AI use case · Public sector
Rule-based design checking, with every finding linked to the rule and the drawing.
For authorities and review teams that check submitted designs against written regulations by hand. Codified rules check each submission, every finding points to its rule and its place on the drawing, and a reviewer decides.
What is automated design compliance checking?
Automated design compliance checking tests a submitted design against regulations written as explicit rules, such as minimum dimensions, required elements or distances between elements. Each finding cites the rule and the location on the drawing. The software approves nothing: a qualified reviewer checks every finding and makes the decision on the submission.
Where this stands: We built a rule-based design-checking demonstrator for a UAE public-sector authority. It is a demonstration, not a live system, and we don't sell a packaged product for this sector; each project is scoped and built around your rules.
The problem
Every submission is checked against the regulations by hand.
Review teams at UAE authorities receive design submissions as drawings and supporting documents, then check them against written regulations. The reviewer reads the drawing, finds the relevant clause and measures or counts by hand.
Many checks are objective: a dimension meets a minimum or it does not, and a required element is present or missing. They still take reviewer time on every submission, and two reviewers can record the same issue in different words.
Applicants then receive comments that are hard to trace back to a clause or a place on the drawing. Each resubmission repeats the cycle, and the reviewer checks everything again.
Measured by hand
Objective checks, such as dimensions and required elements, are measured by hand on each drawing.
Inconsistent comments
Different reviewers record the same issue differently, so applicants receive mixed messages.
Weak traceability
Comments do not always cite the clause or the location, which makes resubmissions harder to check.
How it works
One submission, from upload to a reviewer's decision.
An illustrative workflow based on the approach of our demonstrator. The rules, file formats and review steps are agreed for each project.
Codify the rules
Your specialists and our engineers turn selected clauses into explicit, testable rules. Each rule keeps its clause reference and a named owner.
Submit
The applicant or reviewer uploads the drawings in the agreed format, with the supporting documents.
Read the design
The system reads the elements, dimensions and labels the rules need. Anything it cannot read is listed, not assumed.
Run the checks
Each rule runs against the design. Every finding records the rule, the clause, the measured value and the location on the drawing.
Reviewer decision
A reviewer opens each finding beside the drawing, confirms or dismisses it and adds the checks that need professional judgement.
Report
The confirmed findings become a report for the applicant, each one traceable to its rule and its drawing.
Who does what
What the AI does · What a person checks.
What the AI does
Runs every codified rule against each submission
Links each finding to its rule, its clause and its place on the drawing
Lists the elements and files it could not read
Runs the same checks again on a resubmission
Can draft finding text for the reviewer to edit, as advice only
What a person checks
Every finding, before it reaches the applicant
The checks that need professional judgement, which stay outside the rules
Each new or changed rule, against its clause
The final decision on the submission
First pilot
A bounded first pilot on a small set of rules.
We test on your own past submissions before anyone talks about results. Scope, timeline and price are confirmed in a written proposal.
A small rule set
Choose a limited group of objective, frequently checked clauses, not the whole regulation.
One submission type
Start with one type of design and the file format most applicants already use.
Past submissions
Previous submissions with known review outcomes, shared safely after the first call, become the test set.
Agreed acceptance
Agree what counts as a correct finding, a missed issue and a false alarm, then compare the results with your reviewers' own comments.
Data, hosting and Arabic
Where the data goes, and how Arabic is handled.
Hosting
The application and stored data can be hosted on Azure UAE North, on private servers managed by Vloud, or in your own cloud account. The AI model is used through a provider agreed per project, and its processing region may be outside the UAE.
Access
Applicants, reviewers and rule owners see different things. Only authorised rule owners can change a rule, and every change is versioned.
Arabic
Findings, reports and the review screens can be in Arabic and English. Arabic labels on drawings are tested as their own sample.
Audit trail
Each check records the rule version, the input files and the reviewer's decision, so a result can be explained later. Retention follows your records policy.
Honest limits
Where it does not work yet.
- Only clauses that can be written as clear, measurable rules can be automated. Judgement stays with reviewers.
- Drawing quality matters. Scans and inconsistently prepared drawings may not be readable, and those submissions go to manual review.
- Our demonstrator is a demonstration, not a live system. We claim no results from live use, and we do not quote accuracy or time savings before testing on your submissions.
- Rules need owners. When a regulation changes, the affected rules must be updated and tested again before they are used.
- We don't sell a packaged product for this sector. Each project is built around your rules and your review process.
What drives cost and timeline. The number of rules and how hard they are to codify, drawing formats and quality, and the number of submission types. Integration with your case system, languages, audit requirements and hosting also count. We don't publish a price list; scope, timeline and price come in a written proposal after the discovery call.
Evidence
What we can show you.
Every example is labelled: delivered work, our own product, a demonstrator, or an illustration with synthetic data.
- Demonstrator
A rule-based design-checking demonstrator
A rule-based design-checking demonstrator built for a UAE public-sector authority. It is a demonstration, not a live system. Findings come from codified rules, and AI plays an advisory role only.
- Illustrative
How to test AI document extraction
The held-out sample method we use to evaluate any pilot before results are discussed. It is an article, not a client project.
Before you start
Questions buyers ask.
Is this AI, or rule-based?
The compliance findings come from codified rules, so each one can be explained by its rule and clause. AI can help read documents or draft text for the reviewer, but it does not decide whether a design complies.
Does the software approve designs?
No. It produces findings. A qualified reviewer checks each one and makes the decision on the submission.
Which regulations can it check?
Any clause that can be written as a clear, measurable rule, once your specialists confirm how it should be read. Clauses that need judgement stay with reviewers.
What happens when a regulation changes?
The rule owners update the affected rules, test them again on past submissions and publish a new version. Every result shows which version of each rule was applied.
Is your demonstrator a live system?
No. It is a rule-based design-checking demonstrator built for a UAE public-sector authority. It is a demonstration, not a live system, and we don't sell a packaged product for this sector.
Which drawing formats work?
That is confirmed in the pilot. Structured digital drawings are the most reliable input, while scans and inconsistently prepared files often need manual review.
Who owns the code?
You own the source code and IP of the custom work we build for you. Our own products stay ours, and any licence terms are agreed separately.
Related
Related AI use cases.
Arabic correspondence classification and routing
Classify incoming Arabic and English letters and emails by topic and urgency, suggest the owning department and keep a registry officer on every decision.Submittal and RFI first-pass review
First-pass checks on submittals and RFIs against the specification and drawings, with each point linked to its source and the consultant deciding.Government and public sector
How we build traceable workflows for UAE public-service teams.Technology advisory
Feasibility and scoping advice before you commit to a build.AI and intelligent automation
How we scope, build and review AI-assisted workflows.
Next step
Test it on your own documents.
Tell us what the work looks like today. We reply with the questions we need answered before a pilot can be scoped.
