Vouch

Vouch is a two sided hiring platform for Kuwait and the GCC. Employers pay a monthly subscription. Job seekers pay per service for interview practice, tailored applications and a capped apply-for-me option, and never pay to search or apply.

It was designed end to end from a blank page: market research, product definition, the business model, the name, the brand, then every screen across four surfaces.

The prototypes

Three running prototypes. Each one starts on its own flow, and the sidebars and tab bars navigate, so you can walk the product rather than look at pictures of it.

Website, desktop

Website, mobile

Dashboards

What is in the file

Twelve pages in total, including a sitemap of 57 screens, five end to end flows, a states and edge cases page, five email templates in English and Arabic, and a handover specs page.

The research came before the design, and it changed the brief twice.

The market broke recently

AI made applying free. 78% of job seekers now use AI in their applications and 56% say they apply to more jobs because of it. 77% of hiring teams regularly see AI written applications, up from 53% in early 2024. Employers get buried, screen harder with their own AI, and seekers respond by applying even more.

At the same time nobody can tell what is real. Between 18% and 22% of listings are ghost jobs, roles nobody intends to fill, and 93% of HR professionals say their employer posts them. The average dead end application cycle costs a seeker about nine hours. Going the other way, Gartner expects one in four candidate profiles worldwide to be fake by 2028.

Every existing platform treats one side of this and ignores the other.

Two things in the original brief were wrong

The build order. A two sided marketplace needs jobs to attract seekers and seekers to attract employers, and Indeed, LinkedIn, Bayt and Seek already own that liquidity everywhere this could launch. The answer was to design a seeker product that works against jobs sourced from elsewhere, and open the employer side alongside it rather than after it.

The business model. As briefed, the platform would sell seekers volume applications and sell employers filtering. Those are two sides of the same war. The moment the apply-for-me product works at volume, the employer customers are the ones complaining. LazyApply, the volume leader in that category, sits at 2.4 out of 5 across 105 reviews with about a quarter citing refund failures.

So the seeker product sells fit and preparation instead of volume, and the apply-for-me service is capped, with every application approved by the seeker before it sends.

The position nobody had taken

Ghost jobs are 18 to 22% of listings and no platform labels them. Verified listings plus a published employer response rate is a position none of the incumbents can copy without admitting they were the problem. That became the promise the entire design is built to keep.

The promise, made structural

Every job is real, and every employer answers. Three mechanics carry it, and all three are measurable:

Every employer is verified against a trade licence or commercial register before a listing goes live. Every listing carries a real vacancy attestation with a named person on it. Every listing shows a salary range, and "competitive" is rejected at the form rather than warned about.

A clock starts when a seeker applies and stops on the employer's first real reply. That rate is published on every listing the employer runs, and no setting, support override or premium tier can hide it.

The design system

The signature is one shape called the frank: a corner cut at 45 degrees, the way a processed document is notched. It appears on the mark, on vouched cards, on the lead price tier and on the dark bands, and nowhere else. In Arabic it mirrors to the opposite corner.

Archival ink on a green tinted paper with an oxide stamp red. No blue and no purple anywhere, because that is the default palette of every AI built product. Every colour pairing was computed rather than eyeballed: ink on paper is 14.71:1 and the weakest pairing in the system is 5.13:1.

Type is Readex Pro, which covers Latin and Arabic in one family drawn for both scripts. The wordmark is drawn, not set.

Four variable collections

Nothing in the file hard codes a hex or a size.

The screens that carry the argument

Post a job turns the salary range and the attestation into gates rather than suggestions, with a checklist showing which are unmet.

Candidate detail puts the AI rank next to its reason, discloses in plain language that an automated system produced it, and offers Shortlist, Override and Reject with a reason. No code path rejects anyone automatically.

The AI decision log records every automated decision with its inputs, output, model version and the named human who reviewed or overrode it. Ranking candidates is a high risk system under the EU AI Act and an automated employment decision tool under NYC Local Law 144, so this is a requirement rather than a feature.

Mock interview feedback is what three credits actually buys: a score, three sub metrics, one specific thing to fix, and a question by question breakdown.

The apply-for-me queue shows every drafted application for approval before it sends, and ends with the sentence "we never promise interviews".

The trust page includes a section called "what we do not check", which admits that listings sourced from partner feeds are unverified employers.

Handover

The file ends with a specs page: breakpoints, tokens, RTL rules, accessibility, and seven non negotiables written as build constraints. Every frame is auto layout. Recommended stack in the PRD is Laravel with Filament, Postgres, Meilisearch and Redis, chosen because four items on the build list come nearly free from Filament and because job search needs Arabic tokenisation.