Every free resume builder I tried had the same two walls: create an account, then pay to download the PDF you just made. The data also leaves your machine — the whole resume, name, phone number, employer history, sometimes a photo — and lands in someone's database.
I wanted to know whether the whole thing could be done in the browser with no server at all. It can. This post is about the four technical problems that made it non-obvious, and how I solved them.
Live demo: resumake.online · Source (MIT): github.com/ConnorYyy/Resumake
The constraint
No backend. No database. No upload endpoint. The resume never leaves the browser, and the PDF is produced locally.
That rules out the usual approach (html2canvas → image → jsPDF). That pipeline produces a PDF whose content is a bitmap: it looks right, and applicant tracking systems (ATS) cannot read a single word of it. For a resume, that is fatal — most large companies parse the file before a human sees it.
So the PDF has to be real text. Which means using the browser's own print engine.
1. Composing a standalone document
The editor is a Vue 3 app, but the printed artifact must not depend on the app. So on every change I compose a complete HTML document — <!doctype html> and all — with the template CSS, the user's content and the page settings inlined:
const html = `<!doctype html> <html lang="${locale}"> <head><style>${shellCss}${templateCss}${userCss}</style></head> <body><div class="resume-page">${renderedContent}</div></body> </html>`; That string goes into an iframe via srcdoc. The iframe is the preview and the print target. One artifact, two uses — which means what you see is literally what prints, with no second rendering path to drift.
function print() { const win = frame.value.contentWindow; win.focus(); win.print(); // the browser's print dialog, including "Save as PDF" } No PDF library, no fonts to embed by hand, ~0 KB of extra JavaScript.
2. Paper size, margins and the header/footer problem
The page geometry belongs in CSS, not in JavaScript:
@page { size: Letter; /* or A4 */ margin: 15mm; /* narrow 10mm · normal 15mm · wide 20mm */ } Two things I did not expect:
@page margins and the print dialog's margins are different things. If the user leaves Chrome's "Margins" on Default, their value wins and your layout shifts. The only reliable fix is to tell the user, in the export dialog, to keep margins on Default and pick the same paper size — so I put that in the confirmation step instead of pretending CSS can control it.
Headers and footers cannot be turned off from CSS. Chrome draws the date, page title and URL into the paper margin. @page { margin: 0 } does not remove them; it just moves your content into their way. Every "why does my PDF have a URL at the top" support question has the same answer: uncheck "Headers and footers" in the print dialog. I surface that as the first, highlighted line of the export dialog, because it is the single most common way the output looks wrong.
One more: -webkit-print-color-adjust: exact on html, body. Without it, Chrome drops background colours (the header block in the "Modern" template) when printing.
3. The text-extraction bug that only shows up in an ATS
This one is my favourite.
The layout puts the company name on the left and the date range on the right, using flexbox:
<div class="row"> <span class="company">SpaceX</span> <span class="period">2002.06 – Present</span> </div> Visually: two columns. In the text layer: depending on the order the spans were painted, a parser can read
SpaceX2002.06 – Present because there is no whitespace node between them. An ATS that splits on whitespace or looks for "Company - Role - Dates" gets one glued token.
The fix is one character. Insert a non-breaking space before the period span so there is always a whitespace node in the extraction order, even though the layout is unchanged:
<span class="company">SpaceX</span> <span class="period">2002.06 – Present</span> I found it by exporting a PDF, selecting all and pasting into a text editor — which is also the check I tell users to run:
Open the PDF, select a line, copy it, paste it into a text editor. If it pastes as readable text in the right order, an ATS can read it.
It is a five-second test that catches most "my resume disappeared in the portal" problems, including ones caused by other tools.
4. Saving without a server
localStorage is too small once someone uploads a photo as a data URL, so the document lives in IndexedDB, written on a debounce:
watch(doc, scheduleSave, { deep: true }); function scheduleSave() { clearTimeout(timer); timer = setTimeout(async () => { await dbPut('current-resume', JSON.parse(JSON.stringify(doc.value))); }, 500); } The trade-off is honest: no account means no cloud sync and no password reset. The replacement is a JSON backup file the user owns — download it, import it on another machine, and the resume comes back exactly as it was. It also means there is no server-side copy to leak.
Two smaller things that matter more than they look:
- Undo/redo has to cover form edits and settings, but must not hijack
Ctrl+Zinside a text input. I check whether the event target is an editable element and let the browser handle it there. - Page-break markers: measuring the rendered iframe lets me draw where the PDF will split, and offer "fit to one page" by nudging the font scale between 80% and 120%. Users hate discovering a two-page resume only after saving.
5. The AI button, without shipping a key
There is an optional "rewrite this bullet" button. The key obviously cannot live in the bundle, so a Cloudflare Worker proxies the request:
- the key is a Worker secret, never in the client
- only the textarea being edited is sent, capped at 600 characters — never the photo, contact details or other sections
- the response is returned and discarded; nothing is stored
- the button is hidden entirely when the Worker has no key configured (the client asks
GET /api/ai/statusonce)
That last part matters for open source: anyone can fork it and run it without an AI provider, and the UI degrades cleanly instead of throwing errors.
What it costs, and what it does not do
Numbers, because "blazing fast" means nothing:
- landing pages: ~11 KB gzip in total — 6.4 KB of HTML, 2 KB of CSS, under 2 KB of JavaScript
- the editor: ~105 KB gzip including the Vue runtime
- zero PDF library, zero font files shipped
What it does not do: DOCX export (PDF only), cloud sync, and it is honestly mediocre on a phone — the split form-and-preview layout needs a desktop or tablet. Those are deliberate scope cuts, not oversights.
What I would do differently
- The brand name. I picked a name that collides with an existing product, which means brand searches will never be mine. Check that first.
- Measuring earlier. I spent weeks on SEO before installing analytics that could tell me whether anyone was arriving. Impressions before optimisation, always.
- The print dialog is the product. Half of the perceived quality of a "PDF export" is the five sentences you show before the dialog opens. I treated it as an afterthought for too long.
If you want to see the result: resumake.online — free, no account, no watermark. The code is MIT licensed on GitHub, and the print-CSS and text-extraction notes above are the parts I would most want back if I were starting again.