A recruitment platform where AI does the shortlisting, not the deciding

The problem

Recruitment runs on three parties who need different views of the same thing. The agency sourcing candidates, the company hiring them, and the candidate waiting to hear. Most tools serve one of them well and leave the other two chasing emails.

What we built

One platform with a portal for each party and an admin view over all of it.

Every user belongs to exactly one organisation, and agencies and hiring companies connect to each other explicitly. That connection decides what each side can see, and it is enforced in the database itself rather than by hiding buttons, so a mistake in the interface cannot leak one client’s candidates to another.

The pipeline

Roles go from creation through applications, interview rounds, assessment days, offers and onboarding, with each stage visible to the people who need it. Screening runs two ways: the hiring company sets the questions an agency must answer when it submits someone, and the agency sets the questions a candidate answers when they apply. They are kept as separate systems on purpose, because confusing them is how a client ends up reading the wrong answers.

AI matching that is quick enough to use

The first version of candidate suggestions sent the newest few dozen candidates to an AI model and hoped. We rebuilt it so the database scores the entire pool first, on skill overlap, job title overlap and distance from the experience band, and only the strongest 40 go to Claude for a closer read. The suggestions panel went from 10.9 seconds to 52 milliseconds to first appear, and it now considers everyone rather than whoever applied most recently.

Interviews, summarised carefully

Interviews run on Microsoft Teams, and their transcripts are summarised by Claude for the hiring manager. The summary deliberately makes no hire recommendation. That keeps a person making the decision, and keeps the product clear of the high risk category in the EU AI Act. It was a product decision, made early and on purpose.

Built to connect

A signed handoff takes candidates to the CMC Markets partner programme, and on the way back an email address arrives only as a hash. Agencies moving from Bullhorn can bring their candidates across. A scheduled job applies the data retention rules every night.

The result

A platform that runs the whole hiring cycle for agencies, hiring companies and candidates, with AI doing the sifting and people doing the deciding. The admin area even holds Verichain’s own sales pipeline, so the product is used to sell itself.