From Hours to Minutes: How We Automated Risk Assessment for a Singapore Engineering Firm
A Singapore engineering firm was spending hours on every WSH Risk Assessment. We built an AI system that generates compliant, print-ready documents in minutes — delivered via WhatsApp.
Every engineering firm in Singapore knows the drill.
Before any job starts, you need a Risk Assessment. The WSH guidelines are clear, the format is prescribed, and the cost of getting it wrong is measured in audit findings and stop-work orders.
The problem was never the rules. It was the time it took to follow them.
The Manual Reality
A single job — say, switchboard maintenance at an industrial plant — took a safety coordinator two to four hours.
Pull the template. List every work activity. Identify hazards for each one. Score severity and likelihood against the 5×5 matrix. Check that high-risk items have additional controls. Format the table. Get it reviewed. Sign it.
Then do it again for the next job. And the one after that.
The knowledge from each completed RA sat in a folder somewhere. Rarely opened. Never reused. Every new job started from a blank page.
Why This Is Hard to Automate
Risk assessments are not ordinary documents. A template and a text generator will not cut it.
WSH compliance has hard rules. The matrix must be 5×5. RPN equals Severity multiplied by Likelihood — never summed, never approximated. Controls must follow the Hierarchy of Control, elimination first, PPE last. Anything scoring 15 or above needs additional controls before approval.
Miss one of those and the document is non-compliant. In an audit, that is a serious finding.
So the brief was not only speed. The system had to be faster and more rigorous than a careful human.
What We Built
A conversational AI that takes a job description and returns a compliant, print-ready PDF. Delivered through the messaging apps the team already uses.
The flow is simple:
- The coordinator describes the upcoming job in plain language
- The AI gathers the header details — location, team, dates
- It proposes work activities and hazards from a validated library built out of past RAs
- The coordinator reviews and edits each row
- Automated compliance checks reject anything that breaks a WSH rule
- A formatted PDF comes back, ready to print
Start to finish, a few minutes.
The AI Is Constrained — Deliberately
The AI does not invent safety rules. The WSH RMCP is the source of truth, not the model’s imagination.
Every hazard and control in the library was validated by a human before it was allowed in. The compliance logic is hardcoded. A coordinator still signs off on every row.
The AI’s job is speed and consistency. It surfaces the right hazards faster, catches the things a tired engineer misses at 5pm on a Friday, and enforces the rules without exception.
That boundary is the whole point.
The Result
Two to four hours became roughly five minutes. Scoring errors — routine when RPN was calculated by hand — went to zero.
Every new RA draws on the full history of previous ones. Quality compounds instead of resetting.
The bigger shift was behavioural. When the document is cheap to produce, nobody cuts corners. The RA stopped feeling like paperwork and started doing its actual job: keeping people safe on site.
What This Means for Your Business
If your team produces safety documents, compliance reports, or any structured output that follows a defined set of rules, there is a version of this that fits your operation.
The pattern holds. Take the knowledge sitting in people’s heads. Encode it into a validated library. Enforce the rules in code. Give your team a faster way to apply what they already know.
The technology is the easy part. The hard part is deciding it is time to stop doing it by hand.
Work with us if that conversation is overdue.
Working on something like this?
Work with us →