magiCV

Local CV tailoring tool. Edit your projects in a SQLite-backed web UI, paste a job description, and get a tailored 3–5 page PDF (DE + EN) rendered from HTML via headless Chromium.

Setup

bun install         # JS dependencies only
bun run setup       # downloads the Playwright Chromium build (~100 MB, one-time)
bun dev             # vite on :5174, bun server on :5173, open http://127.0.0.1:5174

bun run setup is a thin wrapper for bunx playwright install chromium. It's kept out of postinstall so bun install stays fast and predictable in CI and sandboxes; you only pay the Chromium download once, when you actually want to render PDFs.

First run starts with an empty DB. Open the UI and fill in your Profile + Projects; the data is persisted to data/magicv.db (gitignored).

Build & run as a single process

bun run build
bun start        # bun serves the SPA + API on :5173

Requirements

• Bun 1.3+
• Claude Code CLI (claude) on PATH — used for the JD-extraction and tailored-summary LLM calls.
• Playwright Chromium — install once with bun run setup (downloads ~100 MB into Playwright's cache).

Optional

• Ghostscript (gs on PATH) — post-processes the generated PDF to shrink + linearize it (~700 KB → ~110 KB). The render pipeline detects gs automatically and skips the step if it's missing.
  • Install: brew install ghostscript (macOS) · apt install ghostscript (Debian/Ubuntu).
  • Tune output size via MAGICV_PDF_PRESET env var: /screen (smallest), /ebook (default), /printer, /prepress.

How it works

[Browser SPA]  ←HTTP→  [Bun server on 127.0.0.1:5173]
                              │
                              ├─ bun:sqlite       → data/magicv.db
                              ├─ child_process    → `claude -p --json-schema …`
                              └─ playwright       → renderHtml() → page.pdf()

• Profile, projects, skills, quotes, runs live in data/magicv.db (SQLite).
• Tailored CV flow: paste JD → claude extracts must-have / should-have skills + writes a tailored 1–3 sentence summary → server matches skills against your project history → renders an HTML document → headless Chromium produces the PDF.
• Default CV flow: skips the JD step, uses summary_default_de/en + featured_skills from your profile.
• PDF features: teal-accented sidebar (photo, contact icons, languages, education, hobbies), skill-match table on page 1 with internal anchor links to project blocks on subsequent pages, page numbers in bottom-right.

Useful scripts

bun run check             # typecheck + lint + format-check + knip
bun run typecheck         # tsc --noEmit + svelte-check
bun run lint              # oxlint
bun run format            # prettier --write .
bun run build             # vite build → web/dist
bun run db:generate       # drizzle-kit: generate a migration from schema changes
bun run db:studio         # drizzle-kit: web UI for inspecting the DB

Layout

magicv/
├── server/
│   ├── index.ts             # thin Hono app: CORS + route mounts
│   ├── routes/              # one Hono sub-app per resource (profile, projects, llm, render, runs, …)
│   ├── db.ts                # drizzle-backed query functions
│   ├── db/
│   │   ├── schema.ts        # drizzle table definitions (source of truth)
│   │   └── connection.ts    # opens SQLite + runs drizzle migrations on boot
│   ├── schema.ts            # zod schemas for API request/response shapes
│   ├── extract.ts           # JD → must/should-have skills via `claude -p --json-schema`
│   ├── summary.ts           # tailored summary via `claude -p --json-schema`
│   ├── match.ts             # buildMatrix + pickProjectsForRender
│   └── render.ts            # renderHtml() + Playwright → PDF
├── drizzle/                 # generated SQL migrations (committed)
├── templates/
│   └── cv.html.ts           # HTML/CSS template (one module, both languages)
├── web/                     # Svelte 5 + Vite SPA
│   └── src/
│       ├── App.svelte       # hash-based router (tailor / projects / profile)
│       ├── routes/
│       └── components/
├── assets/
│   ├── profile_placeholder.svg
│   ├── icons/               # mail/phone/linkedin/github/map-pin (Tabler, tinted teal)
│   └── uploads/             # gitignored — user-uploaded profile photos
├── data/                    # gitignored — magicv.db (your source of truth)
└── out/                     # gitignored — generated PDFs

Backing up your data

data/magicv.db is your source of truth. Copy it anywhere for a backup.

License

MIT — see LICENSE.