Jev PR Reviewer

PR triage with structured decisions

Turn Pull Requests into structured engineering decisions.

Jev analyzes PR context and returns structured decisions for risk, type, human review, and testing — ready to power automated developer workflows.

Review a Pull RequestHow it works
  1. PR Contexttitle, files, diff stats
  2. Jevstructured questions
  3. Structured Decisiontype · risk · review · tests
  4. Developer Actionlabels, reviewers, gates

Review a Pull Request

Demo: enter PR context by hand (or load the demo PR) and ask Jev directly.

Pull Request Context

The signals Jev will evaluate.

Pull Request details
3 files
  • src/auth/middleware.ts
  • src/auth/token.ts
  • tests/auth/token.test.ts

Press Enter to add. Paste a multi-line list (e.g. git diff --name-only) to add many.

PR metrics

Repository signals

Ready to review your Pull Request

Paste your PR context and let Jev turn it into structured engineering decisions.

  • typebug | feature | refactor | …choice
  • risk0 ─── 3score
  • needs_human_review0.0 … 1.0noul
  • needs_tests0.0 … 1.0noul

The demo PR is loaded and ready.

How it works

From PR context to actionable signals.

No walls of generated text. Just structured decisions your workflow can use.

Traditional AI review

Unstructured

  1. Pull Request

  2. Long generated response

  3. Human interprets the response

  4. Human decides what to do

With Jev

Structured

  1. Pull Request

  2. Structured questions

    • type: choice
    • risk: score
    • needs_human_review: noul
    • needs_tests: noul

  3. Structured decisions (example values)

    type.choice
    "bug"
    risk.score
    1.84
    needs_human_review.noul
    0.86
    needs_tests.noul
    0.46

  4. Deterministic workflow

    • label: high-risk
    • request reviewers
    • gate merge

GitHub Actions integration

Illustrative workflow
  1. 01GitHub Pull RequestOpened or updated
  2. 02Collect PR ContextTitle, description, files, diff stats
  3. 03Jev APIPOST /api/v1/decisions
  4. 04Structured Decisiontype · risk · needs_human_review · needs_tests
  5. 05GitHub AutomationLabels, reviewers, merge gates

The questions this app asks Jev

Every analysis sends the same four typed questions. Each answer type has a fixed shape, so the UI — and any automation — can consume it without parsing prose.

  • typechoice

    Classify the type of this pull request.

    • bug
    • feature
    • refactor
    • documentation
    • other

    One of a fixed set, with a probability for each

  • riskscore

    Score the engineering risk of this pull request.

    • 0 Low
    • 1 Moderate
    • 2 High
    • 3 Critical

    A number on an ordered scale, with a distribution

  • needs_human_reviewnoul

    Does this pull request require careful human review before merging?

    • 0.0 … 1.0

    A yes/no probability between 0 and 1

  • needs_testsnoul

    Should additional automated tests be required before merging this pull request?

    • 0.0 … 1.0

    A yes/no probability between 0 and 1

API / Raw Decision

What goes to Jev, and what comes back.

The review above is rendered from this JSON. The raw response is shown exactly as Jev returned it.

  1. BrowserPOST /api/analyze
  2. Next.js routevalidates · holds API key
  3. JevPOST /api/v1/decisions

The Jev API key is read from server environment variables and never sent to the browser.

request.json
{
"state": {
"pr_title": "Fix JWT refresh bug",
"pr_description": "Fixes an issue where expired JWT tokens are not refreshed correctly.",
"changed_files": [
"src/auth/middleware.ts",
"src/auth/token.ts",
"tests/auth/token.test.ts"
],
"files_changed": 3,
"tests_added": true,
"lines_added": 82,
"lines_removed": 24,
"has_database_changes": false,
"has_authentication_changes": true
},
"questions": {}
}
About Jev PR ReviewerWhat this project demonstrates

Jev PR Reviewer demonstrates how structured decisions from Jev can transform Pull Request context into actionable engineering signals.

  1. GitHub Pull Request
  2. PR Context
  3. Jev
  4. Structured Decisions
  5. Developer Workflow

Built as an exploration of structured AI decision-making for developer workflows.

Created by Muhammad Shahrukh Khan · AI Full Stack Engineer

Built by

Muhammad Shahrukh Khan

AI Full Stack Engineer

muhammadshahrukhkhan87@gmail.com

Building AI-powered developer tools and intelligent engineering workflows.