Evidence-driven identity intelligence

Anchor a case to a person you choose in your directory, gather evidence from many sources with its origin attached, review it by hand, and run cited AI analysis. A match never quietly becomes an identification.

For authorized investigators in institutions that run Microsoft Entra ID.

Evidence Review: every item shows its provenance, source, capture time and review state. Illustrative demonstration using fictional data.

Identity, evidence and analysis, held in one case

Identity

Start from a person you explicitly choose in your organization’s directory. Nothing is matched automatically, and a directory match is a possible match, not an identification.

Identity and directory

Evidence

Collect from social, phone and messaging, public-web and institutional sources. Every item is stored with where it came from and when, and nothing is collected until you ask.

Evidence collection

Analysis

Turn approved evidence into cited findings, a synthesis and a graded person analysis, then question the case with an assistant that shows its sources.

Analysis

Provenance you can see

Every block of evidence declares what kind of thing it is, with a coloured rail and a written label. Colour in VERIDEX is reserved for this, so a reader always knows whether they are looking at a recorded fact, a local calculation, something observed, or a model’s reading.

Source fact

A value recorded as a source system holds it, such as an attribute of the directory record.

Example: Department, as listed in the directory

Derived locally

Computed by VERIDEX from source facts. A reading of the record, never checked against an official register.

Example: Birth date decoded from a Codice Fiscale

Observation

Something present in material collected from a source at a known time, kept as a snapshot.

Example: A public post captured during a collection run

AI interpretation

A model’s reading of approved evidence, citing the exact items it relies on and graded for support.

Example: A claim citing two evidence items, partially supported

Context

Information that frames the evidence without being evidence: coverage, limitations, review status.

Example: No collection has been run for this source

A human decides what the analysis may read

Investigations follow the same five steps. Review sits in the middle: analysis reads only the items you have approved, and every decision is written to an audit log.

  1. Choose

    Choose the person in the Internal Directory and start an investigation.

  2. Collect

    Collect evidence from any source, in any order.

  3. ReviewHuman gate

    Review every collected item: approve, reject, or leave pending.

  4. Analyse

    Analyse the approved evidence and ask the AI Assistant.

  5. Report

    Export a report built from a fixed evidence snapshot.

How an investigation works, step by step

Off until you decide otherwise

A fresh deployment reaches no AI service and sends no email. Each external capability is something your institution switches on deliberately.

What is on and off out of the box

  • AI runs on a built-in demo provider Live model output starts only when you connect a provider of your own.
  • Email sending and enrichment allow nobody Both are limited to operators your deployment names explicitly. The default list is empty.
  • Breach data and public-web search stay off They need explicit configuration before they can be used at all.

Built for authorized institutional investigations

See how VERIDEX keeps evidence, review and AI analysis in one accountable place.