← All use cases / Document intelligence
Synthetic data only
Fire-safety engineering · Melbourne
From a strategy session to a defensible compliance register
A Melbourne fire-safety engineering consultancy of around 20 engineers designs performance solutions that have to survive audits years later. The decisions are made out loud, in strategy sessions; turning them into registers, reports and tracked-change documents is where whole days disappear. We built a working prototype that does the assembly, while every decision stays with an engineer.
Explore the actual prototype
- Everything is synthetic: a fabricated project (“Foundry Lane”), invented names, no client data.
- No live AI calls: outputs are pre-computed, so there is nothing to configure and nothing to break.
- Nothing to sign up for: click around exactly as the client did in the demo call.
Opens in a new tab · synthetic data only
The problem
- Performance solutions are decided in meetings, then reconstructed by hand from notes and memory.
- The register lives in Word and SharePoint; assembling it is copy-paste work measured in days.
- Every item needs a defensible audit trail: who proposed it, who reviewed it, what changed.
- Precedents from past projects sit in senior engineers’ heads, not in a searchable library.
- Reports go out as tracked-changes Word documents. Any tool that cannot produce those is dead on arrival.
of the register workflow covered by the Stage-1 prototype, from session audio to a generated tracked-changes document.
decisions made by the AI. Every inferred item is labelled “AI inferred, pending human review” until an engineer signs it off.
What the prototype does
Strategy session in, structured register out
Engineers run the discussion exactly as they do today: upload the audio or record in the room. The AI proposes register items from what was said.
AI drafts
PM review with an audit record
A project manager accepts, edits or rejects each item. The trail (source quote, reviewer, timestamp) is kept for every decision.
Human decides
A live register, per role
Directors, PMs and engineers see the same register through different lenses, including an exposure ledger of what is still open.
Shared view
Real Word documents, tracked changes included
Generate produces the actual .docx the industry expects, tracked changes and all. Not a PDF export that breaks the workflow.
Word-native
A precedent library that remembers
Past performance solutions become searchable: find the project where a similar atrium was solved, and what the authority accepted.
Searchable
A peek inside

The actual Stage-1 prototype, loaded with the fabricated “Foundry Lane” project. Nothing here is client data.
How it runs in production
Australian inference
Claude on Amazon Bedrock with Australian inference profiles. Session audio and registers are processed onshore.
Your tools stay
Word, SharePoint and the existing PM stack remain the system of record. The AI slots in; nothing gets ripped out.
Engineers sign off
The production boundary was agreed up front: the AI assembles and suggests, humans approve. Always.
What this page doesn’t claim
No invented ROI percentages, no anonymous five-star quotes. This is a Stage-1 working prototype built in days on synthetic data, shown to the client as the first step of a fixed-price scope. When a build like this goes to production, the results section gets real numbers, and not before.
Related services
Drowning in documents your team assembles by hand?
Tell us where the hours go. If it looks like a fit, the next thing you see is a working prototype on your workflow, not a proposal.
