The way an investigation actually flows
Five steps, one owned case, and a human in the loop the whole way. These are the same steps the application shows its own users.
-
Choose the person and open a case
Choose the person in the Internal Directory and start an investigation.
Sign in with your institution’s Microsoft account, then search the directory by name, work email or employee number. Results are ranked candidates. You open the exact profile yourself.
Starting an investigation creates a private case owned by you and anchored to that one directory identity. Asking again returns the same open case rather than a duplicate.
Step 1: the directory profile, where an investigation starts. Illustrative demonstration using fictional data. -
Collect evidence
Collect evidence from any source, in any order.
From the case’s Sources section, run the collections the investigation needs: a selected Instagram candidate, one of the subject’s directory phone numbers, a bounded public-web search, the institutional index, or a research run.
Each collection is an explicit action and runs in the background. Results arrive with their source and capture time, and each Instagram run is kept as its own dated snapshot.
-
The human gate
Review every item
Review every collected item: approve, reject, or leave pending.
This is the step that controls everything after it. Analysis reads only the items you approve, and a page recorded as concerning a different person never qualifies, even if it was attached.
Decisions apply once per logical item, so the same content seen in a later collection keeps your earlier decision. Every change is appended to an audit log.
Step 3: Evidence Review, the human gate. Illustrative demonstration using fictional data. -
Analyse and ask
Analyse the approved evidence and ask the AI Assistant.
Analysis needs your explicit confirmation and at least one approved item. It produces evidence-scoped findings, then a cited synthesis, then a person analysis where every claim is graded and cites the evidence behind it.
The AI Assistant answers questions over the same case data, shows its sources, and adds caveats when an answer leans on unreviewed or rejected material.
-
Export a report
Export a report built from a fixed evidence snapshot.
Generate the AI Personal Intelligence Report or the Comprehensive Evidence Report, in Italian, English or Arabic. Each is rendered once from a fixed snapshot and stored.
When the work is done, the owner can archive the case or delete it entirely. Deletion is limited to the owner and recorded.
Step 5: a report rendered once from a fixed snapshot. Illustrative demonstration using fictional data.
A match is not an identification
VERIDEX never claims that a collected profile, a phone response or a web page belongs to your subject. It shows you the evidence and how strong the signal is, and leaves the judgment to you.
For public-web and institutional pages, the rule that decides whether a page is even worth showing you is deterministic and visible. Weak signals can support a candidate, but they can never create one.
Can qualify a page as a candidate
- The person’s institutional email address or alias
- Their employee number
- Their full name, as one continuous phrase
- Their exact phone number, together with a name signal
Only ever corroborates
- Given name on its own
- Surname on its own
- Department
Even a qualifying page is only a candidate. It never attaches to the case by itself, and it never proves who someone is.
See how AI fits into this
Where the model helps, and the limits the platform enforces around it.