From résumé to hire

Seven steps. Two hours. One audit trail.

The agent parses the résumé, scores it against the JD, verifies claims, runs the live interview, scores it, ships a coding challenge to qualifying candidates, and routes to the recruiter — every step model-versioned, prompt-versioned, replayable seven years later.

See pricing

01Résumé parsedimap · 12 KB · 0.42st+0:00
02Matched against JDSonnet · score 82 · 0.6st+0:01
03Delivery verifiedGitHub + web · 5 claims · 1mt+0:03
04Live interviewvoice · 38 turns · 45mt+0:48
05Scorecard4.2 / 5 · 5 criteriat+0:50
06Coding challengesandbox · 60m · GitHub repot+1:50
07Hiring decisionrecruiter · pass · readyt+2:00

Model · Claude Opus 4.7Framework · NYC LL144 · EU AI Act · GDPR · PDPATotal · 2:00

Trusted by recruiting teams shipping audit-grade hiring

01 · Provenance

Each step: model, prompt, artefact.

Every stage emits a JSON artefact under data/{job}/{candidate}/. parse uses imap + pdf extraction. match runs Claude Sonnet against jd.txt and emits analysis.json with score + recommendation + per-criterion profile. verify is an Agent SDK session that cross-checks GitHub, web, and authorship sources. interview runs 45 minutes (Sonnet for Doubao, autonomous voice for OpenAI/Gemini). scorecard is a Post-Interview Agent SDK pass over the transcript. coding is a Sonnet topic generator + Agent SDK evaluator. decision is the recruiter via Lark bot, dashboard, or API. Every artefact is versioned with model + prompt id.

02 · Replay

Replay any step seven years on.

Filesystem (data/) and PostgreSQL (interviews.state JSONB, coding_challenges, recruitment_events) form a parallel record. Default retention is seven years. The audit-trail surface re-renders any decision with the model input + state at that moment — frozen prompt + frozen rubric guarantees deterministic replay. A regulator can replay your hire from 2031 the way you saw it in 2026.

03 · Gating

Coding challenge gates on score ≥ 45.

Only candidates whose interview score clears CHALLENGE_MIN_SCORE (default 45) get the challenge dispatched. Below 50 → automatic rejection. 50–69 → recruiter judgment via human-review queue. ≥70 → recommended hire. The Post-Interview Agent makes the gate call from the transcript evaluation; recruiters can override at any threshold and the override is logged with the same provenance shape as the original decision.

End-to-end

Two hours, résumé to scorecard.

Seven steps, one audit trail. The fastest path from inbox spike to interview-ready candidate — and the slowest path a regulator can still replay seven years on.

2hr

Two hours from résumé to scorecard.

Start free — 10 interviews