Recruitment

Hiring at the speed of intent

A 60-person recruitment firm placing engineering and finance talent

The firm was losing placements to speed. Strong candidates went cold while recruiters worked through a backlog of CVs by hand. They stood up Roster, our talent management product, and it now reads every application, scores it against the live role, and books the first call automatically. They own the system and run it on their own infrastructure.

72%less time spent screening CVs
3.5xfaster first response to applicants
18 daysaverage time to shortlist, down from 41

Representative engagement. Numbers are directional outcomes typical of builds of this type, not figures from a single named client account.

The challenge

Every open role generated hundreds of applications across three job boards, a careers page, and inbound referrals. Recruiters were spending the first half of each day triaging CVs instead of talking to people. By the time a good candidate surfaced, the best ones had already accepted offers elsewhere. The CRM held the data but did none of the thinking, and no one trusted the keyword filters enough to rely on them.

Our approach

We started where the time was bleeding out: the screen. Rather than bolt another filter onto the existing ATS, Roster ingests every applicant, parses the CV into structured evidence, and reasons about fit against the actual requirements of the live role, not a list of keywords. Recruiters stay in the loop on every decision, but they review a ranked shortlist with reasoning attached instead of a raw pile.

What we built
  • Unified intake that pulls applicants from job boards, the careers page, and referral links into one pipeline
  • CV parsing into structured skills, seniority, and trajectory, with an evidence trail for every claim
  • Role-aware scoring that ranks candidates against the live brief and explains why, in plain language
  • Automated first-touch: qualified candidates get a personalized message and a self-serve booking link within minutes
  • Recruiter cockpit with one-click overrides, so humans always make the call and the model learns from it
The outcome

Recruiters now open their day to a ranked shortlist with reasoning, not a backlog. The first response to a strong applicant goes out in minutes instead of days, and the firm stopped losing candidates to faster competitors. Roster runs on the firm’s own infrastructure with unlimited seats, and they own the source outright.

We stopped competing on who had the biggest team and started competing on who answered first. That turned out to be the whole game.

Managing Director, recruitment firm
Stack
Next.jsPythonLLM scoring layerPostgresVector searchSelf-hosted on client AWS
Timeline

Core platform shipped in 6 weeks

Tell us what you're trying to run. We'll show you which product fits, or build the one that doesn't exist yet.