Security

Your lab's data never leaves its own boundary.

Not a compliance claim — a description of how the system is actually built.

How it's actually built

Four things, verified in the codebase, not asserted in a copy deck.

Per-tenant data isolation

Each laboratory's documents are indexed into a separate vector store — QualiBOT can only retrieve from your lab's own uploaded records, never another tenant's.

PII redaction before indexing

Documents pass through a redaction step before they're indexed, including recognizers built for regional identifiers — Tanzanian phone numbers, NIDA national ID format, and TIN numbers — not just generic patterns.

Self-hosted inference

QualiBOT's language model runs on infrastructure we control, not a third-party AI API call for every question — consistent with keeping lab data inside a system you can audit.

Answers are gated, not guessed

Retrieval is scored against a confidence threshold before any answer is generated — below it, QualiBOT says it doesn't have the information, rather than filling the gap.

In development

What's not here yet

No compliance certification (SOC 2, ISO 27001, or similar) is claimed — none has been pursued yet. This page describes real architecture decisions, not an audited compliance posture. If you need a formal security review for an evaluation, talk to the team directly rather than relying on this page alone.