Built to withstand
procurement, compliance, and legal review
Kempian is designed as if every page may be reviewed by a sophisticated buyer, regulator, and counsel on the same day.
Trust pillars
Explainable AI by design
Every AI output includes signal names, factor groups, confidence bands, missing signals, and reasoning access. No score ships without explanation.
See AI Confidence →Human-in-the-loop at every gate
Four mandatory human review gates govern the lifecycle from demand intake to candidate submission. No gate can be programmatically bypassed.
Multi-tenant data isolation
Every DynamoDB operation is scoped to the authenticated tenant via JWT. Cross-tenant data access is architecturally impossible. Not just a policy.
EU AI Act Art.14 alignment posture
Kempian is designed to support human oversight requirements for high-risk AI systems as defined under Annex III of the EU AI Act.
Immutable audit trail
Every decision is recorded in an immutable compliance log per GDPR Art.22. Recruiter ID, timestamp, AI score seen, decision, and override reasoning.
AWS-hosted, single-region
All data resides in AWS us-east-1. No cross-tenant model training. No customer data used for AI improvement without explicit agreement.
AI decision boundaries
Kempian AI is decision support. Not an autonomous decision-maker. Every action that affects a candidate or client requires an explicit human confirmation step.
See AI ConfidenceGovernance FAQ
Does Kempian train its AI on my candidate data?
No. Customer data is not used to train or improve the shared AI model. AI scoring uses the Claude API with prompts constructed from your workflow data. Your data is not retained by the model.
What data does Kempian store and where?
Candidate profiles, requisitions, outreach records, and compliance logs are stored in AWS DynamoDB in us-east-1. Talent profile files in S3. All storage is scoped to your tenant.
How does Kempian handle GDPR data subject rights?
Kempian provides controls for data access, deletion, and export at the tenant level. Candidate opt-out records are maintained and enforced before any outreach step. Contact your account team for a data processing agreement.
Can a recruiter override an AI recommendation?
Yes, always. Override capability is a core requirement. Every human review gate includes a reject or hold option with a mandatory reason field for the audit trail.
How is AI Confidence calculated?
Confidence is calculated by Claude AI (Anthropic Bedrock) using named factor groups: skills, experience, compliance, location, and role-specific signals. The rubric is deterministic and documented per role type. It does not infer demographic data.
Need a full governance review?
We provide security posture documentation, data processing agreements, and architecture review calls for enterprise procurement.