A free ATS friendly resume checker and builder that runs entirely in the browser. It reads the resume you already have, compares it against a job posting, shows you the plain text a parser extracts from your document, and prints a PDF that screening software reads correctly.
Live: resume.nicolaslecocq.com
An applicant tracking system extracts the text from your file, sorts it into fields such as employer, job title and dates, then scores those fields against the requisition. Most resumes fail at the sorting step rather than the scoring step, and the usual reason is a two column layout with a skills sidebar: the extractor reads the page in one pass and interleaves the columns, so your job title ends up filed under your employer. The rejection email arrives before a person has read a line, and it never tells you which step failed.
This tool shows you the step that failed.
| Parser view | Renders the plain text an extractor pulls out of your document, next to the document itself. If a heading or a date looks wrong here, it is wrong in the employer's system. |
| Posting match | Paste the job description. The checker separates required from preferred terms using the posting's own headings and cues, then reports how many of the required ones your resume already contains. |
| PDF import | Drop in your current resume. The text is extracted with pdf.js inside your tab and sorted into fields you then correct. |
| JSON round trip | Export your resume as JSON and import it next time, so a new application costs you the posting step and nothing else. |
| Three templates | All single column with standard headings, because that is what parses. They differ in density, not in structure. |
| Real text layer | Printing goes through the browser, so the PDF keeps a text layer. A rasterised PDF is invisible to every parser. |
It does not write hidden keywords in white text, and that is a deliberate decision rather than a missing feature.
Parsers strip styling and read the raw text layer, so a hidden block lands in the middle of the recruiter's screen looking exactly like what it is. Screening platforms flag the pattern, and a flagged application can be rejected automatically with a fraud marker attached to the candidate record, which follows the applicant across every future role at that company. The same applies to hidden instructions aimed at an AI reader.
Putting the same words in visibly, inside the summary and inside real bullets, scores identically and survives a human reading it. That is what the posting step does. When a required term is missing, it is reported as missing, and the person decides whether it is true of them before it is added.
There is no database, no account, no analytics cookie and no server that receives your file. Your
resume, your photo and the posting you paste are held in the browser tab and in sessionStorage,
which the browser clears when the tab closes. The JSON export is the only copy that outlives the
session, which is why it exists.
The site is a static bundle. Nothing on the page can send your data anywhere, and you can verify that
by reading src/scripts/ or by opening the network tab.
git clone https://github.com/DigiHold/parseable.git
cd parseable
npm install
npm run dev # http://localhost:4321npm run build # static bundle into dist/
npm run preview # serve the built bundle
npm run deploy # build, then wrangler deployThe output is a folder of static files, so any host works. The repository is set up for Cloudflare Workers with static assets, where static asset requests are not billed.
npx wrangler login
npx wrangler deployThen point a hostname at the Worker: Cloudflare dashboard → Workers & Pages → your worker → Settings → Domains & Routes → Add → Custom domain. Cloudflare creates the DNS record and issues the certificate itself, so there is nothing to configure by hand as long as the zone is on Cloudflare.
Change name in wrangler.jsonc and site in astro.config.mjs to your own before deploying.
src/
pages/index.astro the whole page: markup, copy, JSON-LD schema
styles/global.css Tailwind v4 theme tokens and the component layer
styles/sheet.css the resume sheet and the print rules, plain CSS on purpose
scripts/app.ts wizard state, forms, import and export, printing
scripts/parse.ts pdf.js text extraction and the heuristics that sort it into fields
scripts/keywords.ts posting analysis, required against preferred, the match score
scripts/render.ts resume data to sheet HTML, and the parser text approximation
scripts/types.ts the resume model and the import normaliser
Two notes on the choices, since they look inconsistent at first glance:
The resume sheet is styled in plain CSS rather than Tailwind because its markup is generated as an
HTML string at runtime, where Tailwind's scanner cannot see the class names. The same reasoning
explains the small component layer in global.css, while static markup uses utilities directly.
pdf.js is loaded through a dynamic import, so the 400 KB library is fetched only when someone actually drops a PDF in. The initial page ships around 32 KB of JavaScript.
Every control in the tool is exercised end to end with Playwright, against a preview or against production, using a real resume:
npm run build && npm run preview &
npm run test:e2e -- http://localhost:4321/ path/to/your-resume.pdfIt imports the PDF, zooms, switches views, edits fields, adds, moves and removes rows, switches templates, toggles the photo, runs an audit and adds a missing term, exports the JSON and imports it back into a fresh tab, reloads, fills the print root, and drops a file on the hero.
Issues and pull requests are welcome. Two things to know before you open one:
The extraction heuristics in parse.ts are the weakest area, and a failing real world resume is the
most useful bug report there is. Attach the PDF only if you are comfortable making it public, or
describe the layout and paste the text pdftotext -layout gives you.
Pull requests that add hidden text, invisible keywords or prompt injection aimed at AI screeners will be closed. The reasoning is in the section above.
MIT. See LICENSE.
Built by Nicolas Lecocq.