An inert scaffold sat untouched for seven weeks.
The real decision wasn't a line of code — it was choosing to actually start: rebuild it from scratch as a full architecture blueprint, not patch the scaffold. That's where KIS began.
KIS is the real-world pilot for the security track of my own Intelligence Engineering Methodology (IEM-CS) — applying 20 years of program and PMO leadership to a question I kept seeing from the delivery side but never got to solve hands-on: how do you reduce decision-risk when your data is incomplete, disconnected, or low-confidence?
For physical + cyber security, that means one converged system instead of a dozen disconnected vendor dashboards.
In the first four days of real design: a 28-table schema, a cross-vendor telemetry matrix, and a correlation-layer concept for stitching signals across products.
Small detail that mattered more than it looked at the time — customer-name sanitization went into every document on day one, not as an afterthought once real customer data showed up later.
One design rule got locked immediately and never got renegotiated since: the AI narrates, it never decides. Every fact in the system has to come from a deterministic engine first; the LLM's only job is to phrase it into readable language. That one sentence has shaped every AI-touching feature built since.
#IEMCS #AISecurity #SecurityArchitecture #ResponsibleAI
Originally published on LinkedIn.