscription or Qwen
Getting started
See the Try it section of the README; the CLI is jrp.At a glance
- Type
- Model
- Status
- prototype
- Language
- Python
- Runtime
- CLI (Pydantic AI)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Research monitoring: asks Jev Noul gates and Score dimensions per (paper, research question) on each daily fetch through Pydantic AI's typesafe model, keeps sources above a code-side threshold, and hands them to Qwen for question-centric Obsidian notes; offline tests replay recorded cassettes.
yibie/awesome-jevUnofficial experimental research monitor: Jev screens papers and other sources against each open research question (Noul gates, Score dimensions), code applies thresholds, and Qwen writes question-centric notes into an Obsidian vault
AnotiaWang/awesome-decision-models\*\*Code owns the loop, Jev judges, Qwen writes: a daily research monitor for standing questions.\*\*
logicrw/awesome-jev-projectsDaily research monitor: for every fetched paper and each of the author's open research questions, Jev answers Noul gates (on topic, method transferable, evidence compatible) and Score dimensions through Pydantic AI's typesafe: model; code routes each pair to keep / review / drop, and Qwen writes one question-centric Obsidian note per research line.
cobanov/awesome-jevA research note each morning for the topics you follow: your questions steer the search, TypeSafe Jev judges, an LLM you choose writes it (en/zh/ja). CLI: jrp (pilot).
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Whether each new paper helps answer one of your open research questions, via gates and scores.
Who writes the notes?
An LLM you choose; Jev only screens and scores.
How is it integrated?
Through Pydantic AI's typesafe model provider.
Related projects
Classifies 1,000 arXiv AI papers into 24 topics with one Jev Choice each and checks agreement against an LLM judge.
ProjectprototypePythonInteractive explorer that uses Jev to tag TC39 proposals by adoption path, cognitive overhead, web-compat risk and intent archetype.
ProjectprototypeTypeScriptAlibaba's AgentScope 2.0 agent framework, which added a JevClassifierModel for binary, choice and score decisions via the TypeSafe SDK.
ProjectproductionPythonVerifiable coding-agent runtime built on checkable execution graphs, listed with an optional Jev provider for bounded choices like routing.
ServiceproductionTypeScriptAutoGPT agent platform, listed as adding Jev decision blocks for typed routing, filtering, scoring and confidence-gated next actions.
ProjectproductionPythonAx, a DSPy-style multi-language LLM framework, listed with a native TypeSafe provider for Boolean, Choice, Score and raw Jev questions.
PluginproductionTypeScript