← Work & Resources
Internal AI

Internal AI assistants for property operations

Gemini assistants for staff questions, maintenance requests and owner reports.

AI workflow case view showing recommendation, human decision and next action
The related public workflow shows where AI stops and a person takes over.
Context

The work

I developed Gemini assistants using company source material and examples of staff requests. The assistants covered operational questions and recurring communication, including maintenance and owner reporting.

What mattered

A polished answer can still be wrong

A draft owner update had to stay a draft until somebody sent it. A work order needed an actual approval before the assistant could describe it as approved.

One concrete failure involved a past property visit: the workflow could not treat a historical report like a live visit and force a current checkout step. It needed to reconstruct what it could from existing work-order notes and leave missing facts missing.

Work

How I changed the assistants

The instructions defined the job and the source material. They also marked the points where a person had to decide. I tested bad answers and changed the instructions when the assistant handled an exception badly.

Result

The answer follows the recorded facts

The instructions told each assistant which company sources to use, when to ask a person and how to describe an action that was still pending.

KPS

A prepared request still needs approval

The KPS example shows this rule inside the job: a quote exceeds the spending limit, and the owner must approve it before booking.