ihatejob

Live: https://ihatejob.vercel.app  ·  installs as an app  ·  works offline

A free, offline-first CV builder that tells you what is wrong with your CV.

Most free CV tools give you three templates, no guidance, and no way to judge whether what you wrote is any good. This one is built around four problems:

1. Not enough formats to compare. Eight templates, switchable in one click, with the same content.
2. No idea what "professional" means for your field. Forty-one profession packs and twenty regional rule sets change the section order, the section names, the target length and what counts as evidence — and each profession ships a worked sample CV you can look at.
3. No feedback. A rating out of 100 across six named characters, naming every specific problem — the exact bullet, the exact phrase.
4. Text pasted from a chatbot that the applicant cannot evaluate. The checker flags the phrases that read as machine-written and offers plain replacements.

No account, no upload, no tracking, no AI service. Everything runs in the browser and stays on the machine.

Run it

npm --prefix C:/SideProjects/ihatejob run dev

Open http://localhost:5190 for the landing page, or http://localhost:5190/app to go straight to the builder. Set PORT to change the port.

There is no build step and nothing to install. server.mjs is a static file server written against node:http. Edit a file, refresh, done.

What it does

Eight templates

Template	Layout	Best for
Classic	Single column, centred header	General applications
Minimal	Single column, wide margins	Design and senior profiles
ATS Plain	Black only, real disc bullets, photo suppressed	Job portals and parsers
Executive	Headings in a left margin column	Long careers
Modern	Colour sidebar, left	One-page CVs led by skills
Creative	Accent banner, right rail	Portfolios and studios
Academic	Long-form, numbered publications	Research records
Detailed	Dense, boxed, bio-data style	Government and PSU forms

41 profession packs

Across nine groups: Technology (software, data/ML, product, cybersecurity, IT support), Engineering (core, architecture, agriculture), Health & Life sciences (nursing, pharmacy, dentistry, physiotherapy, psychology, lab technology, veterinary), Legal & Public (law, government/PSU, social work/NGO, defence veterans), Business (finance, consulting, marketing, sales, HR, operations, banking/BFSI, insurance/actuarial, real estate), Creative & Media (design, journalism, film/VFX, content writing), Service & Hospitality (hotels, culinary, aviation, retail, skilled trades), Education & Academic, and Early career.

Every one has a worked sample CV written to pass the checker — open it in the preview without touching your own, then use it as a starting point if you want.

Choosing one rewrites the layout, the section order and the section names — "Experience" becomes Clinical Experience for nursing and Academic Appointments for research — and turns on the sections that field needs, such as licences and registrations or publications.

20 region rule sets

India, US, UK, Ireland, Canada, Australia/NZ, Germany/Austria, Netherlands, France, Nordics, UAE/Gulf, Singapore, Malaysia, Philippines, Japan, China, South Korea, South Africa, Nigeria, Brazil/LatAm.

Regions carry consequences, not decoration. A photo is conventional in Germany and disqualifying in the US, so selecting the US switches the photo off and the checker raises a blocker if personal details like date of birth appear anywhere in the document.

The rating

Upload or paste a CV and it is graded out of 100 across six characters, each with its own letter grade:

Character	What it measures
Evidence	Do your claims carry numbers, or only adjectives?
Authenticity	Does it read as written by a person who did the work?
Precision	Tight, consistent, specific lines — not padding
Field fit	Does it contain what this profession expects to see?
Machine-readability	Will screening software parse it and pass it on?
Restraint	Right length, no keyword stuffing, nothing padded

You also get a planet, from Mercury at the bottom to the Sun at the top, with a star rating for each rung — the memorable half of the same number.

It arrives as a flyby: the screen goes to deep space and the rail travels outward from Mercury, decelerating onto the planet you earned. The card lands, the score counts up and the stars pop in one at a time. The whole thing runs 0.9-2 seconds depending on how far you got, so a strong CV takes the longer trip. Click anywhere to skip it, and it is disabled entirely under prefers-reduced-motion.

Splitting the score matters: a CV can be immaculately formatted and say nothing (Machine-readability A, Evidence E), and those are different failures. Findings are then listed as fix before you send, worth fixing and polish, each quoting the offending text with a button that jumps to the field.

A sample of what it catches:

• Phrases that read as generated — leveraged, spearheaded, results-driven, passionate about, robust, seamless — with the plain replacement
• Claims with no number behind them (significantly improved)
• Bullets that describe duties rather than achievements (responsible for)
• The percentage of bullets carrying a figure, against a 40% target
• First person, essay connectives (furthermore), bullets over 34 words
• Missing sections for your profession, missing portfolio link, keyword stuffing
• Bullets that are all suspiciously the same length — one of the more reliable signs of generated text
• Reversed dates, mixed date formats, employment gaps over six months
• Page count against the profession and region target

Every rule is deterministic and runs offline. Nothing is sent anywhere.

Cover letters

A second document sharing the CV's name block, typeface and accent, on its own tab. Fill in the role, company and recipient; the greeting, subject line and address block assemble themselves. Start from a structure lays out the four paragraphs that do the job, with every claim left as a [bracketed prompt] for you to replace — the editor counts how many are still unreplaced, because brackets print exactly as they appear. Exports as PDF, Word, HTML or text under its own filename.

Clean up automatically

The rating tells you what is wrong; Clean up automatically fixes what a machine can fix safely — replacing leveraged with used, deleting empty claims, stripping a leading I, normalising date formats to the region's convention, removing personal details a region forbids, and making bullet punctuation consistent. Each fix is listed with a count and can be unticked.

Two things are deliberately not automated: rewriting "Responsible for developing X" needs verb morphology that turns ran into runned, and adding numbers to bullets would mean inventing facts. Those stay flagged for you. A guard also reverts any rewrite that would leave broken English — a clause that lost its subject, or a line cut to almost nothing.

Import an existing CV

Drop in a PDF, a .docx, or paste the text. The file is read in the browser — .docx is unzipped with DecompressionStream, PDFs are decoded from their content streams. Text is parsed into sections, roles, dates, skills and scores, shown as a report for you to confirm, and then checked.

Getting text out of a PDF takes more care than it sounds:

• DecompressionStream rejects trailing bytes that zlib ignores, and discards the whole result when it does. A PDF stream is delimited by the endstream keyword, so the slice nearly always carries the writer's newline — meaning every compressed stream failed and the reader saw nothing. It now keeps whatever decoded before the complaint.
• Embedded fonts are binary and large enough to contain the bytes Tj by chance, which made them look like content streams and drowned the real text. The object's own dictionary says what it is: /Length1 means a font program.
• Subset fonts store glyph ids, not letters. Each carries a /ToUnicode CMap; the reader resolves every font resource to its map and translates.
• A word processor emits a separate run per formatting change, so breaking the line at each ET split words in half — "Subham" became "S" + "ubham". Line breaks come from the text matrix moving down the page instead.

Parsing the structure out of that text is still approximate, and the app says so. A scanned PDF has no text to find at all — it tells you to paste instead rather than producing nonsense.

Import from LinkedIn

Most people arrive with a LinkedIn profile and no CV. Two routes in, both handed over by the person themselves, because no LinkedIn API returns a member's positions or education — those endpoints are Recruiter and CRM partner only, and scraping the site breaches its terms and risks their account.

Their profile as a PDF (More → Save to PDF) is the front door: instant, and the layout is fixed enough to parse exactly rather than guess at. The parser reads the contact rail, top skills, languages, certifications, the identity block, summary, roles with dates and locations, and education. If it finds no roles it falls through to the generic parser rather than producing a worse result.

Real profiles are messier than the shape suggests, and the parser handles what one actually contains: headlines that wrap over three lines, a profile URL split mid-word, durations printed on their own line, a company named once above several roles beneath it, a description whose last word wraps onto a line that then looks exactly like the next company, degrees spread over three lines with the dates on a fourth, entries the PDF repeats across a page break, and "Page 1 of 2" arriving as four separate lines.

npm test runs the import suite — 36 checks over both LinkedIn routes and the date parser, including the guardrails that stop an ordinary CV being pushed through the LinkedIn one.

Their data archive (Settings → Data privacy → Get a copy of your data) is cleaner — real CSV fields, no guessing — but LinkedIn takes 10 minutes to 72 hours to send it, so it is the second route. Drop the .zip in and it reads Profile, Positions, Education, Skills, Certifications, Languages, Projects, Email Addresses and Phone Numbers.

The archive also contains messages, connections, ad targeting and LinkedIn's inferences about the person. Only the profile files are opened, and the import report names exactly which ones, so nothing is read quietly.

Fix my LinkedIn

The reverse trip, under More. Nothing can write to a LinkedIn profile, so this sizes the text to LinkedIn's own fields instead — headline 220 characters, About 2,600, each role 2,000 — counts against those limits live, and flags filler with the same phrase lists the CV checker uses, via checkPhrases in review.js. One rule set, so the two can never disagree.

It never rewrites anything for you. Inventing a number you did not give would be inventing a fact about your career.

Downloads

• PDF via the print dialog (set Margins to None, Background graphics on)
• Word .doc — opens in Word and Google Docs. Sidebar templates flatten to one column; Classic, Minimal, ATS Plain and Executive convert faithfully
• Self-contained .html
• Plain .txt for portals that reject formatted uploads
• .json backup you can re-open later

Known limitation: multi-page printing

Page padding lives inside the document rather than in @page. That is what lets Modern and Creative bleed colour to the sheet edge — the cost is that a second printed page loses the top padding and the sidebar background. The preview draws a red line at each page break and warns when a sidebar format runs past page one. For longer CVs use Classic, Minimal, ATS Plain, Executive or Academic.

Dark and light

Both themes ship, following your system preference on first load and remembered after that. The CV page itself stays white in both — it is paper, and it has to print.

Install it as an app

The site is a PWA. A banner appears at the top of the page on arrival offering the install, and there are quieter controls for anyone who dismisses it — the toolbar, the mobile menu, a card in the share section, and the builder's More menu — all sharing one data-install contract in public/js/pwa.js, and all appearing only when the browser genuinely offers an install.

The banner has an ×, and dismissing it is remembered: it will not reappear for 30 days, and turning down the browser's own prompt counts as the same answer. It sits in the page flow rather than over it, so it pushes the nav down instead of covering it, and inside the builder it is a flex child of the app shell so the workspace shrinks around it.

The quieter controls open a dialog with Cancel and Install rather than firing the browser's prompt immediately. The prompt can only be shown once, so someone who taps a bare "Install the app" menu item to find out what it means should be able to read that and back out without spending it. Cancel takes focus, not Install.

The banner skips the dialog, because it already says what installing does — its Install goes straight to the browser's prompt.

On iOS Safari, which has no install API at all, the same dialog shows the three Share-sheet taps instead, with the Install button removed — there is nothing for it to call.

Once installed it runs in its own window, starts at the builder, offers Rate my CV and Build a CV as icon shortcuts, and works with no network at all.

Deploying

Already deployed to Vercel. The current deployment is unclaimed — run npx vercel login then npx vercel claim to move it into your account before attaching a domain. Full details, including how to add your domain and when to bump the service worker version: docs/deploying.md.

Admin

Moderate reviews and journeys, read feedback, and check the site's own setup. Full guide: docs/admin.md.

Protected by a username and password checked in middleware.js at the edge, before any file is served — a password checked in browser JavaScript is not a password, it ships to every visitor. Set ADMIN_USER and ADMIN_PASSWORD in the Vercel project settings, and optionally ADMIN_PATH to move it to an address only you know. With no password set, the admin refuses to load rather than standing open.

Saving commits public/data/*.json back to GitHub, so every change is a commit you can revert, and there is no database to run.

It cannot show you anyone's CV or who visited — none of that is collected.

Job journeys

/stories is where visitors write up the whole hunt - how many applications, how long the silence lasted, what finally got a reply - and other people read and reply. It replaced a blog, on the reasoning that nobody comes to a CV tool to read us; they come to find out whether any of this worked for anyone.

Someone submits from the form on /stories, which opens a public issue labelled story. You publish it from the admin. The issue then stays the story's discussion thread, and that is the whole backend for likes and replies:

• The like count is the real +1 and heart reactions on that issue.
• The replies shown under a journey are the real comments on it, fetched from the public GitHub API and rendered read-only.
• Replying goes to the thread, where the person's own account is the identity.

Nothing here is a number we can inflate, no account system exists to build, and if GitHub rate-limits the request the counts are simply absent rather than shown as zero. Comment bodies come from strangers, so they are escaped before any markup is applied and links are restricted to http(s) and root-relative paths.

Drafts stay off /stories until you tick Published. /blog permanently redirects here.

Before you launch it

Three things live in public/js/config.js:

Setting	What it does
SITE.repo	Where the suggestion form and every "open an issue" link point. Change your-username or those links stay disabled.
SITE.contactEmail	Fallback inbox for reviews and suggestions before the repository is public. Set at least one of these two, or visitors have no way to reach you.
STATS.endpoint	A JSON summary of your analytics. Visitor numbers appear once it returns real figures.
STATS.minVisitors	Below this, visitor numbers stay off the page entirely. Default 100.
REVIEWS	Ships empty on purpose.
MIN_REVIEWS	The reviews section appears once you have this many. Default 3.

Where it all arrives: docs/receiving-feedback.md — which page to watch, the labels to create first (GitHub silently drops labels that do not exist), and how to turn on notifications.

How visitors send you a review. Reviews have their own section and their own form, separate from the issue form under Open source. A review and a bug report are different acts by different people; sharing one dropdown between them made both read as paperwork.

The rating is a planet, picked from the same Mercury-to-the-Sun scale the checker uses on a CV, so nobody has to think in numbers. The form writes a fixed header that the admin reads back, which is how the planet, the credit and the consent answer survive the trip through GitHub:

Rating: Jupiter (5 of 10, 2.5 stars)
Credit: Asha R - Staff nurse, Leeds
May be quoted on the site: yes

...the review itself

Stars are always derived from the planet — half a star per rung — rather than stored beside it, so the two cannot drift apart.

Submitting goes to the repository if SITE.repo is set, otherwise to SITE.contactEmail, and if neither is configured it copies to the clipboard and says plainly that nothing was sent. Set one of them before launch.

On the repository link. While SITE.repo still contains your-username, every "open an issue" control detects it and scrolls to the on-page suggestion form instead of sending someone to a GitHub 404, pre-selecting the right category. Point it at a real repository and they become direct issue links.

On visitor numbers. The front page can show visitors and CVs rated, read from whatever you point STATS.endpoint at - Plausible, Umami, GoatCounter and Fathom all expose a summary API, and a Cloudflare Worker holding a counter in KV works too.

Until the endpoint is set, or while the figures are below minVisitors, those cells are not on the page at all - no dash, no placeholder, no explanation. A visitor sees only the four facts that are true from day one. "17 visitors" is worse than saying nothing. The reason is logged to the browser console, where only you will look.

On reviews. No testimonials ship with this. A brand-new site showing five glowing quotes from people who do not exist is the fastest way to lose a visitor, and it is the first thing anyone checks.

The whole section - heading, nav link and all - stays out of the page until MIN_REVIEWS are visible, then appears on its own. The review form stays on the page throughout, so people can send you reviews before the section exists. Each entry needs a name, a role, a planet, and ideally a link to where they said it.

Privacy

Nothing leaves the machine. The CV lives in localStorage under ihatejob.v1. Uploaded photos are centre-cropped to a 420px JPEG and stored in the same record. There is no analytics, no telemetry, no font CDN and no upload endpoint. More → Clear everything wipes it.

Contributing

New professions, regions, templates and checker rules are the most useful contributions, and the first two need field knowledge more than code.

• CONTRIBUTING.md
• Adding a profession
• Adding a template
• Where reviews and suggestions arrive
• Deploying and installing as an app

Layout of the code

server.mjs                static file server (node:http only); "/" is the site, "/app" the builder
public/index.html         landing page
public/app.html           the builder
public/css/site.css       landing page styles
public/js/site.js         landing page logic
public/js/config.js       domain, repo, analytics endpoint, reviews
public/admin.html         admin dashboard (public data only)
public/js/admin.js        admin logic: health, GitHub inbox, integrity checks
middleware.js             edge auth for the admin: password, secret address
api/content.js            saves reviews and journeys back to the repository
public/stories.html       job-journey listing and single-journey view
public/js/stories.js      journeys, GitHub likes and replies, a small Markdown subset
public/data/*.json        reviews and journeys, edited from the admin
public/js/pwa.js          install prompt and service worker registration
public/sw.js              offline cache
public/manifest.webmanifest  app identity, icons, shortcuts
vercel.json               hosting config: routing, headers, caching
public/css/app.css        editor UI
public/css/cv.css         the document: structure, .tpl-* skins, print rules
public/js/schema.js       data model, form schema, sample CV
public/js/professions.js  profession packs and region rules
public/js/templates.js    CV rendering and plain-text export
public/js/review.js       the checker
public/js/import.js       PDF/DOCX extraction and CV text parsing
public/js/form.js         builds the editor from the schema
public/js/main.js         state, events, preview, storage, exports

Licence

Proprietary. All rights reserved — see LICENSE.