We Built an AI Recruiter in Two Days
We Built an AI Recruiter in Two Days
The Problem
Recruiting is broken. Recruiters drown in CVs, candidates write the same cover letter 50 times, and the first interview is usually theater — both sides performing instead of connecting.
The real question in any hire is never "do they have the right keywords on their CV?" It's "how does this person actually think?"
We decided to build something that answers that question without wasting everyone's time.
What We Built
Vibe Check is a two-sided AI recruitment platform. Candidates upload their CV, have a structured 8-stage voice or text conversation with an AI interviewer, and get a report. Recruiters get every candidate ranked by score with green flags, red flags, and a full transcript without sitting through a single screening call.
The AI doesn't ask generic questions. It reads the CV before the interview starts, references specific projects and roles by name, and pushes candidates to go deeper. It moves through structured stages. Technical ability, teamwork, obsessions, a curveball and ends the interview itself.
No cover letters. No bias from a nervous first impression. Just conversation.
How It Works
The candidate flow is simple:
Open a link (unique per job posting)
Upload your CV
The AI reads it before asking anything
8-stage structured interview. voice or text
Report generated automatically
The recruiter sees all candidates ranked. Score, summary, green flags, red flags, full transcript. The best candidates rise to the top without anyone manually reading 200 applications.
The Stack
Built in two days by two people working in parallel — one on backend, one on frontend.
Next.js — framework
Groq API (Llama 3.3) — the AI brain, fast and free
Supabase — database, stores every conversation and report
Web Speech API — browser-native voice, costs nothing
unpdf — CV parsing
Tailwind CSS — styling
Vercel — deployment
Total cost to run: essentially zero at this stage.
Where This Goes
The immediate next step is a recruiter dashboard a single view where hiring managers see all candidates ranked, compare profiles side by side, and add notes as candidates move through interview stages.
After that, the platform becomes two-sided in a deeper way. Instead of recruiters filling in job descriptions, the AI interviews them first "What has made past hires fail? What does success look like in 90 days?" and builds a criteria profile from that conversation. Better input, better output.
The longer vision is an end-to-end hiring intelligence layer. Multi-stage pipeline tracking, interviewer briefings before each stage, collective decision synthesis when multiple people have interviewed the same candidate, and organizational memory so a strong candidate from last year's hiring round isn't lost forever in a spreadsheet.
The core insight that makes this different: conversation reveals more than a CV ever will. Forms and profiles show what people want you to see. A 10-minute conversation, led by an AI that has read their CV and knows what to probe, shows how they actually think.
Current Pilots
We are working with two early clients ***, a program for unconventional builders, and VES, (Vaasa Entrepreneurship Society) to validate the approach with real candidates in real hiring cycles.
If it works there, the same engine applies to any company drowning in applications and tired of the same broken process.
Built with Next.js, Groq, Supabase, and a lot of coffee.
