The standing record · /.well-known/trust.json
Trusted by early teams hiring against the rubric
01
Depth
Most platforms stop after a single probe. Ishavi follows the rubric like a senior interviewer: it asks again, asks differently, asks for the example, asks for the metric. The transcript shows where the candidate got specific and where they hand-waved.
02
Fairness
No narrative-only ratings are allowed in the data model. Each score carries timestamped evidence quotes you can replay. If it is not in the transcript, it is not in the score -- the AI is a recommender, never the deciding voice.
03
Integrity
Knowledge probes, latency-loop detection, and paste signals run on every interview to surface inconsistencies. Heavier observation is opt-in per job and forcibly off in EU jurisdictions. We protect the integrity of the rubric without surveilling the honest majority.
04
Bill of Rights
Practice Mode, Re-Do, Human Escalation, Accessibility Profile, Scorecard Transparency. Five named rights, wired into the database -- not bolted on after a compliance audit. Reviewer rotation is enforced by CHECK constraints, not policy memos.
Hiring decisions should read like court rulings — anchored to evidence, open to challenge, signed by a human. Ishavi is built so that every interview can.
The editorial position
Author the role, the rubric, and any custom prompt layers. Tenant overrides ride on top of platform defaults. Every published version is immutable and audit-logged.
Magic-link email, region-aware, no password to remember. The candidate joins from any browser and speaks through a structured rubric with five-plus levels of follow-up.
Every score carries a timestamped evidence quote you can replay. The AI is a recommender, never the deciding voice -- a human reviewer signs the shortlist.
Bill of Rights for candidates
Practice Mode, Re-Do, Human Escalation, Accessibility Profile, Scorecard Transparency -- wired into the database, not bolted on.
Evidence-anchored scoring
No narrative-only ratings. Each rubric line carries a timestamped quote the reviewer can replay against the recording.
EU AI Act + NYC LL144 compliant
Limited-risk Art. 50 transparency obligations live. Bias-audit programme scheduled. Candidate-notice text published in every jurisdiction.
Reviewer rotation is enforced by a CHECK constraint. Appeal SLAs are timestamps, not promises. Every transition is audit-logged with a required reason field.
Practice Mode
Unlimited free dry runs against the exact AI persona for the role. Score never stored.
Re-Do
Within 24 hours of a tech-failed interview, candidates request a second chance. Recruiter SLA: three business days.
Human Escalation
Candidate says the word and a human takes over the live call. If no human is available in five minutes, the interview pauses and reschedules.
Accessibility Profile
Pre-interview self-disclosure (stutter, ADHD, ESL, vision, motor). Disables vision-based anti-cheat, extends latency budget, slows AI pacing.
Scorecard Transparency
Every candidate gets the plain-language summary, the evidence quotes, and a one-click path to request human review.
No. Ishavi produces evidence-anchored recommendations, never decisions. The IL AIVIA + GDPR Art. 22 framing is baked into the data model: a human reviewer signs every shortlist.
Candidates open an appeal from their dashboard. A different reviewer than the original decision-maker is assigned automatically; reviewer rotation is enforced in the database, not just in policy text.
Tenants are tagged with a region from day one. Year-1 of the platform deploys in Mumbai (ap-south-1) for Indian customers; EU and other regions land on the same residency-routing layer when demand calls for it.
Yes. A four-layer prompt composition engine -- system, industry, company, job -- lets you author overrides at any level. Every published version is immutable and audit-logged with a required reason field.
Region-aware plans, billed in your local currency. A free tier exists. See the pricing page for the current matrix.