Everyone has an AI interviewer now. Kavron is the one your legal team, your candidates, and your hiring managers all trust — deep anti-cheat, provable consent, and the full hire-to-pay loop in a single platform.
The AI-interview category's biggest liability is how it handles consent, biometrics, and candidate data. We turned that liability into the product's strongest moat.
Consent
Four un-bundled consents (recording, biometric proctoring, data processing) gate every interview — and using interview data to train models is a separate toggle that's OFF by default. Every grant is an append-only, version-stamped audit record.
Most AI interviewers bundle consent — or train on candidate interviews — with no clear, separable opt-in. That's the exact ground competitors are getting sued on.
Retention
Face and eye analysis runs entirely in the candidate's browser: facial landmarks and video frames are never transmitted or stored. Only behavioral flags and a derived integrity verdict are kept — and a daily job purges those plus the transcript after 30 days (the configured window), leaving only the numeric decision record. Paired with GDPR export and password-reauthenticated erasure.
Indefinite retention of voice and biometric data is the norm — and a growing BIPA / GDPR liability your legal team inherits.
Privacy
Candidates never see scores, recommendations, or per-question grades — enforced at the database query, the API response shape, and the UI. Three independent layers, so a future refactor can't open a hole.
Accidental score exposure through an over-broad API include is a classic, recurring leak in screening tools.
More vetting modalities, more domains, and a verdict a hiring manager can trust — and contest.
Anti-cheat
Vision proctoring, a linguistic AI-assist detector (delay, zero disfluencies, essay cadence, uniform answers), 2× spot-error integrity traps, and a CV-vs-interview gap detector — four orthogonal signals fused into one verdict.
Single-modality webcam proctoring is defeated outright by “ChatGPT open behind the camera, read aloud.” Our linguistic detector targets exactly that blind spot.
Coverage
Question depth is calibrated to each candidate's real background (no fixed tier), with domain-universal traps for medicine, finance, law, software and more. Language-specific roles run entirely in-language with a native-fluency + RLHF-viability score.
Most AI screeners are tuned for software engineers. Native-fluency grading for RLHF / data-labeling hiring is a market incumbents don't serve.
Fairness
Borderline answers can get a small, capped benefit-of-the-doubt for nerves/transcription — never on integrity traps — and the exact points and reason are surfaced to recruiters. The model can't hide an adjustment.
AI scores are usually a black box, which fuels the “automated hiring is unaccountable” complaint. Ours is auditable and contestable.
The whole hire-to-pay loop, and the reliability and economics to run it at scale.
Marketplace
Stripe Connect escrow contracts with milestone funding and release (idempotent, with refund/dispute/clawback handling) — plus a compounding 20%-per-milestone referral reward that grows your candidate supply.
Most marketplaces stop at matching and hand payment to external tools, with at best a flat one-time referral bonus.
Reliability
Reasoning routes to Claude, bulk extraction to a cheaper model, with retries and a fully local heuristic fallback — so interviews and grading keep working through a provider outage, and a per-model cost ledger shows real margin.
Single-vendor LLM lock-in means an outage or price hike takes the product down with it.
LLM-parsed résumé + GitHub/LinkedIn enrichment build one profile, reused across every role.
A ~25-min role-calibrated voice interview — proctored, integrity-trapped, graded for substance and fluency.
Optional written / MCQ / coding / DSA stage, graded automatically — no scheduling, no panel.
Recruiter reviews one scorecard; employer funds escrow milestones and releases payment on delivery.
Spin up a role, run the AI interview, and watch the scorecard, proctoring verdict, and integrity check land in the recruiter panel.