For the officer who wants to try this
You have used AI and want to build something useful
Start here. One job, written in one sentence, and a clear line about what the companion is never allowed to decide.
Build an AI companion · session 1 of 4
Most companions fail because they were asked to do everything. This session is about choosing one job - and the line the companion never crosses.
Six decisions before building: writing the one job in a sentence, refusing to give it everything, keeping access decisions with people, feeding it information you already hold, checking an answer against a source. The details that never go in.
Corporate pricing on request. Course funding application status: pending.
The scenario
An officer builds an AI companion the way anybody builds one: a free tool, an hour of curiosity. A feeling that the answers look clever. The first week is fun. The trouble arrives later, when the companion is answering with confidence about things it was never told.
The fix is unglamorous. Write down the one job. Give it the information you already hold. Decide the three questions it must always hand back to a person.
Do that and you have a companion you can trust with a task, because you know exactly where it stops.
The six decisions are the ones that actually go wrong on a quiet shift. They are, in the order the run puts them:
What changes, and for whom
For the officer who wants to try this
Start here. One job, written in one sentence, and a clear line about what the companion is never allowed to decide.
For the supervisor
This session sets the boundary before the enthusiasm: no decisions about people, no personal details, no site layout. The rest is genuinely useful.
For the agency
Better a short, plain line that teaches the boundary: one job, real information, no personal data. That beats an officer learning it from a mistake.
These are descriptions of the work the training addresses. They are not client results and not a promised saving.
What the officer does, in order
About ten minutes at the post, on the officer's own phone, with a supervisor on site.
One printed marker at the guard house desk. The officer builds on their own phone during quiet time.
Six decisions that narrow the companion down: what it is for, what it is never for, and what it needs to know.
The overlay shows the three questions a companion must always send back to a person: access, identity and anything confidential.
One row, the same five fields as every run in the department.
Every session writes the same record, so four sessions make one page of that officer's training history, and the same five fields are asked of the newest officer and the most senior one - which is the part of it that is fair. The run is written to the agency's list in the five fields the checklist reads.
This page is the session. The run itself is on the player: open it here - the practice run needs no camera. It is session 1 of 4 in Build an AI companion.
This is post training, so there is no assessor to satisfy - but the record matters anyway: it is how an agency shows what its officers were trained on, in what order, and when.
| Field | What it carries |
|---|---|
| Name | As written in the agency's own staff record. Nothing else is stored about the officer. |
| NRIC / FIN | The last four characters only. The full number is never entered and never leaves the phone. |
| Training provider | Whichever name the checklist route puts here: the agency's own name when the agency delivers the training in-house, or the service provider's name when a provider delivers it - in that case the checklist also asks the agency to submit the provider's invoice showing simulation training was used. The field is left blank on purpose: it is typed, not assumed. |
| Course title | States the technology: the training programme or session plus Augmented Reality (AR) Simulation Training. |
| Date of training | The date the run was completed, written from the device at the time of the run. |
The record lives on the officer's own phone until it is exported or printed, so the habit is: run it, then save the list. The house standard is 80 per cent, which is a fair bar for six decisions, and a failed run is a re-run rather than a mark against an officer.
Every course here is honest about what it is and what it is not, because a package that oversells itself is the one that lets an agency down in front of an assessor. Read these before you book anything.
Questions about this session
Build an AI companion · session 1 of 4
Fix one marker at the real point, open the session on a phone, and see whether the overlay holds where your officers actually stand. That is a fair test, and it takes three minutes.
Corporate pricing on request. Course funding application status: pending.