GeoSurv runs well. The systems work. So this isn't a tool to fix what isn't broken; it's a system that addresses the part Josh said is the real bottleneck — people. Five-year locators leaving in four-week windows. Knowledge walking out the door. Vision that's clear at the top but uneven in the field.
Five things we committed to
Outcomes, not tooling. Every feature here answers the question "without this, how long would X take? With this, how long does X take?" If we can't answer it, the feature isn't here.
AI throughout, with guard rails. 15 headless agents, 8 conversational. All in Sydney. All cost-capped. All produce drafts you sign off — never silent state changes.
Australian-resident data path. Postgres in ap-southeast-2. Bedrock inference in Sydney via the AU-Geo cross-region inference profile. No Opus calls (US-only). Sub-processor register on the public footer; updated quarterly.
Domain accuracy. Real entities. Real standards. AS 5488.1:2022 quality levels in their proper places. NATA Site 25927. TfNSW TAO 224. RIW. BYDA. We don't hand-wave compliance.
UI as if you have a job to do. Cmd-K opens anywhere. Every page loads in <1.5 s. Empty states teach. Errors are actionable. Tables show the work; modals don't get in the way.
What this is not
Not a chatbot. Most of the AI runs without you asking — flight risk surfaces before notice, capture sessions schedule themselves, calibration books before hard-stops.
Not generic SaaS. Open Simpro / WorkflowMax / Procore in another tab. Compare the language. Compare what's on the dashboard. Compare what's missing.
Not an "iteration" of current tools. The end-of-shift wrap is 30 seconds. The quote-from-RFP is 90 seconds. The knowledge transfer board exists at all. These don't show up in a slightly nicer Excel.
How we know it works
Every AI feature logs the outcome. The Outcomes Register lists each feature, the baseline (without it), the target (with it), and the metric. We measure ourselves against it. Outcomes register →
And we'll tell you when something doesn't deliver. The sub-processor register, the audit log on every row, the kill-switches per AI feature — these aren't decoration. They're the trust contract.
In short
If we get this right, your best people stay longer, ramp faster, and walk out with their knowledge captured — not their skills.