AI Hiring Copilot (Clarifit)
Two startups inside an IT holding, around 80 people, hiring all the time with a small recruiting team. I was their People Partner, and the internal AI Challenge was a good excuse to fix something that annoyed everyone.
Problem
Two recruiters reading the same CV came to different conclusions. Interview plans were copied from the last vacancy. After the call someone had to turn the whole conversation into a comment in PeopleForce and, for most candidates, a polite rejection. Per candidate that was about 35 minutes of routine work, and the quality depended on who did it and how tired they were.
Solution
Built Clarifit, a recruiting copilot that reads the CV against the job, drafts the interview plan, helps during the call and turns the transcript into an ATS comment and a rejection note.
- saved per candidate (estimate)
- ~23 min
- saved per candidate (estimate)
- modes across the hiring loop
- 4
- modes across the hiring loop
- internal AI Challenge
- 3rd place
- internal AI Challenge
How it works
Resume
Paste the JD, the CV, the hiring manager's brief and what worries you ("no GraphQL", "English below the bar"). Pick strictness. Out: hard and soft skills against the JD, risks, and what to dig into at the interview.
Plan
Same inputs plus seniority and up to 4 company values. Out: interview sections (hard skills, soft skills, values) with questions and timing.
Interview
During the call you log answers; it rates each one 1–4, flags what's missing and suggests a follow-up and the next question from the plan.
Analysis
Paste the transcript. Out: a verdict, a comment ready to paste into PeopleForce, and a personalised rejection note. Nothing is stored on a server: candidate data stays in the recruiter's browser.
Outcome
- About 23 minutes saved per candidate (35 → 12 min) by our estimate: CV review 5 → 2, interview plan 15 → 5, write-up 15 → 5.
- 3rd place in the internal AI Challenge.
- Pilot with the recruiting lead: positive feedback and a clear list of what to fix next: pull vacancies and hiring manager comments from PeopleForce, make strictness levels really differ, link the interview plan to the CV analysis, and give the tool a memory of the candidate.
What I did
- Built the app myself: Next.js, AI SDK with structured output, gpt-5-mini, built in Cursor.
- Designed four modes that follow the hiring loop: Resume, Plan, Interview, Analysis.
- Wrote the prompts and output schemas, including strictness levels and company values.
- Ran a pilot with the recruiting lead, wrote a guide, and pitched it at the internal AI Challenge.
Screens
My role
Product, prompts and code, by myself.
Stack
Next.js · Vercel AI SDK · OpenAI (gpt-5-mini) and Google models · structured output · Railway · built with Cursor.
Next case
Employer Brand Studio