Gatekeeper
A zero-trust safety framework and WHS governance architecture for construction. Gatekeeper generates SWMS, risk assessments, and WHS control packs — and refuses to let anything reach a worker until it has passed through explicit trust gates.
The problem
Safety documentation in construction is high-stakes and high-volume. SWMS and risk assessments are often copy-pasted, generic, or rushed — which erodes trust and, worse, misses the hazards that actually matter on a specific job. The bottleneck isn't writing documents; it's producing documents a competent person can confidently sign off.
The approach
Gatekeeper treats document generation as a governed pipeline, not a single AI call. It combines deterministic inference (job classification, hazard-family selection, control logic) with AI-assisted assembly, then gates the output behind review before anything is issued.
- Job classified before hazard selection — retrofit, new build, demolition, civil, maintenance treated distinctly
- Deterministic hazard and control logic, not generic AI wording
- Issue-gate checks catch trust failures — blank responsible-person fields, placeholder text, unresolved emergency details
- Review-before-download: output stays draft-for-review until a competent person approves
- Regression discipline protects closed benchmark streams from silent drift
Why "zero-trust"
Nothing is trusted by default. Every generated document is treated as unverified until it clears the gates — the same posture a zero-trust network takes with every request. This keeps human judgement as the final decision point while automation removes the repetitive load.