The interview that interviews back.
How Punch built Zwilt a conversational AI interviewer that talks to candidates, listens, and hands hiring teams a scored, evidence-backed read — at the scale of every applicant, not just the lucky few.
"…so rather than lock the whole table I'd version the row and reconcile on write — it costs a little memory but it keeps the hot path read-only."
Every signal is traceable to the moment in the interview that produced it. Hiring managers review the quote, not just the number — and can override with a reason that feeds back into calibration.
One rubric, every candidate.
Structured interviews mean everyone is measured on the same signals — the consistency a tired human panel can't promise.
No score without a quote.
Scores that can't point to what the candidate actually said don't ship. Auditable by design.
The model recommends. People decide.
Every recommendation is an input to a human decision — with overrides captured for calibration, never silently discarded.
A thousand applicants. Time to properly talk to twelve.
Screening is a bottleneck: recruiters skim résumés and most candidates never get a real conversation. Zwilt wanted to give every applicant a genuine first-round interview — and give hiring teams a consistent, comparable read on all of them.
Punch built the engine: a voice agent that runs a structured interview, transcribes and understands it live, then scores each answer against a role rubric — with the transcript quote behind every score.
Case study PDF ↓IN THE NEWS
ZWILT.COM · OFFICIAL
Zwilt — the platform, live. ↗
LINKEDIN · OFFICIAL
Zwilt on LinkedIn. ↗
