Shortlist

AI-powered job search, end to end. Shortlist scrapes listings from six sources every morning, scores them against your profile with AI, lets you tailor your resume with one click, and tracks your entire pipeline — all in one place.

Built by John Moorman as a portfolio project. Currently single-user; SaaS roadmap in progress.

---

Features

• Scored feed — Every listing is evaluated against your background and given a 0–100 match score with a plain-English verdict: GO, EXAMINE, or NO_GO. Weak matches are hidden automatically; strong ones surface to the top.
• AI resume tailoring — Side-by-side job description and resume editor. Click Generate and Claude streams a tailored draft in real time. Writing rules let you protect specific phrases, ban clichés, and prevent the AI from inventing achievements you don't have.
• PDF export — Export any tailored resume directly to PDF from the browser. No third-party service.
• Pipeline tracker — Kanban-style table: Saved → Applied → Interview → Offer. Tracks follow-up dates, notes, and the tailored resume used for each application.
• Import anything — Paste a URL or raw text from any job listing. Claude extracts the structured fields. Works with any company, not just supported ATS platforms.
• Multi-profile — One account, multiple independent job searches. "Frontend Berlin" and "Automation Remote" get completely separate feeds, criteria, and pipelines.
• Daily scraping — Six sources (Greenhouse, Lever, Ashby, USAJobs, Adzuna, Arbeitnow) scraped every morning via Vercel Cron. Pool-first architecture deduplicates globally before matching per-profile.

---

Tech Stack

Concern	Choice
Framework	Next.js 15 App Router
Language	TypeScript (strict)
Styling	Tailwind CSS v4
Database	Neon (PostgreSQL)
ORM	Prisma
Auth	Clerk
AI	OpenRouter — multiple models (see AI pipeline below)
Scraping	Greenhouse, Lever, Ashby, USAJobs, Adzuna, Arbeitnow
State	Zustand (client), React 19 useOptimistic
Testing	Playwright (E2E), Vitest + React Testing Library (unit)
CI/CD	GitHub Actions (typecheck → lint → unit → Playwright)
Scheduling	Vercel Cron (daily at 7am UTC)
PDF	@react-pdf/renderer
Markdown	@uiw/react-md-editor
Deployment	Vercel

---

Architecture

Pool-first scraping

The scraper writes to a global JobPool table, deduplicated by (source, externalId). A separate matching pass then creates Job junction rows — one per profile that matched the listing. This means the same posting is never fetched or stored twice regardless of how many profiles exist, and changing matching logic means re-running the match, not re-scraping.

JobPool  ──────────────────────────────────────
  └── Job (junction: one per profile match)
        ├── aiScore, feedStatus, userNotes
        └── Application → TailoredResume[]

User (Clerk ID)
  └── Profile (search context)
        ├── targetRoles, targetLocations, skills
        ├── curriculumVitae, masterResume
        ├── protectedPhrases, bannedPhrases     ← writing rules
        └── neverClaim, verifiedMetrics

AI pipeline

• Scoring — Batches of 5 jobs, 500ms between batches, anthropic/claude-haiku-4.5. Pre-filter rejects excluded keywords without an API call. Response: { score, status, summary, matchPoints[], gapPoints[] }.
• Tailoring — Streaming, qwen/qwen3.5-397b-a17b. Uses the full CV as content source and the master resume as format template. Writing rules are injected as hard constraints.
• Extraction — One-shot, anthropic/claude-haiku-4.5. Fetches a URL, strips HTML noise, converts to markdown, extracts structured fields.
• Triage — Batch classification of borderline candidates, google/gemini-2.5-flash.

All models are user-overridable per profile via Advanced Settings.

---

Local Development

Prerequisites

• Node.js 20+
• pnpm (npm i -g pnpm)
• A Neon database (free tier works)
• A Clerk application
• An OpenRouter API key

Setup

git clone https://github.com/mojoro/shortlist
cd shortlist
pnpm install

Copy the environment template and fill in your values:

cp .env.example .env.local

Run the database migration and start the dev server:

pnpm prisma migrate dev
pnpm dev

Seed the database with realistic mock data (optional):

curl -X POST http://localhost:3000/api/dev/seed

Environment Variables

Variable	Description
DATABASE_URL	Neon pooled connection string
DATABASE_URL_UNPOOLED	Neon direct connection string (migrations only)
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY	Clerk publishable key
CLERK_SECRET_KEY	Clerk secret key
CLERK_WEBHOOK_SECRET	Clerk webhook signing secret (optional)
OPENROUTER_API_KEY	OpenRouter API key
CRON_SECRET	Random string — protects /api/scrape and /api/analyze
USAJOBS_API_KEY	USAJobs scraper (optional)
USAJOBS_EMAIL	USAJobs scraper (optional)
ADZUNA_APP_ID	Adzuna scraper (optional)
ADZUNA_APP_KEY	Adzuna scraper (optional)
NEXT_PUBLIC_APP_URL	Full URL of the deployment (e.g. http://localhost:3000)
NEXT_PUBLIC_DEFAULT_THEME	light | dark | system

Triggering a scrape manually

curl -X POST http://localhost:3000/api/scrape \
  -H "Authorization: Bearer YOUR_CRON_SECRET"

Add ?skipPool=1 to skip re-scraping and re-run the matching pass against the existing pool.

---

Project Structure

src/
  app/
    (auth)/               # Clerk sign-in / sign-up pages
    (dashboard)/
      dashboard/          # Job feed — default view
      jobs/[id]/          # Job detail + match analysis
      tailor/[jobId]/     # Resume tailor — JD vs resume, streaming, export
      pipeline/           # Application tracker (table + Kanban board)
      settings/           # Profile + Account tabs
    api/
      scrape/             # POST — pool scrape + profile matching
      analyze/            # POST — AI scoring for unscored jobs
      tailor/             # POST — streams tailored resume
      tailor/save/        # POST — persists tailored resume draft
      jobs/extract/       # POST — AI field extraction from URL/text
      jobs/import/        # POST — saves a custom job listing
      webhooks/clerk/     # POST — Clerk webhook (user lifecycle)
    onboarding/           # Onboarding wizard for new users
    page.tsx              # Landing page
  components/
    dashboard/            # FeedToolbar, ProfileSwitcher
    jobs/                 # JobCard, JobFeed, ScoreBadge, ImportJobModal
    tailor/               # TailorPanel, GeneratePane, ResumePDFDocument, PDFPreview
    pipeline/             # PipelineTable, KanbanBoard, ApplicationDrawer, StatusSelect
    settings/             # SettingsClient, UsageSection, FeedbackForm
    layout/               # AppNav (collapsible sidebar + mobile tabs)
    landing/              # HeroDemoPreview, FeatureRow, LandingNav
    onboarding/           # OnboardingWizard
  lib/
    prisma.ts             # Prisma client singleton
    openrouter.ts         # OpenRouter client (server-only)
    models.ts             # Model constants + getModels() helper (client-safe)
    match-sql.ts          # SQL-based pool matching + rematch
    normalize.ts          # Source raw data → JobPool schema
    validations.ts        # Zod schemas for all API request bodies
    store.ts              # Zustand store (dashboard state)
    scrapers/             # greenhouse, lever, ashby, usajobs, adzuna, arbeitnow
  config/
    app.ts                # APP_CONFIG — app name lives here only
    companies.ts          # Company lists + search configs for all scrapers
prisma/
  schema.prisma           # Source of truth for DB schema
tests/
  *.spec.ts               # Playwright E2E tests
  unit/                   # Vitest unit tests
  global-setup.ts         # Playwright global setup — seeds test data
.github/
  workflows/ci.yml        # CI: typecheck → lint → unit → Playwright

---

CI/CD

Every push to main and every pull request runs the full CI pipeline via GitHub Actions:

Step	What it does
Type check	pnpm tsc --noEmit
Lint	ESLint
Unit tests	Vitest + React Testing Library
E2E tests	Playwright (runs against a Neon dev branch database)

Playwright tests run after all other checks pass. Test reports are uploaded as artifacts and retained for 14 days. Vercel auto-deploys main to production and PR branches to preview URLs.

---

Contributing

Contributions are welcome. This project uses the Business Source License (see below), so you can run it locally and submit pull requests freely — but you may not deploy it commercially without permission.

If you want to contribute:

1. Open an issue first for anything non-trivial. Describe what you want to change and why. This avoids wasted effort on PRs that won't be merged.
2. Fork, branch, and PR. Branch names follow type/kebab-description (e.g. feature/email-notifications, fix/feed-filter).
3. One concern per PR. Don't bundle a bug fix with a refactor.
4. Pass CI. Run pnpm tsc --noEmit and pnpm test:unit before submitting.

Areas where contributions are especially useful:

• New scraper sources (job boards with public APIs)
• Accessibility improvements
• Mobile UX
• Test coverage

---

Roadmap

• Clerk webhook — user lifecycle events (created, updated, deleted)
• Onboarding wizard for new users
• Multi-profile support with independent feeds and pipelines
• USAJobs, Adzuna, and Arbeitnow scrapers (6 sources total)
• Kanban board view for pipeline
• User-configurable AI model overrides
• Email notifications for new high-score matches
• LinkedIn scraper
• SaaS billing (Stripe)

---

License

This project is licensed under the Business Source License 1.1 (BUSL-1.1).

What this means in practice:

• You can read, clone, fork, and contribute to this repository freely.
• You can run it locally for personal, non-commercial use.
• You cannot deploy it as a hosted service or use it commercially without a separate agreement.
• On March 17, 2028 (two years after initial release), the license automatically converts to Apache 2.0 — fully open source, no restrictions.

BUSL is not an OSI-certified open source license. It is source-available. The intent is to keep the code readable and contributable while reserving commercial rights until a potential SaaS launch.

See LICENSE for the full text.

---

Author

Built by John Moorman.