How this works, and why it's built this way

JobHunter AI — how it works, and why it's built this way

The two-sided problem, honestly stated

A job board is worthless without both sides at once. You cannot post jobs nobody applies to, and you cannot collect seekers with nothing to apply to. Most new job boards solve this by faking the empty side. We did the opposite.

Job data: we index live postings from public job boards (currently Bayt and LinkedIn) and send seekers to the original listing to apply. We are a wayfinder, not a middleman, for jobs we did not receive directly.

Candidate data: there is no fake candidate pool. A profile only exists because a person created it. Employers see opt-in profiles only, and the number of real candidates is displayed honestly, even when it is zero.

That means the product is useful on day one (hundreds of real, current roles) and becomes more valuable as it accumulates its own postings and profiles.

What we do with AI, and what we refuse to do

Three hard rules, enforced in the code rather than promised in a blog post:

1. 1No invented experience. Every generated line is grounded in a line of the person's own CV. If a bullet needs a number the CV does not contain, the tool asks for it.
2. 2Human approval before anything leaves. CVs, cover letters and outreach emails are drafts until the person clicks send. Daily caps prevent accidental spam.
3. 3Nothing protected is ranked. The matcher never scores age, gender, nationality, religion, marital status or photos. Every score can be overridden by a human, and the override is logged.

The trail

The product is organised as a trail with five waypoints, because a job hunt is a sequence, not a feed:

Discover → Prepare → Apply → Interview → Offer

Each waypoint has one job. Discover ranks and explains. Prepare builds and checks the CV. Apply drafts and waits for approval. Interview rehearses. Offer negotiates. Nothing tries to do all five at once.

CV Hub

• Builder — six guided steps with a live preview, autosaved locally, export to PDF. Enforces the rules from the guides: one column, dates as years, measurable bullets, no "responsible for".
• Analyzer — paste a CV and a posting; get keyword coverage, format risks that break parsing, and a score. Runs entirely in the browser.
• Guides — eight specific pieces covering structure, ATS keywords, achievement bullets, employment gaps, fresh graduates, photos, country conventions and common mistakes.

Data handling

• Harvested job data is rejected at insert time if the posting is older than seven days. Stale listings never reach the interface.
• CV drafts from guests live in the browser's local storage. When you're signed in, the builder also saves your CV to your account so it follows you across devices; you can delete it any time. The analyzer reads your file in the browser and does not upload it.
• Candidate profiles, once accounts exist, are opt-in and deletable. Data export and deletion are part of the product, not a support request.

What exists today

• Marketing site: home, employers, job seekers, CV hub
• Job directory with search, country and function filters, and pagination
• Job detail pages with an honest outbound-apply flow
• Employer landing page and a three-step post-a-job wizard
• CV Hub: builder, analyzer, and eight guides
• Help, pricing, about, contact, privacy, terms, glossary, tools and blog pages
• Full bilingual EN/AR with light and dark themes

What is deliberately not here yet

• Accounts and authentication
• A real database for employer posts and candidate profiles
• Payment processing
• AI generation of CVs and cover letters (the builder guides; it does not write for you yet)

Those are the next phases, in that order.