Curriculum Generator

Edit your content, not your layout. A self-hosted CV builder with live PDF preview, an ATS score, and a job-search toolkit built in.

Features · Quickstart · Architecture · Deployment · Roadmap

---

Why

Most CV tools hand you a design and a paywall. This one gives you a typed data model, three templates driven by the same design tokens, and checks that tell you how a parser and a recruiter will read the result. Your data stays on your server: no SaaS lock-in, no analytics SDK, no telemetry.

Features

Write

• Live PDF preview while you type, no rebuild loop
• Three templates: ats (single column, parser-friendly), premium (single column) and modern (two columns)
• One design system for PDF and HTML: colors, fonts, spacing and accent style come from a single Theme
• Single-page auto-fit that reports exactly what it trims, or natural multi-page pagination
• Per-CV locale (fr / en): section headings, date formats and labels follow the language
• Editor comforts: drag and drop reordering, undo/redo (50 steps), autosave, snapshot history, ⌘S save, ⌘P preview, ⌘K command palette

Check

• ATS score (0 to 100) from deterministic checks, no AI call: missing sections, summary length, skill overload, unproven skills, weak or missing verbs, metrics, date gaps and overlaps, stack per job
• ATS text preview: the plain text a parser extracts from your PDF
• Job match: paste an offer, see which keywords are proven in your experience and which are only listed in your skills

Export and share

• PDF (A4, @react-pdf/renderer) with proper metadata and a download name based on your name
• DOCX export, localized like the PDF
• Public share link with an HTML view and a PDF download, set to noindex

Assist (optional, needs ANTHROPIC_API_KEY)

• OCR import: drop an old CV (PDF or image) and get a structured draft
• Chat assistant with tool calling: rewrite bullets, add experiences, retarget the CV. Every change is shown as a diff before it is applied. See docs/ai.md

Search

• Job sources and postings, company discovery and enrichment, application tracking, and outreach with batch sending, delivery webhooks and one-click unsubscribe

Run it yourself

• Multi-user sign-in with GitHub OAuth or magic link (Better Auth)
• Docker Compose + Caddy on your own VPS, secrets encrypted with SOPS

Stack

Layer	Tools
App	Next.js 15 (App Router), React 19, TypeScript, Tailwind CSS v4
Data	PostgreSQL 16, Drizzle ORM, Zod
Documents	@react-pdf/renderer, docx
Auth	Better Auth
AI	Anthropic SDK
Quality	Vitest, Playwright, ESLint, Prettier, GitHub Actions
Ops	Docker, Caddy, SOPS + age

Quickstart

Requires Node 20+, pnpm 9+ and Docker.

pnpm install
pnpm db:up                      # PostgreSQL in Docker
cp .env.example .env            # then set BETTER_AUTH_SECRET: openssl rand -hex 32
pnpm db:migrate
pnpm dev                        # http://localhost:3000

Without SMTP_HOST, magic-link emails are printed in the dev console: copy the link from the terminal to sign in. Without ANTHROPIC_API_KEY, only OCR import and the chat assistant are disabled.

Browse /showcase to see the templates and design tokens without signing in.

Architecture

app/                 Routes: auth, dashboard, public CV, API
  api/cv/[id]/pdf     Streamed PDF, with an X-CV-Fit header
  api/cv/[id]/docx    DOCX export
  api/cv/[id]/chat    AI assistant (NDJSON streaming, tool-calling loop)
  api/cv/import       OCR extraction (PDF, PNG, JPG, WebP)
  api/jobs, companies, outreach, cron, webhooks
lib/
  cv/                 Domain: Zod schemas, ATS checks, job match, labels, repository   (pure)
  pdf/                React-PDF templates, tokens, fit cascade, render                  (pure)
  ai/                 Anthropic client, OCR extractor, agent tools
  jobs/ companies/ outreach/   Job-search toolkit
  auth/ db/           Better Auth config, Drizzle schema and client
components/           cms/ (editor), preview/ (PDF and HTML), ui/ (primitives)
docker/               Entrypoint and Caddyfile
secrets/              SOPS-encrypted env files (templates committed)
tests/                unit/ (Vitest), e2e/ (Playwright smoke)

Three rules keep it tidy:

• lib/cv/ has no React and no PDF dependency.
• lib/pdf/ takes validated CvData and knows nothing about the database.
• UI components never reach the database directly: they go through Server Actions or route handlers.

The ats template follows parser constraints on purpose: text only, no icons, no tracked headings (which extract as P R O F I L), standard section names and a body size that never goes below 9 pt.

Development

pnpm test          # unit tests (Vitest)
pnpm test:e2e      # Playwright smoke test, starts its own dev server
pnpm typecheck
pnpm lint

The ATS thresholds live in lib/cv/ats-checks.ts and are documented against docs/research/cv-watch-2026.md.

Deployment

Designed for a small VPS running Docker.

One-time setup

1. Install Docker and Docker Compose on the VPS.

2. Install SOPS and age locally and on the VPS.

3. Generate a key with age-keygen -o ~/.config/sops/age/keys.txt. Put the public key in .sops.yaml and copy the private key to the same path on the VPS.

4. Fill in secrets/.env.production.example and secrets/.env.postgres.example, save them without the .example suffix, then encrypt:

   sops -e -i secrets/.env.production
   sops -e -i secrets/.env.postgres

5. Commit the encrypted files. Plaintext ones are gitignored.

Deploy. GitHub Actions builds the image, pushes it to GHCR, then connects to the VPS over SSH to decrypt the secrets and run docker compose pull && docker compose up -d --remove-orphans.

Required Actions secrets: DEPLOY_HOST, DEPLOY_USER, DEPLOY_SSH_KEY, DEPLOY_PATH, plus SOPS_AGE_KEY if you decrypt in CI rather than on the VPS.

Backups: see docs/ops/backup.md.

Roadmap

• Profile photo upload with a resize pipeline
• Targeted CV variants (full stack, freelance, lead) sharing one pool of experiences
• Import from LinkedIn export and JSON Resume
• Per-user AI usage tracking and rate limits
• Observability stack (Loki, Grafana)

License

MIT © 2026 sycatle