Check locally
A document is selected and the prototype checks the first ten pages for a limited set of patterns.
AI ADOPTION · DATA PROTECTION · HUMAN-IN-THE-LOOP WORKFLOW
ClearGate
A B2B SaaS prototype for deciding when an AI document request can continue, when it needs review and what data is actually necessary.
Independent fictional portfolio prototype. It performs a limited browser-based pattern check; it is not a production security scanner, legal control or deployed customer system.
THE OPERATING PROBLEM
AI adoption fails when teams cannot tell the difference between a normal request and a high-risk exception.
Employees need AI to stay useful for ordinary work. Customer administrators and privacy teams need clear controls when documents may contain identifiers, contact details or other sensitive information.
Product principle Let ordinary work move; slow down only when the request gives a concrete reason to do so.
PROTOTYPE SCOPE
A visible decision path instead of an unexplained block.
INPUTEmployee chooses a document and states the task outcome.CHECKLocal prototype check looks for PAN-style IDs, Aadhaar-style numbers, emails and phone numbers.ROUTENo selected pattern found: proceed. Pattern found: request review and propose minimisation.BOUNDARYOnly a limited demonstration. It cannot replace policy, compliance review or production detection tooling.
WORKFLOW DESIGN
Useful AI begins with a decision people can understand.
Explain the route
The employee sees either a standard path or a clear reason why review is required.
Minimise and review
A sensitive request asks for task context and proposes the smallest approved data set.
Keep a human option
A reviewer or specialised data-processing team can complete an exception safely.
DATA MINIMISATION
Approval should be specific to the task, not a blanket permission for a document.
Original document
- Protected by default
- Not sent to AI before a decision
- Available to the human review path when needed
Task context
- Employee explains the intended outcome
- Customer policy informs the route
- Review is limited to genuine exceptions
Approved AI input
- Only data needed for the task
- Example: verification status and date
- Exclude ID number, image and address unless truly required
WHAT A FUTURE PILOT WOULD TEST
Measure safety and usability before claiming adoption impact.
SafetyShare of restricted requests caught and correctly routedSpeedTime to review or human handoff for exceptionsTrustWhether employees and customer admins understand the decision
No pilot has been run. These are proposed validation measures, not measured results.
INTERACTIVE PROTOTYPE
Try the decision path yourself.
Choose a non-sensitive sample PDF, run the local pattern check, then compare the standard path with the review-and-minimisation flow.