CV Builder

• 

  Keep one profile, not five CVs

  Every role, project, skill and school lives in one profile, whether or not it fits on a page. A CV is a selection from it, so tailoring one for an application is picking, not retyping.

• 

  Start from a template that already has your content

  Twenty templates render with your own profile straight away, so you are choosing a layout you can actually read. Colours and fonts swap before anything is created.

• 

  Edit on a real page

  Blocks are dragged onto an A4 canvas, the layers tree keeps the structure visible, and the inspector edits whatever is selected, down to margins in millimetres.

• 

  Export a PDF that matches the screen

  The PDF is the same paginated document the editor draws, with selectable vector text. There is no second layout engine, so nothing shifts between preview and file.

• 

  Keep a CV per application

  Every CV in one list, each card drawn from the document itself, with its page count and when it was last edited. Duplicate one and adjust it for the next role.

App

• React 19
• TypeScript
• Vite
• React Router
• TanStack Query
• Tailwind CSS
• Base UI

Editor

• Zustand
• Immer
• dnd-kit
• Tiptap

Document

• Zod schemas
• Migrations
• Paginator

Backend

• Supabase Auth
• PostgreSQL
• Row-level security
• Storage
• Edge Functions

Server

• Cloudflare Workers
• Browser Rendering
• Workers KV
• Claude API

Quality

• Vitest
• ESLint
• TypeScript strict

Hosting

• Vercel
• Cloudflare

From a tree of blocks to A4 pages

1. One document

   A normalised tree of nodes, validated by Zod and saved as a single jsonb document. A schema version and code migrations keep old CVs opening.

2. Measure

   The tree is rendered once as a continuous layout, and every block is measured.

3. Paginate

   The paginator splits it into A4 fragments: between the children of a box, column by column inside rows, between list items. Keep together, keep with next and page breaks are honoured.

4. Render anywhere

   The same paged document draws the editor canvas, the preview and a bare print route.

5. Print

   A Cloudflare Worker opens the print route in headless Chromium and returns a vector PDF. Without it, the browser print dialog does the same job.

Reading an old CV back in

1. Read in the browser

   pdf.js (loaded only when needed, so the editor never carries it) extracts the text and rebuilds lines from glyph positions.

2. Ask for structure

   The Worker sends the text to Claude with a strict JSON schema and gets the semantic CV data back. The API key never reaches the browser.

3. Choose a look

   The parsed CV fills the template gallery, so every template previews with your own content.

4. Create

   Nothing is stored until you press Create CV. Then it opens in the editor, fully editable.

Decisions that keep it on free tiers

• Postgres as a document store

  A CV is one jsonb column whose shape belongs to the Zod schema, so a new element type never touches SQL. Relational columns exist only for the list page and ownership, and the list never downloads the documents.

• Saves that notice a second tab

  Every save goes through a small Postgres function that only writes if the revision is the one the editor loaded. If another tab got there first, the editor asks whether to reload or overwrite instead of silently losing work.

• One schema package for app and Worker

  @cv/schema holds the schemas, defaults, presets, templates and migrations, and is shared by the web app and the Worker, so both agree on what a document is.

• Limits where the free tier has them

  PDF rendering is capped per user per day with counters in Workers KV, falling back to browser print. Supabase pauses idle free projects, so the Worker runs a daily cron that keeps it awake.

• Auth links that explain themselves

  Every e-mail link lands on one callback route that finishes the sign-in, or says clearly that the link expired, was already used or was opened in another browser, instead of bouncing to the sign-in page.