Trust & Compliance

AI in hiring should show its receipts.

Hiring is one of the highest-stakes things software can touch. These commitments are structural — built into how the product works, not promises bolted on after.

A human makes every decision.

AI ranks, evidences and drafts. It never advances or rejects anyone — on any plan, in any mode. In the code, the decision field has exactly one writer, and it is only reachable from a person clicking a button.

Recommendations are off by default.

Turn them on and they're advisory, clearly labelled, and logged — including who turned them on. They're available from the Expert plan up, and the toggle itself writes an audit event.

Every score links to evidence.

One tap from any number to the moment, message or piece of work that produced it. A score cannot be saved without it: it's enforced in the application, by a NOT NULL column, and by a database constraint that also rejects a blank.

Integrity signals are questions, never verdicts.

Paste, tab-switch and cadence detection prompt a conversation. They never label anyone a cheat — and disclosed AI use is logged neutrally, because using tools well can be the skill you're hiring for.

Candidates always know where they stand.

Sims are disclosed as AI, async sessions stay short, and a human reviews everything they produce.

Built for a regulated space

Designed around the rules, not despite them.

NYC LOCAL LAW 144

New York regulates automated employment decision tools, with bias-audit and candidate-notice requirements when tools substantially assist decisions. Our human-decides design keeps AI advisory — and session exports give your counsel the decision logs and notice trail they’ll ask for.

EU AI ACT

European law treats hiring AI as high-risk, with obligations around human oversight, transparency and logging now phasing in. Here, oversight is structural — nothing auto-decides — every session produces an evidence trail, and AI involvement is disclosed to candidates by default.

We design to support your obligations — we can’t discharge them for you, and this page isn’t legal advice.

Kandevo AI runs on Anthropic’s Claude models — named here, in your procurement pack, and to every candidate. No mystery AI.

AUDIT TRAIL · EVERY SESSION EXPORTS ONE

14:02Score 4/5 · Judgment — linked: "Named the client's exec exposure before the fix" (chat)

14:07AI recommendation enabled by T. Chen · advisory

14:12Decision: Advance — made by T. Chen (human)

SCORE → EVIDENCE → HUMAN DECISION, EVERY TIME

The trail is append-only at the database level: updates and deletes are rejected, not just discouraged.

Cheat-resistant by construction

Generated, not library.

Question banks leak onto forums within weeks. Our content doesn’t exist until you paste the JD — scenarios, sims and mandates are generated per role, per session. There’s nothing to memorise, nothing to trade, and regenerating mid-interview takes one tap if anything feels rehearsed.

The candidate’s side of the deal

TIME

Async sessions stay short.

Respectful work samples — not an unpaid project.

DISCLOSURE

AI is always disclosed.

Candidates know when they're talking to a sim and what's being observed.

REVIEW

A human reviews everything.

Nothing they produce is judged by a machine alone — ever.

What we haven’t done yet

We are not SOC 2 certified, and we have no psychometric validation study — we can show you evidence behind every score, but not yet a correlation between those scores and who succeeded in the role. Bias-audit support, a DPA and a security review are part of the Enterprise compliance pack and are scheduled against real Enterprise conversations rather than shipped in advance.

We’d rather you read that here than discover it in procurement.

See the evidence trail for yourself.

Start free See pricing